Merge origin/main into garrytan/slim-gstack-skills

VERSION → 1.15.0.0 (MINOR bump on top of main's v1.14.0.0). Branch's
v1.13.1.0 work (preamble compression + real-PTY harness + 5 plan-mode
tests passing) consolidated with v1.15.0.0 work (6 new E2E tests on the
harness + parseNumberedOptions + budget regression utils) into a single
release entry — v1.13.1.0 never landed on main, so its content rolls
into the final shippable version per the never-orphan rule in
CLAUDE.md.

Conflicts resolved:
- VERSION: 1.13.1.0 (HEAD) + 1.14.0.0 (main) → 1.15.0.0
- package.json: matching 1.15.0.0
- CHANGELOG.md: replaced HEAD's 1.13.1.0 entry with a consolidated
  1.15.0.0 entry above main's untouched 1.14.0.0 entry. Itemized
  changes split per-version (no shared header).

CLAUDE.md adds "Scale-aware bumps — use common sense" guidance under
CHANGELOG + VERSION style. Big diffs (>2K LOC, new capability) bump
MINOR; PATCH is for fixes/small adds; MAJOR for breaking changes.
Codified after a v1.14.1.0 PATCH attempt got correctly pushed back on
for a ~10K-line additions / -24K-line removals release.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Garry Tan
2026-04-26 04:53:23 -07:00
35 changed files with 3049 additions and 5120 deletions
+4
View File
@@ -20,6 +20,10 @@ bin/gstack-global-discover
.gbrain/
.context/
extension/.auth.json
# xterm assets are vendored from npm at build time; not source-of-truth.
extension/lib/xterm.js
extension/lib/xterm.css
extension/lib/xterm-addon-fit.js
.gstack-worktrees/
/tmp/
*.log
+78 -7
View File
@@ -1,14 +1,14 @@
# Changelog
## [1.13.1.0] - 2026-04-25
## [1.15.0.0] - 2026-04-26
## **Skill prompts get a 25% haircut. Plan-mode E2E tests work for the first time ever.**
## **Skill prompts get a 25% haircut. Plan-mode E2E coverage doubles, and AUQ rendering is now testable.**
Two pieces of work in one release. First, every preamble resolver got compressed: 18 resolvers (Voice, Writing Style, AskUserQuestion Format, Completeness Principle, Plan Mode Info, Brain Sync, Routing Injection, and 11 more) lost a third of their prose without losing a single semantic rule. The full corpus of generated `SKILL.md` files dropped from 3.08 MB to 2.30 MB across 47 outputs. Second, the 5 plan-mode E2E tests added in v1.11.1.0 and rewritten in v1.12.1.0 turned out to have never actually passed. The SDK harness they used couldn't observe Claude's plan-mode confirmation UI, so `result.askUserQuestions.length` was always 0. They fail on `origin/main`. They fail on v1.0.0.0. This release ships a real-PTY harness that drives the actual `claude` binary, watches the rendered terminal, and gets all 5 to green.
Three pieces of work in one release. First, every preamble resolver got compressed: 18 resolvers (Voice, Writing Style, AskUserQuestion Format, Completeness Principle, Plan Mode Info, Brain Sync, Routing Injection, and 11 more) lost a third of their prose without losing a single semantic rule. The full corpus of generated `SKILL.md` files dropped from 3.08 MB to 2.30 MB across 47 outputs. Second, the 5 plan-mode E2E tests added in v1.11.1.0 and rewritten in v1.12.1.0 turned out to have never actually passed — the SDK harness they used couldn't observe Claude's plan-mode confirmation UI. This release ships a real-PTY harness that drives the actual `claude` binary, watches the rendered terminal, and gets all 5 to green. Third, on top of that harness, 6 new E2E tests cover behaviors no test could reach before: AUQ format compliance, plan-design UI-scope detection (positive path), tool-budget regression, /ship idempotency end-to-end, /plan-ceo answer-routing, and /autoplan phase ordering.
### The numbers that matter
Token-level reduction comes from regenerating every `SKILL.md` against the slim resolvers (`bun run gen:skill-docs --host all`). Plan-mode E2E numbers come from `EVALS=1 EVALS_TIER=gate bun test test/skill-e2e-plan-*-plan-mode.test.ts` on a clean working tree.
Token-level reduction comes from regenerating every `SKILL.md` against the slim resolvers (`bun run gen:skill-docs --host all`). Plan-mode E2E numbers come from `EVALS=1 EVALS_TIER=gate bun test test/skill-e2e-plan-*-plan-mode.test.ts` on a clean working tree. New E2E test verification uses the same gate flag against the new test files.
| Metric | Before | After | Δ |
|---|---|---|---|
@@ -17,23 +17,38 @@ Token-level reduction comes from regenerating every `SKILL.md` against the slim
| `plan-ceo-review` preamble | 54 KB | 31 KB | 43% |
| Plan-mode E2E tests passing | 0/5 | 5/5 | +5 |
| Plan-mode E2E wall time | ∞ (never green) | 790 s (sequential) | proven |
| Real-PTY E2E test count | 5 | 11 | +6 |
| Gate-tier paid E2E added | 0 | 3 | auq-format, design-with-ui, budget-regression |
| Periodic-tier paid E2E added | 0 | 3 | mode-routing, ship-idempotency, autoplan-chain |
| New helper unit tests | 0 | 23 | parser + budget regression coverage |
| Skill class | Old preamble | New preamble | Δ |
|---|---|---|---|
| Tier-2+ review skills | ~50 KB | ~30 KB | 40% |
| Tier-1 quick skills | ~12 KB | ~9 KB | 25% |
The biggest wins are the tier-≥3 plan reviews that load full preamble surface (Brain Sync, Context Recovery, Routing Injection): they keep all the load-bearing functionality and lose almost half the bytes. Every gstack invocation is now ~50K tokens lighter.
The biggest wins are the tier-≥3 plan reviews that load full preamble surface (Brain Sync, Context Recovery, Routing Injection): they keep all the load-bearing functionality and lose almost half the bytes. Every gstack invocation is now ~50K tokens lighter, and the test harness can finally observe what users actually see in the terminal.
### What this means for builders
Faster every-skill startup, cheaper prompt-cache pricing on cold reads, more headroom inside the 200K context window for actual work. And for anyone who tried `/plan-ceo-review` in plan mode and watched it silently write a plan file: those tests now actually verify that doesn't happen. Run `bun run gen:skill-docs --host all` after pulling. The 5 plan-mode tests will run in CI on the next gate-tier eval pass.
Faster every-skill startup, cheaper prompt-cache pricing on cold reads, more headroom inside the 200K context window for actual work. The plan-mode E2E tests now actually verify the skill doesn't silently write a plan file when `/plan-ceo-review` runs in plan mode. And the 3 new gate-tier tests catch a class of regression that was previously invisible: AUQ format drift (`Recommendation:` line missing), UI-scope misdetection (positive path), and tool-call budget bloat (a skill burning 3× the tools it used to). Run `bun run gen:skill-docs --host all` after pulling. The 11 plan-mode tests will run in CI on the next gate-tier eval pass.
### Itemized changes
#### Added
- `test/helpers/claude-pty-runner.ts`: real-PTY test harness using `Bun.spawn({terminal:})` (Bun 1.3.10+ has built-in PTY — no `node-pty`, no native modules). Exposes `launchClaudePty()` for raw session control and `runPlanSkillObservation()` as the high-level contract for plan-mode skill tests.
- `parseNumberedOptions(visible)` and `isPermissionDialogVisible(visible)` helpers in `claude-pty-runner.ts`. Tests can now look up an option index by its label without hard-coding positions, and auto-grant Claude Code's file-edit / workspace-trust / bash-permission dialogs that fire during preamble side-effects.
- `findBudgetRegressions()` and `assertNoBudgetRegression()` in `test/helpers/eval-store.ts`. Pure functions returning tests that grew >2× in tools or turns vs the prior eval run, with floors at 5 prior tools / 3 prior turns to avoid noise. Env override `GSTACK_BUDGET_RATIO`.
- 6 new real-PTY E2E tests on the harness:
- `skill-e2e-auq-format-compliance.test.ts` (gate, ~$0.50/run): asserts every gstack `AskUserQuestion` rendering contains the 7 mandated format elements (ELI10, Recommendation, Pros/Cons with ✅/❌, Net, `(recommended)` label).
- `skill-e2e-plan-design-with-ui.test.ts` (gate, ~$0.80/run): positive coverage for `/plan-design-review` UI-scope detection. Counterpart to the existing no-UI early-exit test — without it, a regression that flips the detector to "early-exit always" would ship undetected.
- `skill-budget-regression.test.ts` (gate, free): branch-scoped library-only assertion that no skill burns >2× tools or turns vs its prior recorded run.
- `skill-e2e-plan-ceo-mode-routing.test.ts` (periodic, ~$3/run): verifies AUQ answer routing — HOLD SCOPE picks routes to rigor language, SCOPE EXPANSION picks route to expansion language.
- `skill-e2e-ship-idempotency.test.ts` (periodic, ~$3/run): runs `/ship` end-to-end against a real git fixture with `STATE: ALREADY_BUMPED` baked in; asserts no double-bump, no double-commit, no fixture mutation.
- `skill-e2e-autoplan-chain.test.ts` (periodic, ~$8/run): asserts `/autoplan` phase ordering by tee'ing timestamps as each `**Phase N complete.**` marker appears.
- `test/helpers-unit.test.ts`: 23 unit tests covering `parseNumberedOptions` edge cases (empty, partial paint, >9 options, stale-vs-fresh anchoring) and `findBudgetRegressions` (noise floor, env override, missing tool data).
- `test/fixtures/plans/ui-heavy-feature.md`: planted plan with explicit UI scope keywords for the new design-with-UI test.
- Auto-handling of the workspace-trust dialog so tests run in temp directories without manual intervention.
- Outcome contract: `asked` | `plan_ready` | `silent_write` | `exited` | `timeout`. Tests pass on `asked` or `plan_ready`, fail on the rest.
@@ -43,6 +58,7 @@ Faster every-skill startup, cheaper prompt-cache pricing on cold reads, more hea
- All 47 generated `SKILL.md` files regenerated; 3 ship golden fixtures regenerated.
- Plan-* skills retain full preamble surface (Brain Sync, Context Recovery, Routing Injection) — the early slim attempt that cut these was reverted after diagnosing them as load-bearing.
- 5 plan-mode E2E tests rewritten on the new harness with a 300s observation budget.
- `isNumberedOptionListVisible` regex tolerates whitespace collapse from TTY cursor-positioning escapes (`\x1b[40C`) which `stripAnsi` removes — `\b2\.` was failing on word-to-word transitions where stripped output read `text2.`.
#### Fixed
@@ -56,9 +72,64 @@ Faster every-skill startup, cheaper prompt-cache pricing on cold reads, more hea
#### For contributors
- `test/helpers/touchfiles.ts`: 5 plan-mode test selections + e2e-harness-audit selection now point at `claude-pty-runner.ts` instead of the deleted helper.
- `test/helpers/touchfiles.ts`: 5 plan-mode test selections + e2e-harness-audit selection now point at `claude-pty-runner.ts` instead of the deleted helper. 6 new entries (`auq-format-pty`, `plan-ceo-mode-routing`, `plan-design-with-ui-scope`, `budget-regression-pty`, `ship-idempotency-pty`, `autoplan-chain-pty`) with tier classifications: 3 gate, 3 periodic.
- `test/e2e-harness-audit.test.ts`: recognizes `runPlanSkillObservation` as a valid coverage path alongside the legacy `canUseTool` / `runPlanModeSkillTest` patterns.
- New unit test: `test/gen-skill-docs.test.ts` asserts plan-review preambles stay under 33 KB and the slim Voice section preserves its load-bearing semantic contract (lead-with-the-point, name-the-file, user-outcome framing, no-corporate, no-AI-vocab, user-sovereignty).
- `test/touchfiles.test.ts`: skill-specific change selection count updated 15 → 18 to match the 6 new touchfile entries that depend on `plan-ceo-review/**`.
## [1.14.0.0] - 2026-04-25
## **The gstack browser sidebar is now an interactive Claude Code REPL with live tab awareness.**
Open the side panel and Claude Code is right there in a real terminal. Type, watch the agent work, switch browser tabs and Claude sees the change. The old one-shot chat queue is gone. Two-way conversation, slash commands, `/resume`, ANSI colors, all of it. Plus a `$B tab-each` command that fans out a single browse command across every open tab and returns per-tab JSON results.
### The numbers that matter
| Metric | Before | After | Δ |
|---|---|---|---|
| Sidebar surfaces | Chat (one-shot `claude -p`) + 3 debug | Terminal (live PTY) + 3 debug | -1 surface, +interactive |
| Subprocesses spawned per session | Many (one per chat message) | One (PTY claude, lazy-spawned) | -N |
| Lines in `extension/sidepanel.js` | 1969 | 1042 | -47% |
| Total diff | — | 27 files, +2875 / -3885 | -1010 net |
| New unit + integration + regression tests | 0 | 56+ | +56 |
| Live `tabs.json` push latency | n/a (no live state) | <50ms after `chrome.tabs` event | new capability |
### What this means for builders
Open the sidebar, type. Real PTY means slash commands, `/resume`, real ANSI rendering, real claude process lifecycle. Switch browser tabs while Claude is running and `<stateDir>/tabs.json` + `active-tab.json` update in place — Claude reads them, no need to ask `$B tabs`. Need to do the same thing on every tab? `$B tab-each <command>` returns a JSON array, original active tab restored when done, no OS focus stealing.
The old chat queue is gone. `sidebar-agent.ts`, `/sidebar-command`, `/sidebar-chat`, `/sidebar-agent/event` all deleted. The Cleanup / Screenshot / Cookies toolbar buttons survive in the Terminal pane — Cleanup pipes its prompt straight into the live PTY via `window.gstackInjectToTerminal()` instead of spawning yet another `claude -p`.
### Itemized changes
#### Added
- **Interactive Terminal sidebar tab.** xterm.js + a non-compiled `terminal-agent.ts` Bun process that spawns claude with `Bun.spawn({terminal: {rows, cols, data}})`. Auto-connects when the side panel opens, no keypress needed.
- **`$B tab-each <command>`** — fan-out helper for multi-tab work. Returns `{command, args, total, results: [{tabId, url, title, status, output}]}`. Skips chrome:// pages, scope-checks the inner command before iterating, restores the original active tab in a `finally` block, never pulls focus away from the user's foreground app.
- **Live tab state files.** `<stateDir>/tabs.json` (full list with id, url, title, active, pinned, audible, windowId) and `<stateDir>/active-tab.json` (current active). Updated atomically on every `chrome.tabs` event (activated, created, removed, URL/title change). Claude reads on demand instead of running `$B tabs`.
- **Tab-awareness system prompt** injected via `claude --append-system-prompt` at spawn so the model knows about the state files and the `$B tab-each` command without being told.
- **Always-visible Restart button** in the Terminal toolbar. Force-restart claude any time, not just from the "session ended" state.
#### Changed
- **Sidebar is Terminal-only.** No more `Terminal | Chat` primary tab nav. Activity / Refs / Inspector still live behind the `debug` toggle in the footer. Quick-actions (🧹 Cleanup / 📸 Screenshot / 🍪 Cookies) moved into the Terminal toolbar.
- **WebSocket auth uses `Sec-WebSocket-Protocol`** instead of cookies. Browsers can't set `Authorization` on WS upgrades, and `SameSite=Strict` cookies don't survive the cross-port jump from server.ts:34567 to the agent's random port from a chrome-extension origin. The token rides on `new WebSocket(url, [`gstack-pty.<token>`])` and the agent echoes the protocol back (Chromium closes connections that don't pick a protocol).
- **Cleanup button now drives the live PTY.** Clicking "🧹 Cleanup" injects the cleanup prompt straight into claude via `window.gstackInjectToTerminal()`. The Inspector "Send to Code" action uses the same path. No more `/sidebar-command` POSTs.
- **Repaint after debug-tab close.** xterm.js doesn't auto-redraw when its container flips from `display: none` back to `display: flex`. A MutationObserver on `#tab-terminal`'s class attribute now forces a `fitAddon.fit() + term.refresh() + resize` push when the pane becomes visible.
#### Removed
- **`browse/src/sidebar-agent.ts`** — the one-shot `claude -p` queue worker. ~900 lines.
- **Server endpoints**: `/sidebar-command`, `/sidebar-chat[/clear]`, `/sidebar-agent/{event,kill,stop}`, `/sidebar-tabs[/switch]`, `/sidebar-session{,/new,/list}`, `/sidebar-queue/dismiss`. ~600 lines.
- **Chat-related state** in server.ts: `ChatEntry`, `SidebarSession`, `TabAgentState`, `pickSidebarModel`, `addChatEntry`, `processAgentEvent`, `killAgent`, the agent-health watchdog, `chatBuffer`, the per-tab agent map.
- **Chat UI in sidepanel.html**: primary-tab nav, `<main id="tab-chat">`, the chat input bar, the experimental "Browser co-pilot" banner, the security event banner, the `clear-chat` footer button.
- **Five obsolete test files**: `sidebar-agent.test.ts`, `sidebar-agent-roundtrip.test.ts`, `security-e2e-fullstack.test.ts`, `security-review-fullstack.test.ts`, `security-review-sidepanel-e2e.test.ts`. Plus 5 chat-only describe blocks inside surviving security tests (loadSession session-ID validation, switchChatTab DocumentFragment, pollChat reentrancy, sidebar-tabs URL sanitization, agent queue security).
#### For contributors
- **`browse/src/pty-session-cookie.ts`** mirrors `sse-session-cookie.ts`. Same TTL, same opportunistic pruning, separate registry (PTY tokens must never be valid as SSE tokens or vice versa).
- **`docs/designs/SIDEBAR_MESSAGE_FLOW.md`** rewritten around the Terminal flow: WebSocket upgrade, dual-token model (`AUTH_TOKEN` for `/pty-session`, `gstack-pty.<token>` for `/ws`, `INTERNAL_TOKEN` for server↔agent loopback), threat-model boundary (Terminal tab bypasses the prompt-injection stack on purpose; user keystrokes are the trust source).
- **`browse/test/terminal-agent.test.ts`** (16 tests) + `terminal-agent-integration.test.ts` (real `/bin/bash` PTY round-trip, raw `Sec-WebSocket-Protocol` upgrade verification) + `tab-each.test.ts` (10 tests with mock `BrowserManager`) + `sidebar-tabs.test.ts` (27 structural assertions locking the chat-rip invariants).
- **CLAUDE.md** updated with the dual-token model, the cookie-vs-protocol rationale, and the cross-pane injection pattern.
- **`vendor:xterm`** build step copies `xterm@5.x` and `xterm-addon-fit` from `node_modules/` into `extension/lib/` at build time. xterm files are gitignored.
- **TODOS.md** carries three v1.1+ follow-ups: PTY session survival across sidebar reload (Issue 1C deferred), `/health` `AUTH_TOKEN` distribution audit (codex finding, pre-existing soft leak), and dropping the now-dead `security-classifier.ts` ML pipeline.
## [1.13.0.0] - 2026-04-25
+54 -6
View File
@@ -225,12 +225,35 @@ When you need to interact with a browser (QA, dogfooding, cookie setup), use the
project uses.
**Sidebar architecture:** Before modifying `sidepanel.js`, `background.js`,
`content.js`, `sidebar-agent.ts`, or sidebar-related server endpoints, read
`docs/designs/SIDEBAR_MESSAGE_FLOW.md`. It documents the full initialization
timeline, message flow, auth token chain, tab concurrency model, and known
failure modes. The sidebar spans 5 files across 2 codebases (extension + server)
with non-obvious ordering dependencies. The doc exists to prevent the kind of
silent failures that come from not understanding the cross-component flow.
`content.js`, `terminal-agent.ts`, or sidebar-related server endpoints,
read `docs/designs/SIDEBAR_MESSAGE_FLOW.md`. The sidebar has one primary
surface — the **Terminal** pane (interactive `claude` PTY) — with
Activity / Refs / Inspector as debug overlays behind the footer's
`debug` toggle. The chat queue path was ripped once the PTY proved out;
`sidebar-agent.ts` and the `/sidebar-command` / `/sidebar-chat` /
`/sidebar-agent/event` endpoints are gone. The doc covers the WS auth
flow, dual-token model, and threat-model boundary — silent failures
here usually trace to not understanding the cross-component flow.
**WebSocket auth uses Sec-WebSocket-Protocol, not cookies.** Browsers
can't set `Authorization` on a WebSocket upgrade, but they CAN set
`Sec-WebSocket-Protocol` via `new WebSocket(url, [token])`. The agent
reads it, validates against `validTokens`, and MUST echo the protocol
back in the upgrade response — without the echo, Chromium closes the
connection immediately. `Set-Cookie: gstack_pty=...` is kept as a
fallback for non-browser callers (the cross-port `SameSite=Strict`
cookie path doesn't survive from a chrome-extension origin).
**Cross-pane PTY injection.** The toolbar's Cleanup button and the
Inspector's "Send to Code" action both pipe text into the live claude
PTY via `window.gstackInjectToTerminal(text)`, exposed by
`sidepanel-terminal.js`. No `/sidebar-command` POST — the live REPL is
the only execution surface in the sidebar now.
**`/health` MUST NOT surface any shell-grant token.** It already leaks
`AUTH_TOKEN` to localhost callers in headed mode (a v1.1+ TODO). Don't
make that worse by adding the PTY session token there. PTY auth flows
through `POST /pty-session` only.
**Transport-layer security** (v1.6.0.0+). When `pair-agent` starts an ngrok tunnel,
the daemon binds two HTTP listeners: a local listener (127.0.0.1, full command
@@ -437,6 +460,31 @@ claims v1.7.0.0 as a MINOR and branch B is also a MINOR, B lands at v1.8.0.0
`bin/gstack-next-version` advances within the chosen bump level rather than
repicking the level when collisions happen.
**Scale-aware bumps — use common sense.** When the diff is big, bump MINOR (or
MAJOR), not PATCH. PATCH is for bug fixes and small additions; MINOR is for
substantial new capability or substantial reduction; MAJOR is for breaking
changes. Rough guideposts (don't treat as rules, treat as smell-checks):
- **PATCH (X.Y.Z+1.0)**: bug fix, doc tweak, small additive change, single
test/file added. Net diff under ~500 lines, no new user-facing capability.
- **MINOR (X.Y+1.0.0)**: new capability shipped (skill, harness, command, big
refactor), substantial code reduction (compression, migration), or coordinated
multi-file change. Net diff over ~2000 lines added/removed, OR a user-visible
feature you'd put in a tweet.
- **MAJOR (X+1.0.0.0)**: breaking change to public surface (CLI flag rename,
skill removed, config format changed), OR a release big enough to be the
headline of a blog post.
If you find yourself debating "is 10K added + 24K removed really a PATCH?" — it
isn't. Bump MINOR. Same for "this adds a whole new test harness with 6 new E2E
tests + helper utilities" — MINOR. The bump level is communication to the user
about what kind of release this is; don't undersell it.
When merging origin/main brings a higher VERSION, re-evaluate the bump level
against the SCALE of your branch's work, not just whether main moved forward.
If main bumped MINOR and your branch is also a substantial change, you bump
MINOR again on top (e.g., main at v1.14.0.0, your branch lands v1.15.0.0).
**VERSION and CHANGELOG are branch-scoped.** Every feature branch that ships gets its
own version bump and CHANGELOG entry. The entry describes what THIS branch adds —
not what was already on main.
+1
View File
@@ -880,6 +880,7 @@ Refs are invalidated on navigation — run `snapshot` again after `goto`.
| `closetab [id]` | Close tab |
| `newtab [url] [--json]` | Open new tab. With --json, returns {"tabId":N,"url":...} for programmatic use (make-pdf). |
| `tab <id>` | Switch to tab |
| `tab-each <command> [args...]` | Run a command on every open tab. Returns JSON with per-tab results. |
| `tabs` | List open tabs |
### Server
+52
View File
@@ -1,5 +1,57 @@
# TODOS
## Sidebar Terminal (cc-pty-import follow-ups)
### v1.1: PTY session survives sidebar reload
**What:** Today the Terminal tab's PTY dies with the WebSocket — sidebar
reload, side-panel close, even a quick navigate-away in another tab close
the session. v1.1 should key the PTY on a tab/session id so a reload
reattaches to the existing claude process and you keep `/resume` history.
**Why:** Mid-task resilience. When you've been pair-programming with claude
for 20 minutes and an accidental Cmd-R blows it away, the cost is real.
**Pros:** Better UX, fewer interrupted sessions. **Cons:** Session-tracking
state, ghost-process risk, lifecycle bugs (when DOES the PTY actually go
away?). v1 chose the simple "PTY dies with WS" model deliberately.
**Context:** /plan-eng-review Issue 1C decision (cc-pty-import branch,
2026-04-25). v1 ships with phoenix's lifecycle. **Depends on:**
cc-pty-import landed.
**Priority:** P2 (nice-to-have).
**Effort:** M. Likely needs a per-tab session map keyed by chrome.tabs.id
plus a TTL so abandoned PTYs eventually exit.
---
### v1.1+: Audit `/health` token distribution
**What:** Codex's outside-voice review on cc-pty-import flagged that
`/health` already surfaces `AUTH_TOKEN` to any localhost caller in headed
mode (`server.ts:1657`). That's a pre-existing soft leak — anything
running on localhost gets the root token by hitting `/health`.
**Why:** cc-pty-import sidesteps it by NOT putting the PTY token there
(uses an HttpOnly cookie path instead). But the underlying leak is still
shippable surface. A second extension or a localhost web app could
currently scrape `AUTH_TOKEN` and hit any browse-server endpoint.
**Pros:** Closes a real privilege-escalation path on multi-extension
machines. **Cons:** Either we tighten the gate (Origin must be OUR
extension id, not just any chrome-extension://) or we move bootstrap
discovery off `/health` entirely. Either has migration cost for tests
and the existing extension.
**Context:** codex finding #2 on cc-pty-import plan-eng review. Not in
scope of that PR; deliberately deferred to keep PTY-import small.
**Priority:** P2.
**Effort:** M.
---
## Testing
## P1: Structural STOP-Ask forcing function across all skills
+1 -1
View File
@@ -1 +1 @@
1.13.1.0
1.15.0.0
+1
View File
@@ -804,6 +804,7 @@ $B prettyscreenshot --cleanup --scroll-to ".pricing" --width 1440 ~/Desktop/hero
| `closetab [id]` | Close tab |
| `newtab [url] [--json]` | Open new tab. With --json, returns {"tabId":N,"url":...} for programmatic use (make-pdf). |
| `tab <id>` | Switch to tab |
| `tab-each <command> [args...]` | Run a command on every open tab. Returns JSON with per-tab results. |
| `tabs` | List open tabs |
### Server
+32 -47
View File
@@ -853,7 +853,7 @@ Refs: After 'snapshot', use @e1, @e2... as selectors:
// Delete stale state file
safeUnlinkQuiet(config.stateFile);
console.log('Launching headed Chromium with extension + sidebar agent...');
console.log('Launching headed Chromium with extension + terminal agent...');
try {
// Start server in headed mode with extension auto-loaded
// Use a well-known port so the Chrome extension auto-connects
@@ -882,56 +882,41 @@ Refs: After 'snapshot', use @e1, @e2... as selectors:
const status = await resp.text();
console.log(`Connected to real Chrome\n${status}`);
// Auto-start sidebar agent
// __dirname is inside $bunfs in compiled binaries — resolve from execPath instead
let agentScript = path.resolve(__dirname, 'sidebar-agent.ts');
if (!fs.existsSync(agentScript)) {
agentScript = path.resolve(path.dirname(process.execPath), '..', 'src', 'sidebar-agent.ts');
// sidebar-agent.ts spawn was here. Ripped alongside the chat queue —
// the Terminal pane runs an interactive PTY now, no more one-shot
// claude -p subprocesses to multiplex.
// Auto-start terminal agent (non-compiled bun process). Owns the PTY
// WebSocket for the sidebar Terminal pane.
let termAgentScript = path.resolve(__dirname, 'terminal-agent.ts');
if (!fs.existsSync(termAgentScript)) {
termAgentScript = path.resolve(path.dirname(process.execPath), '..', 'src', 'terminal-agent.ts');
}
try {
if (!fs.existsSync(agentScript)) {
throw new Error(`sidebar-agent.ts not found at ${agentScript}`);
if (fs.existsSync(termAgentScript)) {
// Kill old terminal-agents so a stale port file can't trick the
// server into routing /pty-session at a dead listener.
try {
const { spawnSync } = require('child_process');
spawnSync('pkill', ['-f', 'terminal-agent\\.ts'], { stdio: 'ignore', timeout: 3000 });
} catch (err: any) {
if (err?.code !== 'ENOENT') throw err;
}
const termProc = Bun.spawn(['bun', 'run', termAgentScript], {
cwd: config.projectDir,
env: {
...process.env,
BROWSE_STATE_FILE: config.stateFile,
BROWSE_SERVER_PORT: String(newState.port),
},
stdio: ['ignore', 'ignore', 'ignore'],
});
termProc.unref();
console.log(`[browse] Terminal agent started (PID: ${termProc.pid})`);
}
// Clear old agent queue
const agentQueue = path.join(process.env.HOME || '/tmp', '.gstack', 'sidebar-agent-queue.jsonl');
try {
fs.mkdirSync(path.dirname(agentQueue), { recursive: true, mode: 0o700 });
fs.writeFileSync(agentQueue, '', { mode: 0o600 });
} catch (err: any) {
if (err?.code !== 'EACCES') throw err;
}
// Resolve browse binary path the same way — execPath-relative
let browseBin = path.resolve(__dirname, '..', 'dist', 'browse');
if (!fs.existsSync(browseBin)) {
browseBin = process.execPath; // the compiled binary itself
}
// Kill any existing sidebar-agent processes before starting a new one.
// Old agents have stale auth tokens and will silently fail to relay events,
// causing the server to mark the agent as "hung".
try {
const { spawnSync } = require('child_process');
spawnSync('pkill', ['-f', 'sidebar-agent\\.ts'], { stdio: 'ignore', timeout: 3000 });
} catch (err: any) {
if (err?.code !== 'ENOENT') throw err;
}
const agentProc = Bun.spawn(['bun', 'run', agentScript], {
cwd: config.projectDir,
env: {
...process.env,
BROWSE_BIN: browseBin,
BROWSE_STATE_FILE: config.stateFile,
BROWSE_SERVER_PORT: String(newState.port),
},
stdio: ['ignore', 'ignore', 'ignore'],
});
agentProc.unref();
console.log(`[browse] Sidebar agent started (PID: ${agentProc.pid})`);
} catch (err: any) {
console.error(`[browse] Sidebar agent failed to start: ${err.message}`);
console.error(`[browse] Run manually: bun run ${agentScript}`);
// Non-fatal: chat still works without the terminal agent.
console.error(`[browse] Terminal agent failed to start: ${err.message}`);
}
} catch (err: any) {
console.error(`[browse] Connect failed: ${err.message}`);
+2 -1
View File
@@ -30,7 +30,7 @@ export const WRITE_COMMANDS = new Set([
]);
export const META_COMMANDS = new Set([
'tabs', 'tab', 'newtab', 'closetab',
'tabs', 'tab', 'tab-each', 'newtab', 'closetab',
'status', 'stop', 'restart',
'screenshot', 'pdf', 'responsive',
'chain', 'diff',
@@ -144,6 +144,7 @@ export const COMMAND_DESCRIPTIONS: Record<string, { category: string; descriptio
'tab': { category: 'Tabs', description: 'Switch to tab', usage: 'tab <id>' },
'newtab': { category: 'Tabs', description: 'Open new tab. With --json, returns {"tabId":N,"url":...} for programmatic use (make-pdf).', usage: 'newtab [url] [--json]' },
'closetab':{ category: 'Tabs', description: 'Close tab', usage: 'closetab [id]' },
'tab-each':{ category: 'Tabs', description: 'Run a command on every open tab. Returns JSON with per-tab results.', usage: 'tab-each <command> [args...]' },
// Server
'status': { category: 'Server', description: 'Health check' },
'stop': { category: 'Server', description: 'Shutdown server' },
+102
View File
@@ -285,6 +285,108 @@ export async function handleMetaCommand(
return `Closed tab${id ? ` ${id}` : ''}`;
}
case 'tab-each': {
// Fan out a single command across every open tab. Returns a JSON
// object: { results: [{tabId, url, title, status, output}], total }.
// Restores the originally active tab when done so the user's view
// doesn't shift under them.
//
// Usage: $B tab-each <command> [args...]
// $B tab-each snapshot -i → snapshot every tab
// $B tab-each text → grab clean text from every tab
// $B tab-each goto https://x.y → load the same URL in every tab
if (args.length === 0) {
throw new Error(
'Usage: browse tab-each <command> [args...]\n' +
'Example: browse tab-each snapshot -i'
);
}
const innerRaw = args[0];
const innerName = canonicalizeCommand(innerRaw);
const innerArgs = args.slice(1);
// Scope check the inner command before fanning out, so a single
// permission failure aborts the whole batch instead of partially
// mutating tabs.
if (tokenInfo && tokenInfo.clientId !== 'root' && !checkScope(tokenInfo, innerName)) {
throw new Error(
`tab-each rejected: subcommand "${innerRaw}" not allowed by your token scope (${tokenInfo.scopes.join(', ')}).`
);
}
const tabs = await bm.getTabListWithTitles();
const originalActive = tabs.find(t => t.active)?.id ?? bm.getActiveTabId();
const executeCmd = opts?.executeCommand;
const results: Array<{
tabId: number;
url: string;
title: string;
status: number;
output: string;
}> = [];
try {
for (const tab of tabs) {
// Skip chrome:// internal pages — they aren't useful targets and
// many commands fail outright on them.
if (tab.url.startsWith('chrome://') || tab.url.startsWith('chrome-extension://')) {
results.push({
tabId: tab.id,
url: tab.url,
title: tab.title || '',
status: 0,
output: 'skipped: internal page',
});
continue;
}
// Switch to the tab. Don't pull focus away — we're a background
// operation; the user shouldn't see the OS window jump.
bm.switchTab(tab.id, { bringToFront: false });
let status = 0;
let output = '';
if (executeCmd) {
const r = await executeCmd(
{ command: innerName, args: innerArgs, tabId: tab.id },
tokenInfo,
);
status = r.status;
output = r.result;
if (status !== 200) {
try { output = JSON.parse(output).error || output; } catch (err: any) { if (!(err instanceof SyntaxError)) throw err; }
}
} else {
// Fallback path (CLI / test harness without a server context).
// We don't recurse through read/write/meta directly here because
// tab-each is only meaningful with the live server; surface a
// clear error.
status = 500;
output = 'tab-each requires the browse server (no executeCommand context)';
}
results.push({
tabId: tab.id,
url: tab.url,
title: tab.title || '',
status,
output,
});
}
} finally {
// Restore the original active tab so the user's view is unchanged.
try { bm.switchTab(originalActive, { bringToFront: false }); } catch {}
}
return JSON.stringify({
command: innerName,
args: innerArgs,
total: results.length,
results,
}, null, 2);
}
// ─── Server Control ────────────────────────────────
case 'status': {
const page = bm.getPage();
+122
View File
@@ -0,0 +1,122 @@
/**
* Session cookie registry for the Terminal sidebar tab's PTY WebSocket.
*
* Why this exists: WebSocket clients in browsers cannot send Authorization
* headers on the upgrade request. The terminal-agent's /ws upgrade therefore
* authenticates via cookie. We never put the PTY token in /health (codex
* outside-voice finding #2: /health already leaks AUTH_TOKEN to any
* localhost caller in headed mode; reusing that path for shell access would
* widen an existing bug). Instead, the extension does an authenticated
* POST /pty-session with the bootstrap AUTH_TOKEN; the server mints a
* short-lived cookie scoped to this terminal session and pushes it to the
* agent via loopback. The browser then carries the cookie automatically on
* the WS upgrade.
*
* Design mirrors `sse-session-cookie.ts` deliberately. Same TTL, same
* scoped-token-must-not-be-valid-as-root invariant, same opportunistic
* pruning. Two registries instead of one because the cookie names are
* different (`gstack_sse` vs `gstack_pty`) and the token spaces must not
* overlap an SSE-read cookie must never grant PTY access, and vice versa.
*/
import * as crypto from 'crypto';
interface Session {
createdAt: number;
expiresAt: number;
}
const TTL_MS = 30 * 60 * 1000; // 30 minutes — matches SSE cookie
const MAX_SESSIONS = 10_000;
const sessions = new Map<string, Session>();
export const PTY_COOKIE_NAME = 'gstack_pty';
/** Mint a fresh PTY session token. */
export function mintPtySessionToken(): { token: string; expiresAt: number } {
const token = crypto.randomBytes(32).toString('base64url');
const now = Date.now();
const expiresAt = now + TTL_MS;
sessions.set(token, { createdAt: now, expiresAt });
pruneExpired(now);
return { token, expiresAt };
}
/**
* Validate a token. Returns true only if the token exists AND is not expired.
* Lazily removes expired entries; opportunistically prunes a few more on
* every call so the registry stays bounded under reconnect pressure.
*/
export function validatePtySessionToken(token: string | null | undefined): boolean {
if (!token) return false;
const s = sessions.get(token);
if (!s) {
pruneExpired(Date.now());
return false;
}
if (Date.now() > s.expiresAt) {
sessions.delete(token);
pruneExpired(Date.now());
return false;
}
return true;
}
/**
* Drop a session token (called on WS close so a leaked cookie can't be
* replayed against a new PTY).
*/
export function revokePtySessionToken(token: string | null | undefined): void {
if (!token) return;
sessions.delete(token);
}
/** Parse the PTY session token from a Cookie header. */
export function extractPtyCookie(req: Request): string | null {
const cookieHeader = req.headers.get('cookie');
if (!cookieHeader) return null;
for (const part of cookieHeader.split(';')) {
const [name, ...valueParts] = part.trim().split('=');
if (name === PTY_COOKIE_NAME) {
return valueParts.join('=') || null;
}
}
return null;
}
/**
* Build the Set-Cookie header value for the PTY session cookie.
* - HttpOnly: not readable from JS (mitigates XSS exfiltration).
* - SameSite=Strict: not sent on cross-site requests (mitigates CSWSH).
* - Path=/: scope to whole origin so /ws and /pty-session both see it.
* - Max-Age matches the TTL.
*
* Secure is intentionally omitted: the daemon binds to 127.0.0.1 over plain
* HTTP; setting Secure would prevent the browser from ever sending it back.
*/
export function buildPtySetCookie(token: string): string {
const maxAge = Math.floor(TTL_MS / 1000);
return `${PTY_COOKIE_NAME}=${token}; HttpOnly; SameSite=Strict; Path=/; Max-Age=${maxAge}`;
}
/** Clear the PTY session cookie. */
export function buildPtyClearCookie(): string {
return `${PTY_COOKIE_NAME}=; HttpOnly; SameSite=Strict; Path=/; Max-Age=0`;
}
function pruneExpired(now: number): void {
let checked = 0;
for (const [token, session] of sessions) {
if (checked++ >= 20) break;
if (session.expiresAt <= now) sessions.delete(token);
}
while (sessions.size > MAX_SESSIONS) {
const first = sessions.keys().next().value;
if (!first) break;
sessions.delete(first);
}
}
// Test-only reset.
export function __resetPtySessions(): void {
sessions.clear();
}
+148 -905
View File
File diff suppressed because it is too large Load Diff
-947
View File
@@ -1,947 +0,0 @@
/**
* Sidebar Agent polls agent-queue from server, spawns claude -p for each
* message, streams live events back to the server via /sidebar-agent/event.
*
* This runs as a NON-COMPILED bun process because compiled bun binaries
* cannot posix_spawn external executables. The server writes to the queue
* file, this process reads it and spawns claude.
*
* Usage: BROWSE_BIN=/path/to/browse bun run browse/src/sidebar-agent.ts
*/
import { spawn } from 'child_process';
import * as fs from 'fs';
import * as path from 'path';
import { safeUnlink } from './error-handling';
import {
checkCanaryInStructure, logAttempt, hashPayload, extractDomain,
combineVerdict, writeSessionState, readSessionState, THRESHOLDS,
readDecision, clearDecision, excerptForReview,
type LayerSignal,
} from './security';
import {
loadTestsavant, scanPageContent, checkTranscript,
shouldRunTranscriptCheck, getClassifierStatus,
loadDeberta, scanPageContentDeberta,
type ToolCallInput,
} from './security-classifier';
const QUEUE = process.env.SIDEBAR_QUEUE_PATH || path.join(process.env.HOME || '/tmp', '.gstack', 'sidebar-agent-queue.jsonl');
const KILL_FILE = path.join(path.dirname(QUEUE), 'sidebar-agent-kill');
const SERVER_PORT = parseInt(process.env.BROWSE_SERVER_PORT || '34567', 10);
const SERVER_URL = `http://127.0.0.1:${SERVER_PORT}`;
const POLL_MS = 200; // 200ms poll — keeps time-to-first-token low
const B = process.env.BROWSE_BIN || path.resolve(__dirname, '../../.claude/skills/gstack/browse/dist/browse');
const CANCEL_DIR = path.join(process.env.HOME || '/tmp', '.gstack');
function cancelFileForTab(tabId: number): string {
return path.join(CANCEL_DIR, `sidebar-agent-cancel-${tabId}`);
}
interface QueueEntry {
prompt: string;
args?: string[];
stateFile?: string;
cwd?: string;
tabId?: number | null;
message?: string | null;
pageUrl?: string | null;
sessionId?: string | null;
ts?: string;
canary?: string; // session-scoped token; leak = prompt injection evidence
}
function isValidQueueEntry(e: unknown): e is QueueEntry {
if (typeof e !== 'object' || e === null) return false;
const obj = e as Record<string, unknown>;
if (typeof obj.prompt !== 'string' || obj.prompt.length === 0) return false;
if (obj.args !== undefined && (!Array.isArray(obj.args) || !obj.args.every(a => typeof a === 'string'))) return false;
if (obj.stateFile !== undefined) {
if (typeof obj.stateFile !== 'string') return false;
if (obj.stateFile.includes('..')) return false;
}
if (obj.cwd !== undefined) {
if (typeof obj.cwd !== 'string') return false;
if (obj.cwd.includes('..')) return false;
}
if (obj.tabId !== undefined && obj.tabId !== null && typeof obj.tabId !== 'number') return false;
if (obj.message !== undefined && obj.message !== null && typeof obj.message !== 'string') return false;
if (obj.pageUrl !== undefined && obj.pageUrl !== null && typeof obj.pageUrl !== 'string') return false;
if (obj.sessionId !== undefined && obj.sessionId !== null && typeof obj.sessionId !== 'string') return false;
if (obj.canary !== undefined && typeof obj.canary !== 'string') return false;
return true;
}
let lastLine = 0;
let authToken: string | null = null;
// Per-tab processing — each tab can run its own agent concurrently
const processingTabs = new Set<number>();
// Active claude subprocesses — keyed by tabId for targeted kill
const activeProcs = new Map<number, ReturnType<typeof spawn>>();
let activeProc: ReturnType<typeof spawn> | null = null;
// Kill-file timestamp last seen — avoids double-kill on same write
let lastKillTs = 0;
// ─── File drop relay ──────────────────────────────────────────
function getGitRoot(): string | null {
try {
const { execSync } = require('child_process');
return execSync('git rev-parse --show-toplevel', { encoding: 'utf-8', stdio: ['pipe', 'pipe', 'pipe'] }).trim();
} catch (err: any) {
console.debug('[sidebar-agent] Not in a git repo:', err.message);
return null;
}
}
function writeToInbox(message: string, pageUrl?: string, sessionId?: string): void {
const gitRoot = getGitRoot();
if (!gitRoot) {
console.error('[sidebar-agent] Cannot write to inbox — not in a git repo');
return;
}
const inboxDir = path.join(gitRoot, '.context', 'sidebar-inbox');
fs.mkdirSync(inboxDir, { recursive: true, mode: 0o700 });
const now = new Date();
const timestamp = now.toISOString().replace(/:/g, '-');
const filename = `${timestamp}-observation.json`;
const tmpFile = path.join(inboxDir, `.${filename}.tmp`);
const finalFile = path.join(inboxDir, filename);
const inboxMessage = {
type: 'observation',
timestamp: now.toISOString(),
page: { url: pageUrl || 'unknown', title: '' },
userMessage: message,
sidebarSessionId: sessionId || 'unknown',
};
fs.writeFileSync(tmpFile, JSON.stringify(inboxMessage, null, 2), { mode: 0o600 });
fs.renameSync(tmpFile, finalFile);
console.log(`[sidebar-agent] Wrote inbox message: ${filename}`);
}
// ─── Auth ────────────────────────────────────────────────────────
async function refreshToken(): Promise<string | null> {
// Read token from state file (same-user, mode 0o600) instead of /health
try {
const stateFile = process.env.BROWSE_STATE_FILE ||
path.join(process.env.HOME || '/tmp', '.gstack', 'browse.json');
const data = JSON.parse(fs.readFileSync(stateFile, 'utf-8'));
authToken = data.token || null;
return authToken;
} catch (err: any) {
console.error('[sidebar-agent] Failed to refresh auth token:', err.message);
return null;
}
}
// ─── Event relay to server ──────────────────────────────────────
async function sendEvent(event: Record<string, any>, tabId?: number): Promise<void> {
if (!authToken) await refreshToken();
if (!authToken) return;
try {
await fetch(`${SERVER_URL}/sidebar-agent/event`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'Authorization': `Bearer ${authToken}`,
},
body: JSON.stringify({ ...event, tabId: tabId ?? null }),
});
} catch (err) {
console.error('[sidebar-agent] Failed to send event:', err);
}
}
// ─── Claude subprocess ──────────────────────────────────────────
function shorten(str: string): string {
return str
.replace(new RegExp(B.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'), 'g'), '$B')
.replace(/\/Users\/[^/]+/g, '~')
.replace(/\/conductor\/workspaces\/[^/]+\/[^/]+/g, '')
.replace(/\.claude\/skills\/gstack\//g, '')
.replace(/browse\/dist\/browse/g, '$B');
}
function describeToolCall(tool: string, input: any): string {
if (!input) return '';
// For Bash commands, generate a plain-English description
if (tool === 'Bash' && input.command) {
const cmd = input.command;
// Browse binary commands — the most common case
const browseMatch = cmd.match(/\$B\s+(\w+)|browse[^\s]*\s+(\w+)/);
if (browseMatch) {
const browseCmd = browseMatch[1] || browseMatch[2];
const args = cmd.split(/\s+/).slice(2).join(' ');
switch (browseCmd) {
case 'goto': return `Opening ${args.replace(/['"]/g, '')}`;
case 'snapshot': return args.includes('-i') ? 'Scanning for interactive elements' : args.includes('-D') ? 'Checking what changed' : 'Taking a snapshot of the page';
case 'screenshot': return `Saving screenshot${args ? ` to ${shorten(args)}` : ''}`;
case 'click': return `Clicking ${args}`;
case 'fill': { const parts = args.split(/\s+/); return `Typing "${parts.slice(1).join(' ')}" into ${parts[0]}`; }
case 'text': return 'Reading page text';
case 'html': return args ? `Reading HTML of ${args}` : 'Reading full page HTML';
case 'links': return 'Finding all links on the page';
case 'forms': return 'Looking for forms';
case 'console': return 'Checking browser console for errors';
case 'network': return 'Checking network requests';
case 'url': return 'Checking current URL';
case 'back': return 'Going back';
case 'forward': return 'Going forward';
case 'reload': return 'Reloading the page';
case 'scroll': return args ? `Scrolling to ${args}` : 'Scrolling down';
case 'wait': return `Waiting for ${args}`;
case 'inspect': return args ? `Inspecting CSS of ${args}` : 'Getting CSS for last picked element';
case 'style': return `Changing CSS: ${args}`;
case 'cleanup': return 'Removing page clutter (ads, popups, banners)';
case 'prettyscreenshot': return 'Taking a clean screenshot';
case 'css': return `Checking CSS property: ${args}`;
case 'is': return `Checking if element is ${args}`;
case 'diff': return `Comparing ${args}`;
case 'responsive': return 'Taking screenshots at mobile, tablet, and desktop sizes';
case 'status': return 'Checking browser status';
case 'tabs': return 'Listing open tabs';
case 'focus': return 'Bringing browser to front';
case 'select': return `Selecting option in ${args}`;
case 'hover': return `Hovering over ${args}`;
case 'viewport': return `Setting viewport to ${args}`;
case 'upload': return `Uploading file to ${args.split(/\s+/)[0]}`;
default: return `Running browse ${browseCmd} ${args}`.trim();
}
}
// Non-browse bash commands
if (cmd.includes('git ')) return `Running: ${shorten(cmd)}`;
let short = shorten(cmd);
return short.length > 100 ? short.slice(0, 100) + '…' : short;
}
if (tool === 'Read' && input.file_path) {
// Skip Claude's internal tool-result file reads — they're plumbing, not user-facing
if (input.file_path.includes('/tool-results/') || input.file_path.includes('/.claude/projects/')) return '';
return `Reading ${shorten(input.file_path)}`;
}
if (tool === 'Edit' && input.file_path) return `Editing ${shorten(input.file_path)}`;
if (tool === 'Write' && input.file_path) return `Writing ${shorten(input.file_path)}`;
if (tool === 'Grep' && input.pattern) return `Searching for "${input.pattern}"`;
if (tool === 'Glob' && input.pattern) return `Finding files matching ${input.pattern}`;
try { return shorten(JSON.stringify(input)).slice(0, 80); } catch { return ''; }
}
// Keep the old name as an alias for backward compat
function summarizeToolInput(tool: string, input: any): string {
return describeToolCall(tool, input);
}
/**
* Scan a Claude stream event for the session canary. Returns the channel where
* it leaked, or null if clean. Covers every outbound channel: text blocks,
* text deltas, tool_use arguments (including nested URL/path/command strings),
* and result payloads.
*/
function detectCanaryLeak(event: any, canary: string, buf?: DeltaBuffer): string | null {
if (!canary) return null;
if (event.type === 'assistant' && event.message?.content) {
for (const block of event.message.content) {
if (block.type === 'text' && typeof block.text === 'string' && block.text.includes(canary)) {
return 'assistant_text';
}
if (block.type === 'tool_use' && checkCanaryInStructure(block.input, canary)) {
return `tool_use:${block.name}`;
}
}
}
if (event.type === 'content_block_start' && event.content_block?.type === 'tool_use') {
if (checkCanaryInStructure(event.content_block.input, canary)) {
return `tool_use:${event.content_block.name}`;
}
}
if (event.type === 'content_block_delta' && event.delta?.type === 'text_delta') {
if (typeof event.delta.text === 'string') {
// Rolling buffer: an attacker can ask Claude to emit the canary split
// across two deltas (e.g., "CANARY-" then "ABCDEF"). A per-delta
// substring check misses this. Concatenate the previous tail with
// this chunk and search, then trim the tail to last canary.length-1
// chars for the next event.
const combined = buf ? buf.text_delta + event.delta.text : event.delta.text;
if (combined.includes(canary)) return 'text_delta';
if (buf) buf.text_delta = combined.slice(-(canary.length - 1));
}
}
if (event.type === 'content_block_delta' && event.delta?.type === 'input_json_delta') {
if (typeof event.delta.partial_json === 'string') {
const combined = buf ? buf.input_json_delta + event.delta.partial_json : event.delta.partial_json;
if (combined.includes(canary)) return 'tool_input_delta';
if (buf) buf.input_json_delta = combined.slice(-(canary.length - 1));
}
}
if (event.type === 'content_block_stop' && buf) {
// Block boundary — reset the rolling buffer so a canary straddling
// two independent tool_use blocks isn't inferred.
buf.text_delta = '';
buf.input_json_delta = '';
}
if (event.type === 'result' && typeof event.result === 'string' && event.result.includes(canary)) {
return 'result';
}
return null;
}
/** Rolling-window tails for delta canary detection. See detectCanaryLeak. */
interface DeltaBuffer {
text_delta: string;
input_json_delta: string;
}
interface CanaryContext {
canary: string;
pageUrl: string;
onLeak: (channel: string) => void;
deltaBuf: DeltaBuffer;
}
interface ToolResultScanContext {
scan: (toolName: string, text: string) => Promise<void>;
}
/**
* Per-tab map of tool_use_id tool name. Lets the tool_result handler
* know what tool produced the content (Read, Grep, Glob, Bash $B ...) so
* we can tag attack logs with the ingress source.
*/
const toolUseRegistry = new Map<string, { toolName: string; toolInput: unknown }>();
/**
* Extract plain-text content from a tool_result block. The Claude stream
* encodes it as either a string or an array of content blocks (text, image).
* We care about text images can't carry prompt injection at this layer.
*/
function extractToolResultText(content: unknown): string {
if (typeof content === 'string') return content;
if (!Array.isArray(content)) return '';
const parts: string[] = [];
for (const block of content) {
if (block && typeof block === 'object') {
const b = block as Record<string, unknown>;
if (b.type === 'text' && typeof b.text === 'string') parts.push(b.text);
}
}
return parts.join('\n');
}
/**
* Tools whose outputs should be ML-scanned. Bash/$B outputs already get
* scanned via the page-content flow. Read/Glob/Grep outputs have been
* uncovered Codex review flagged this gap. Adding coverage here closes it.
*/
const SCANNED_TOOLS = new Set(['Read', 'Grep', 'Glob', 'Bash', 'WebFetch']);
async function handleStreamEvent(event: any, tabId?: number, canaryCtx?: CanaryContext, toolResultScanCtx?: ToolResultScanContext): Promise<void> {
// Canary check runs BEFORE any outbound send — we never want to relay
// a leaked token to the sidepanel UI.
if (canaryCtx) {
const channel = detectCanaryLeak(event, canaryCtx.canary, canaryCtx.deltaBuf);
if (channel) {
canaryCtx.onLeak(channel);
return; // drop the event — never relay content that leaked the canary
}
}
if (event.type === 'system' && event.session_id) {
// Relay claude session ID for --resume support
await sendEvent({ type: 'system', claudeSessionId: event.session_id }, tabId);
}
if (event.type === 'assistant' && event.message?.content) {
for (const block of event.message.content) {
if (block.type === 'tool_use') {
// Register the tool_use so we can correlate tool_results back to
// the originating tool when they arrive in the next user-role message.
if (block.id) toolUseRegistry.set(block.id, { toolName: block.name, toolInput: block.input });
await sendEvent({ type: 'tool_use', tool: block.name, input: summarizeToolInput(block.name, block.input) }, tabId);
} else if (block.type === 'text' && block.text) {
await sendEvent({ type: 'text', text: block.text }, tabId);
}
}
}
// Tool results come back in user-role messages. Content can be a string
// or an array of typed content blocks.
if (event.type === 'user' && event.message?.content) {
for (const block of event.message.content) {
if (block && typeof block === 'object' && block.type === 'tool_result') {
const meta = block.tool_use_id ? toolUseRegistry.get(block.tool_use_id) : null;
const toolName = meta?.toolName ?? 'Unknown';
const text = extractToolResultText(block.content);
// Scan this tool output with the ML classifier if the tool is in
// the SCANNED_TOOLS set and the content is non-trivial.
if (SCANNED_TOOLS.has(toolName) && text.length >= 32 && toolResultScanCtx) {
// Fire-and-forget — never block the stream handler. If BLOCK
// fires, onToolResultBlock handles kill + emit.
toolResultScanCtx.scan(toolName, text).catch(() => {});
}
if (block.tool_use_id) toolUseRegistry.delete(block.tool_use_id);
}
}
}
if (event.type === 'content_block_start' && event.content_block?.type === 'tool_use') {
if (event.content_block.id) {
toolUseRegistry.set(event.content_block.id, {
toolName: event.content_block.name,
toolInput: event.content_block.input,
});
}
await sendEvent({ type: 'tool_use', tool: event.content_block.name, input: summarizeToolInput(event.content_block.name, event.content_block.input) }, tabId);
}
if (event.type === 'content_block_delta' && event.delta?.type === 'text_delta' && event.delta.text) {
await sendEvent({ type: 'text_delta', text: event.delta.text }, tabId);
}
// Relay tool results so the sidebar can show what happened
if (event.type === 'content_block_delta' && event.delta?.type === 'input_json_delta') {
// Tool input streaming — skip, we already announced the tool
}
if (event.type === 'result') {
await sendEvent({ type: 'result', text: event.result || '' }, tabId);
}
// Tool result events — summarize and relay
if (event.type === 'tool_result' || (event.type === 'assistant' && event.message?.content)) {
// Tool results come in the next assistant turn — handled above
}
}
/**
* Fire the prompt-injection-detected event to the server. This terminates
* the session from the sidepanel's perspective and renders the canary leak
* banner. Also logs locally (salted hash + domain only) and fires telemetry
* if configured.
*/
async function onCanaryLeaked(params: {
tabId: number;
channel: string;
canary: string;
pageUrl: string;
}): Promise<void> {
const { tabId, channel, canary, pageUrl } = params;
const domain = extractDomain(pageUrl);
console.warn(`[sidebar-agent] CANARY LEAK detected on ${channel} for tab ${tabId} (domain=${domain || 'unknown'})`);
// Local log — salted hash + domain only, never the payload
logAttempt({
ts: new Date().toISOString(),
urlDomain: domain,
payloadHash: hashPayload(canary), // hash the canary, not the payload (which might be leaked content)
confidence: 1.0,
layer: 'canary',
verdict: 'block',
});
// Broadcast to sidepanel so it can render the approved banner
await sendEvent({
type: 'security_event',
verdict: 'block',
reason: 'canary_leaked',
layer: 'canary',
channel,
domain,
}, tabId);
// Also emit agent_error so the sidepanel's existing error surface
// reflects that the session terminated. Keeps old clients working.
await sendEvent({
type: 'agent_error',
error: `Session terminated — prompt injection detected${domain ? ` from ${domain}` : ''}`,
}, tabId);
}
/**
* Pre-spawn ML scan of the user message. If the classifier fires at BLOCK,
* we log the attempt, emit a security_event to the sidepanel, and DO NOT
* spawn claude. Returns true if the scan blocked the session.
*
* Fail-open: any classifier error or degraded state returns false (safe) so
* the sidebar keeps working. The architectural controls (XML framing +
* command allowlist, live in server.ts:554-577) still defend.
*/
async function preSpawnSecurityCheck(entry: QueueEntry): Promise<boolean> {
const { message, canary, pageUrl, tabId } = entry;
if (!message || message.length === 0) return false;
const tid = tabId ?? 0;
// L4: scan the user message for direct injection patterns (TestSavantAI)
// L4c: also scan with DeBERTa-v3 when ensemble is enabled (opt-in)
const [contentSignal, debertaSignal] = await Promise.all([
scanPageContent(message),
scanPageContentDeberta(message),
]);
const signals: LayerSignal[] = [contentSignal, debertaSignal];
// L4b: only bother with Haiku if another layer already lit up at >= LOG_ONLY.
// Saves ~70% of Haiku calls per plan §E1 "gating optimization".
if (shouldRunTranscriptCheck(signals)) {
const transcriptSignal = await checkTranscript({
user_message: message,
tool_calls: [], // no tool calls yet at session start
});
signals.push(transcriptSignal);
}
const result = combineVerdict(signals);
if (result.verdict !== 'block') return false;
// BLOCK verdict. Log + emit + refuse to spawn.
const domain = extractDomain(pageUrl ?? '');
const leaderSignal = signals.reduce((a, b) => (a.confidence > b.confidence ? a : b));
logAttempt({
ts: new Date().toISOString(),
urlDomain: domain,
payloadHash: hashPayload(message),
confidence: result.confidence,
layer: leaderSignal.layer,
verdict: 'block',
});
console.warn(`[sidebar-agent] Pre-spawn BLOCK (${result.reason}) for tab ${tid}, confidence=${result.confidence.toFixed(3)}`);
await sendEvent({
type: 'security_event',
verdict: 'block',
reason: result.reason ?? 'ml_classifier',
layer: leaderSignal.layer,
confidence: result.confidence,
domain,
}, tid);
await sendEvent({
type: 'agent_error',
error: `Session blocked — prompt injection detected${domain ? ` from ${domain}` : ' in your message'}`,
}, tid);
return true;
}
async function askClaude(queueEntry: QueueEntry): Promise<void> {
const { prompt, args, stateFile, cwd, tabId, canary, pageUrl } = queueEntry;
const tid = tabId ?? 0;
processingTabs.add(tid);
await sendEvent({ type: 'agent_start' }, tid);
// Pre-spawn ML scan: if the user message trips the ensemble, refuse to
// spawn claude. Fail-open on classifier errors.
if (await preSpawnSecurityCheck(queueEntry)) {
processingTabs.delete(tid);
return;
}
return new Promise((resolve) => {
// Canary context is set after proc is spawned (needs proc reference for kill).
let canaryCtx: CanaryContext | undefined;
let canaryTriggered = false;
// Use args from queue entry (server sets --model, --allowedTools, prompt framing).
// Fall back to defaults only if queue entry has no args (backward compat).
// Write doesn't expand attack surface beyond what Bash already provides.
// The security boundary is the localhost-only message path, not the tool allowlist.
let claudeArgs = args || ['-p', prompt, '--output-format', 'stream-json', '--verbose',
'--allowedTools', 'Bash,Read,Glob,Grep,Write'];
// Validate cwd exists — queue may reference a stale worktree
let effectiveCwd = cwd || process.cwd();
try { fs.accessSync(effectiveCwd); } catch (err: any) {
console.warn('[sidebar-agent] Worktree path inaccessible, falling back to cwd:', effectiveCwd, err.message);
effectiveCwd = process.cwd();
}
// Clear any stale cancel signal for this tab before starting
const cancelFile = cancelFileForTab(tid);
safeUnlink(cancelFile);
const proc = spawn('claude', claudeArgs, {
stdio: ['pipe', 'pipe', 'pipe'],
cwd: effectiveCwd,
env: {
...process.env,
BROWSE_STATE_FILE: stateFile || '',
// Connect to the existing headed browse server, never start a new one.
// BROWSE_PORT tells the CLI which port to check.
// BROWSE_NO_AUTOSTART prevents spawning an invisible headless browser
// if the headed server is down — fail fast with a clear error instead.
BROWSE_PORT: process.env.BROWSE_PORT || '34567',
BROWSE_NO_AUTOSTART: '1',
// Pin this agent to its tab — prevents cross-tab interference
// when multiple agents run simultaneously
BROWSE_TAB: String(tid),
},
});
// Track active procs so kill-file polling can terminate them
activeProcs.set(tid, proc);
activeProc = proc;
proc.stdin.end();
// Now that proc exists, set up the canary-leak handler. It fires at most
// once; on fire we kill the subprocess, emit security_event + agent_error,
// and let the normal close handler resolve the promise.
if (canary) {
canaryCtx = {
canary,
pageUrl: pageUrl ?? '',
deltaBuf: { text_delta: '', input_json_delta: '' },
onLeak: (channel: string) => {
if (canaryTriggered) return;
canaryTriggered = true;
onCanaryLeaked({ tabId: tid, channel, canary, pageUrl: pageUrl ?? '' });
try { proc.kill('SIGTERM'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; }
setTimeout(() => {
try { proc.kill('SIGKILL'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; }
}, 2000);
},
};
}
// Tool-result ML scan context. Addresses the Codex review gap: Read,
// Grep, Glob, and WebFetch outputs enter Claude's context without
// passing through the Bash $B pipeline that content-security.ts
// already wraps. Scan them here.
let toolResultBlockFired = false;
const toolResultScanCtx: ToolResultScanContext = {
scan: async (toolName: string, text: string) => {
if (toolResultBlockFired) return;
// Parallel L4 + L4c ensemble scan (DeBERTa no-op when disabled).
// We run L4/L4c AND Haiku in parallel on tool outputs regardless of
// L4's score, because BrowseSafe-Bench shows L4 (TestSavantAI) has
// low recall on browser-agent-specific attacks (~15% at v1). Gating
// Haiku on L4 meant our best signal almost never ran. The cost is
// ~$0.002 + ~300ms per tool output, bounded by the Haiku timeout
// and offset by Haiku actually seeing the real attack context.
//
// Haiku only runs when the Claude CLI is available (checkHaikuAvailable
// caches the probe). In environments without it, the call returns a
// degraded signal and the verdict falls back to L4 alone.
const [contentSignal, debertaSignal, transcriptSignal] = await Promise.all([
scanPageContent(text),
scanPageContentDeberta(text),
checkTranscript({
user_message: queueEntry.message ?? '',
tool_calls: [{ tool_name: toolName, tool_input: {} }],
tool_output: text,
}),
]);
const signals: LayerSignal[] = [contentSignal, debertaSignal, transcriptSignal];
const result = combineVerdict(signals, { toolOutput: true });
if (result.verdict !== 'block') return;
toolResultBlockFired = true;
const domain = extractDomain(pageUrl ?? '');
const payloadHash = hashPayload(text.slice(0, 4096));
// Log pending — if the user overrides, we'll update via a separate
// log line. The attempts.jsonl is append-only so both entries survive.
logAttempt({
ts: new Date().toISOString(),
urlDomain: domain,
payloadHash,
confidence: result.confidence,
layer: 'testsavant_content',
verdict: 'block',
});
console.warn(`[sidebar-agent] Tool-result BLOCK on ${toolName} for tab ${tid} (confidence=${result.confidence.toFixed(3)}) — awaiting user decision`);
// Surface a REVIEWABLE block event. Sidepanel renders the suspected
// text + layer scores + [Allow and continue] / [Block session] buttons.
// The user has 60s to decide; default is BLOCK (safe fallback).
const layerScores = signals
.filter((s) => s.confidence > 0)
.map((s) => ({ layer: s.layer, confidence: s.confidence }));
await sendEvent({
type: 'security_event',
verdict: 'block',
reason: 'tool_result_ml',
layer: 'testsavant_content',
confidence: result.confidence,
domain,
tool: toolName,
reviewable: true,
suspected_text: excerptForReview(text),
signals: layerScores,
}, tid);
// Poll for the user's decision. Default to BLOCK on timeout.
const REVIEW_TIMEOUT_MS = 60_000;
const POLL_MS = 500;
clearDecision(tid); // clear any stale decision from a prior session
const deadline = Date.now() + REVIEW_TIMEOUT_MS;
let decision: 'allow' | 'block' = 'block';
let decisionReason = 'timeout';
while (Date.now() < deadline) {
const rec = readDecision(tid);
if (rec?.decision === 'allow' || rec?.decision === 'block') {
decision = rec.decision;
decisionReason = rec.reason ?? 'user';
break;
}
await new Promise((r) => setTimeout(r, POLL_MS));
}
clearDecision(tid);
if (decision === 'allow') {
// User overrode. Log the override so the audit trail captures it.
// toolResultBlockFired stays true so we don't re-prompt within the
// same message — one override per BLOCK event.
logAttempt({
ts: new Date().toISOString(),
urlDomain: domain,
payloadHash,
confidence: result.confidence,
layer: 'testsavant_content',
verdict: 'user_overrode',
});
await sendEvent({
type: 'security_event',
verdict: 'user_overrode',
reason: 'tool_result_ml',
layer: 'testsavant_content',
confidence: result.confidence,
domain,
tool: toolName,
}, tid);
console.warn(`[sidebar-agent] Tab ${tid}: user overrode BLOCK — session continues`);
// Let the block stay consumed; reset the flag so subsequent tool
// results get scanned fresh.
toolResultBlockFired = false;
return;
}
// User chose BLOCK (or timed out). Kill the session as before.
await sendEvent({
type: 'agent_error',
error: `Session terminated — prompt injection detected in ${toolName} output${decisionReason === 'timeout' ? ' (review timeout)' : ''}`,
}, tid);
try { proc.kill('SIGTERM'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; }
setTimeout(() => {
try { proc.kill('SIGKILL'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; }
}, 2000);
},
};
// Poll for per-tab cancel signal from server's killAgent()
const cancelCheck = setInterval(() => {
try {
if (fs.existsSync(cancelFile)) {
console.log(`[sidebar-agent] Cancel signal received for tab ${tid} — killing claude subprocess`);
try { proc.kill('SIGTERM'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; }
setTimeout(() => { try { proc.kill('SIGKILL'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; } }, 3000);
fs.unlinkSync(cancelFile);
clearInterval(cancelCheck);
}
} catch (err: any) { if (err?.code !== 'ENOENT') throw err; }
}, 500);
let buffer = '';
proc.stdout.on('data', (data: Buffer) => {
buffer += data.toString();
const lines = buffer.split('\n');
buffer = lines.pop() || '';
for (const line of lines) {
if (!line.trim()) continue;
try { handleStreamEvent(JSON.parse(line), tid, canaryCtx, toolResultScanCtx); } catch (err: any) {
console.error(`[sidebar-agent] Tab ${tid}: Failed to parse stream line:`, line.slice(0, 100), err.message);
}
}
});
let stderrBuffer = '';
proc.stderr.on('data', (data: Buffer) => {
stderrBuffer += data.toString();
});
proc.on('close', (code) => {
clearInterval(cancelCheck);
activeProc = null;
activeProcs.delete(tid);
if (buffer.trim()) {
try { handleStreamEvent(JSON.parse(buffer), tid, canaryCtx, toolResultScanCtx); } catch (err: any) {
console.error(`[sidebar-agent] Tab ${tid}: Failed to parse final buffer:`, buffer.slice(0, 100), err.message);
}
}
const doneEvent: Record<string, any> = { type: 'agent_done' };
if (code !== 0 && stderrBuffer.trim()) {
doneEvent.stderr = stderrBuffer.trim().slice(-500);
}
sendEvent(doneEvent, tid).then(() => {
processingTabs.delete(tid);
resolve();
});
});
proc.on('error', (err) => {
clearInterval(cancelCheck);
activeProc = null;
const errorMsg = stderrBuffer.trim()
? `${err.message}\nstderr: ${stderrBuffer.trim().slice(-500)}`
: err.message;
sendEvent({ type: 'agent_error', error: errorMsg }, tid).then(() => {
processingTabs.delete(tid);
resolve();
});
});
// Timeout (default 300s / 5 min — multi-page tasks need time)
const timeoutMs = parseInt(process.env.SIDEBAR_AGENT_TIMEOUT || '300000', 10);
setTimeout(() => {
try { proc.kill('SIGTERM'); } catch (killErr: any) {
console.warn(`[sidebar-agent] Tab ${tid}: Failed to kill timed-out process:`, killErr.message);
}
setTimeout(() => { try { proc.kill('SIGKILL'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; } }, 3000);
const timeoutMsg = stderrBuffer.trim()
? `Timed out after ${timeoutMs / 1000}s\nstderr: ${stderrBuffer.trim().slice(-500)}`
: `Timed out after ${timeoutMs / 1000}s`;
sendEvent({ type: 'agent_error', error: timeoutMsg }, tid).then(() => {
processingTabs.delete(tid);
resolve();
});
}, timeoutMs);
});
}
// ─── Poll loop ───────────────────────────────────────────────────
function countLines(): number {
try {
return fs.readFileSync(QUEUE, 'utf-8').split('\n').filter(Boolean).length;
} catch (err: any) {
console.error('[sidebar-agent] Failed to read queue file:', err.message);
return 0;
}
}
function readLine(n: number): string | null {
try {
const lines = fs.readFileSync(QUEUE, 'utf-8').split('\n').filter(Boolean);
return lines[n - 1] || null;
} catch (err: any) {
console.error(`[sidebar-agent] Failed to read queue line ${n}:`, err.message);
return null;
}
}
async function poll() {
const current = countLines();
if (current <= lastLine) return;
while (lastLine < current) {
lastLine++;
const line = readLine(lastLine);
if (!line) continue;
let parsed: unknown;
try { parsed = JSON.parse(line); } catch (err: any) {
console.warn(`[sidebar-agent] Skipping malformed queue entry at line ${lastLine}:`, line.slice(0, 80), err.message);
continue;
}
if (!isValidQueueEntry(parsed)) {
console.warn(`[sidebar-agent] Skipping invalid queue entry at line ${lastLine}: failed schema validation`);
continue;
}
const entry = parsed;
const tid = entry.tabId ?? 0;
// Skip if this tab already has an agent running — server queues per-tab
if (processingTabs.has(tid)) continue;
console.log(`[sidebar-agent] Processing tab ${tid}: "${entry.message}"`);
// Write to inbox so workspace agent can pick it up
writeToInbox(entry.message || entry.prompt, entry.pageUrl, entry.sessionId);
// Fire and forget — each tab's agent runs concurrently
askClaude(entry).catch((err) => {
console.error(`[sidebar-agent] Error on tab ${tid}:`, err);
sendEvent({ type: 'agent_error', error: String(err) }, tid);
});
}
}
// ─── Main ────────────────────────────────────────────────────────
function pollKillFile(): void {
try {
const stat = fs.statSync(KILL_FILE);
const mtime = stat.mtimeMs;
if (mtime > lastKillTs) {
lastKillTs = mtime;
if (activeProcs.size > 0) {
console.log(`[sidebar-agent] Kill signal received — terminating ${activeProcs.size} active agent(s)`);
for (const [tid, proc] of activeProcs) {
try { proc.kill('SIGTERM'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; }
setTimeout(() => { try { proc.kill('SIGKILL'); } catch (err: any) { if (err?.code !== 'ESRCH') throw err; } }, 2000);
processingTabs.delete(tid);
}
activeProcs.clear();
}
}
} catch {
// Kill file doesn't exist yet — normal state
}
}
async function main() {
const dir = path.dirname(QUEUE);
fs.mkdirSync(dir, { recursive: true, mode: 0o700 });
if (!fs.existsSync(QUEUE)) fs.writeFileSync(QUEUE, '', { mode: 0o600 });
try { fs.chmodSync(QUEUE, 0o600); } catch (err: any) { if (err?.code !== 'ENOENT') throw err; }
lastLine = countLines();
await refreshToken();
console.log(`[sidebar-agent] Started. Watching ${QUEUE} from line ${lastLine}`);
console.log(`[sidebar-agent] Server: ${SERVER_URL}`);
console.log(`[sidebar-agent] Browse binary: ${B}`);
// If GSTACK_SECURITY_ENSEMBLE=deberta is set, also warm the DeBERTa-v3
// ensemble classifier. Fire-and-forget alongside TestSavantAI — they
// warm in parallel. No-op when the env var is unset.
loadDeberta((msg) => console.log(`[security-classifier] ${msg}`))
.catch((err) => console.warn('[sidebar-agent] DeBERTa warmup failed:', err?.message));
// Warm up the ML classifier in the background. First call triggers a 112MB
// download (~30s on average broadband). Non-blocking — the sidebar stays
// functional on cold start; classifier just reports 'off' until warmed.
//
// On warmup completion (success or failure), write the classifier status to
// ~/.gstack/security/session-state.json so server.ts's /health endpoint can
// report it to the sidepanel for shield icon rendering.
loadTestsavant((msg) => console.log(`[security-classifier] ${msg}`))
.then(() => {
const s = getClassifierStatus();
console.log(`[sidebar-agent] Classifier warmup complete: ${JSON.stringify(s)}`);
const existing = readSessionState();
writeSessionState({
sessionId: existing?.sessionId ?? String(process.pid),
canary: existing?.canary ?? '',
warnedDomains: existing?.warnedDomains ?? [],
classifierStatus: s,
lastUpdated: new Date().toISOString(),
});
})
.catch((err) => console.warn('[sidebar-agent] Classifier warmup failed (degraded mode):', err?.message));
setInterval(poll, POLL_MS);
setInterval(pollKillFile, POLL_MS);
}
main().catch(console.error);
+556
View File
@@ -0,0 +1,556 @@
/**
* Terminal Agent PTY-backed Claude Code terminal for the gstack browser
* sidebar. Translates the phoenix gbrowser PTY (cmd/gbd/terminal.go) into
* Bun, with a few changes informed by codex's outside-voice review:
*
* - Lives in a separate non-compiled bun process from sidebar-agent.ts so
* a bug in WS framing or PTY cleanup can't take down the chat path.
* - Binds 127.0.0.1 only never on the dual-listener tunnel surface.
* - Origin validation on the WS upgrade is REQUIRED (not defense-in-depth)
* because a localhost shell WS is a real cross-site WebSocket-hijacking
* target.
* - Cookie-based auth via /internal/grant from the parent server, not a
* token in /health.
* - Lazy spawn: claude PTY is not spawned until the WS receives its first
* data frame. Sidebar opens that never type don't burn a claude session.
* - PTY dies with WS close (one PTY per WS). v1.1 may add session
* survival; for v1 we match phoenix's lifecycle.
*
* The PTY uses Bun's `terminal:` spawn option (verified at impl time on
* Bun 1.3.10): pass cols/rows + a data callback; write input via
* `proc.terminal.write(buf)`; resize via `proc.terminal.resize(cols, rows)`.
*/
import * as fs from 'fs';
import * as path from 'path';
import * as crypto from 'crypto';
import { safeUnlink } from './error-handling';
const STATE_FILE = process.env.BROWSE_STATE_FILE || path.join(process.env.HOME || '/tmp', '.gstack', 'browse.json');
const PORT_FILE = path.join(path.dirname(STATE_FILE), 'terminal-port');
const BROWSE_SERVER_PORT = parseInt(process.env.BROWSE_SERVER_PORT || '0', 10);
const EXTENSION_ID = process.env.BROWSE_EXTENSION_ID || ''; // optional: tighten Origin check
const INTERNAL_TOKEN = crypto.randomBytes(32).toString('base64url'); // shared with parent server via env at spawn
// In-memory cookie token registry. Parent posts /internal/grant after
// /pty-session; we validate WS cookies against this set.
const validTokens = new Set<string>();
// Active PTY session per WS. One terminal per connection. Codex finding #4:
// uncaught handlers below catch bugs in framing/cleanup so they don't kill
// the listener loop.
process.on('uncaughtException', (err) => {
console.error('[terminal-agent] uncaughtException:', err);
});
process.on('unhandledRejection', (reason) => {
console.error('[terminal-agent] unhandledRejection:', reason);
});
interface PtySession {
proc: any | null; // Bun.Subprocess once spawned
cols: number;
rows: number;
cookie: string;
spawned: boolean;
}
const sessions = new WeakMap<any, PtySession>(); // ws -> session
/** Find claude on PATH. */
function findClaude(): string | null {
// Test-only override. Lets the integration tests spawn /bin/bash instead
// of requiring claude to be installed on every CI runner. NEVER read in
// production (sidebar UI). Documented in browse/test/terminal-agent-integration.test.ts.
const override = process.env.BROWSE_TERMINAL_BINARY;
if (override && fs.existsSync(override)) return override;
// Bun.which is sync and respects PATH. Falls back to a small list of
// common install locations if PATH is stripped (e.g., launched from
// Conductor with a minimal env).
const which = (Bun as any).which?.('claude');
if (which) return which;
const candidates = [
'/opt/homebrew/bin/claude',
'/usr/local/bin/claude',
`${process.env.HOME}/.local/bin/claude`,
`${process.env.HOME}/.bun/bin/claude`,
`${process.env.HOME}/.npm-global/bin/claude`,
];
for (const c of candidates) {
try { fs.accessSync(c, fs.constants.X_OK); return c; } catch {}
}
return null;
}
/** Probe + persist claude availability for the bootstrap card. */
function writeClaudeAvailable(): void {
const stateDir = path.dirname(STATE_FILE);
try { fs.mkdirSync(stateDir, { recursive: true, mode: 0o700 }); } catch {}
const found = findClaude();
const status = {
available: !!found,
path: found || undefined,
install_url: 'https://docs.anthropic.com/en/docs/claude-code',
checked_at: new Date().toISOString(),
};
const target = path.join(stateDir, 'claude-available.json');
const tmp = path.join(stateDir, `.tmp-claude-${process.pid}`);
try {
fs.writeFileSync(tmp, JSON.stringify(status, null, 2), { mode: 0o600 });
fs.renameSync(tmp, target);
} catch {
safeUnlink(tmp);
}
}
/**
* System-prompt hint passed to claude via --append-system-prompt. Tells
* claude what tab-awareness affordances exist in this session so it
* doesn't have to discover them by trial. The user can override anything
* here just by saying so system prompt is a soft hint, not a contract.
*
* Two paths claude has:
* 1. Read live state from <stateDir>/tabs.json + active-tab.json
* (updated continuously by the gstack browser extension).
* 2. Run $B tab, $B tabs, $B tab-each <command> to act on tabs. The
* tab-each helper fans a single command across every open tab and
* returns per-tab results as JSON.
*/
function buildTabAwarenessHint(stateDir: string): string {
const tabsFile = path.join(stateDir, 'tabs.json');
const activeFile = path.join(stateDir, 'active-tab.json');
return [
'You are running inside the gstack browser sidebar with live access to the user\'s browser tabs.',
'',
'Tab state files (kept fresh automatically by the extension):',
` ${tabsFile} — all open tabs (id, url, title, active, pinned)`,
` ${activeFile} — the currently active tab`,
'Read these any time the user asks about "tabs", "the current page", or anything multi-tab. Do NOT shell out to $B tabs just to learn what\'s open — read the file.',
'',
'Tab manipulation commands (via $B):',
' $B tab <id> — switch to a tab',
' $B newtab [url] — open a new tab',
' $B closetab [id] — close a tab (current if no id)',
' $B tab-each <command> — fan out a command across every tab; returns JSON results',
'',
'When the user asks for multi-tab work, prefer $B tab-each. Examples:',
' $B tab-each snapshot -i — grab a snapshot from every tab',
' $B tab-each text — pull clean text from every tab',
' $B tab-each title — list every tab\'s title',
'',
'You\'re in a real terminal with a real PTY — slash commands, /resume, ANSI colors all work as in a normal claude session.',
].join('\n');
}
/** Spawn claude in a PTY. Returns null if claude not on PATH. */
function spawnClaude(cols: number, rows: number, onData: (chunk: Buffer) => void) {
const claudePath = findClaude();
if (!claudePath) return null;
// Match phoenix env so claude knows which browse server to talk to and
// doesn't try to autostart its own. BROWSE_HEADED=1 keeps the existing
// headed-mode browser; BROWSE_NO_AUTOSTART prevents claude's gstack
// tooling from racing to spawn another server.
const env: Record<string, string> = {
...process.env as any,
BROWSE_PORT: String(BROWSE_SERVER_PORT),
BROWSE_STATE_FILE: STATE_FILE,
BROWSE_NO_AUTOSTART: '1',
BROWSE_HEADED: '1',
TERM: 'xterm-256color',
COLORTERM: 'truecolor',
};
// --append-system-prompt is the right injection surface (per `claude --help`):
// it gets appended to the model's system prompt, so claude treats this as
// contextual guidance, not a user message. Don't use a leading PTY write
// for this — that would show up as if the user typed the hint, polluting
// the visible transcript.
const stateDir = path.dirname(STATE_FILE);
const tabHint = buildTabAwarenessHint(stateDir);
const proc = (Bun as any).spawn([claudePath, '--append-system-prompt', tabHint], {
terminal: {
rows,
cols,
data(_terminal: any, chunk: Buffer) { onData(chunk); },
},
env,
});
return proc;
}
/** Cleanup a PTY session: SIGINT, then SIGKILL after 3s. */
function disposeSession(session: PtySession): void {
try { session.proc?.terminal?.close?.(); } catch {}
if (session.proc?.pid) {
try { session.proc.kill?.('SIGINT'); } catch {}
setTimeout(() => {
try {
if (session.proc && !session.proc.killed) session.proc.kill?.('SIGKILL');
} catch {}
}, 3000);
}
session.proc = null;
session.spawned = false;
}
/**
* Build the HTTP server. Two routes:
* POST /internal/grant parent server pushes a fresh cookie token
* GET /ws extension upgrades to WebSocket (PTY transport)
*
* Everything else returns 404. The listener binds 127.0.0.1 only.
*/
function buildServer() {
return Bun.serve({
hostname: '127.0.0.1',
port: 0,
idleTimeout: 0, // PTY connections are long-lived; default idleTimeout would kill them
fetch(req, server) {
const url = new URL(req.url);
// /internal/grant — loopback-only handshake from parent server.
if (url.pathname === '/internal/grant' && req.method === 'POST') {
const auth = req.headers.get('authorization');
if (auth !== `Bearer ${INTERNAL_TOKEN}`) {
return new Response('forbidden', { status: 403 });
}
return req.json().then((body: any) => {
if (typeof body?.token === 'string' && body.token.length > 16) {
validTokens.add(body.token);
}
return new Response('ok');
}).catch(() => new Response('bad', { status: 400 }));
}
// /internal/revoke — drop a token (called on WS close or bootstrap reload)
if (url.pathname === '/internal/revoke' && req.method === 'POST') {
const auth = req.headers.get('authorization');
if (auth !== `Bearer ${INTERNAL_TOKEN}`) {
return new Response('forbidden', { status: 403 });
}
return req.json().then((body: any) => {
if (typeof body?.token === 'string') validTokens.delete(body.token);
return new Response('ok');
}).catch(() => new Response('bad', { status: 400 }));
}
// /claude-available — bootstrap card hits this when user clicks "I installed it".
if (url.pathname === '/claude-available' && req.method === 'GET') {
writeClaudeAvailable();
const found = findClaude();
return new Response(JSON.stringify({ available: !!found, path: found }), {
status: 200,
headers: { 'Content-Type': 'application/json' },
});
}
// /ws — WebSocket upgrade. CRITICAL gates:
// (1) Origin must be chrome-extension://<id>. Cross-site WS hijacking
// defense — required, not optional.
// (2) Token must be in validTokens. We accept the token via two
// transports for compatibility:
// - Sec-WebSocket-Protocol (preferred for browsers — the only
// auth header settable from the browser WebSocket API)
// - Cookie gstack_pty (works for non-browser callers and
// same-port browser callers; doesn't survive the cross-port
// jump from server.ts:34567 to the agent's random port
// when SameSite=Strict is set)
// Either path works; both verify against the same in-memory
// validTokens Set, populated by the parent server's
// authenticated /pty-session → /internal/grant chain.
if (url.pathname === '/ws') {
const origin = req.headers.get('origin') || '';
const isExtensionOrigin = origin.startsWith('chrome-extension://');
if (!isExtensionOrigin) {
return new Response('forbidden origin', { status: 403 });
}
if (EXTENSION_ID && origin !== `chrome-extension://${EXTENSION_ID}`) {
return new Response('forbidden origin', { status: 403 });
}
// Try Sec-WebSocket-Protocol first. Format: a single token, possibly
// with a `gstack-pty.` prefix (which we strip). Browsers send a
// comma-separated list when multiple were requested; we pick the
// first that matches a known token.
const protoHeader = req.headers.get('sec-websocket-protocol') || '';
let token: string | null = null;
let acceptedProtocol: string | null = null;
for (const raw of protoHeader.split(',').map(s => s.trim()).filter(Boolean)) {
const candidate = raw.startsWith('gstack-pty.') ? raw.slice('gstack-pty.'.length) : raw;
if (validTokens.has(candidate)) {
token = candidate;
acceptedProtocol = raw;
break;
}
}
// Fallback: Cookie gstack_pty (legacy / non-browser callers).
if (!token) {
const cookieHeader = req.headers.get('cookie') || '';
for (const part of cookieHeader.split(';')) {
const [name, ...rest] = part.trim().split('=');
if (name === 'gstack_pty') {
const candidate = rest.join('=') || null;
if (candidate && validTokens.has(candidate)) {
token = candidate;
}
break;
}
}
}
if (!token) {
return new Response('unauthorized', { status: 401 });
}
const upgraded = server.upgrade(req, {
data: { cookie: token },
// Echo the protocol back so the browser accepts the upgrade.
// Required when the client sends Sec-WebSocket-Protocol — the
// server MUST select one of the offered protocols, otherwise
// the browser closes the connection immediately.
...(acceptedProtocol ? { headers: { 'Sec-WebSocket-Protocol': acceptedProtocol } } : {}),
});
return upgraded ? undefined : new Response('upgrade failed', { status: 500 });
}
return new Response('not found', { status: 404 });
},
websocket: {
message(ws, raw) {
let session = sessions.get(ws);
if (!session) {
session = {
proc: null,
cols: 80,
rows: 24,
cookie: (ws.data as any)?.cookie || '',
spawned: false,
};
sessions.set(ws, session);
}
// Text frames are control messages: {type: "resize", cols, rows} or
// {type: "tabSwitch", tabId, url, title}. Binary frames are raw input
// bytes destined for the PTY stdin.
if (typeof raw === 'string') {
let msg: any;
try { msg = JSON.parse(raw); } catch { return; }
if (msg?.type === 'resize') {
const cols = Math.max(2, Math.floor(Number(msg.cols) || 80));
const rows = Math.max(2, Math.floor(Number(msg.rows) || 24));
session.cols = cols;
session.rows = rows;
try { session.proc?.terminal?.resize?.(cols, rows); } catch {}
return;
}
if (msg?.type === 'tabSwitch') {
handleTabSwitch(msg);
return;
}
if (msg?.type === 'tabState') {
handleTabState(msg);
return;
}
// Unknown text frame — ignore.
return;
}
// Binary input. Lazy-spawn claude on the first byte.
if (!session.spawned) {
session.spawned = true;
const proc = spawnClaude(session.cols, session.rows, (chunk) => {
try { ws.sendBinary(chunk); } catch {}
});
if (!proc) {
try {
ws.send(JSON.stringify({
type: 'error',
code: 'CLAUDE_NOT_FOUND',
message: 'claude CLI not on PATH. Install: https://docs.anthropic.com/en/docs/claude-code',
}));
ws.close(4404, 'claude not found');
} catch {}
return;
}
session.proc = proc;
// Watch for child exit so the WS closes cleanly when claude exits.
proc.exited?.then?.(() => {
try { ws.close(1000, 'pty exited'); } catch {}
});
}
try {
// raw is a Uint8Array; Bun.Terminal.write accepts string|Buffer.
// Convert to Buffer for safety.
session.proc?.terminal?.write?.(Buffer.from(raw as Uint8Array));
} catch (err) {
console.error('[terminal-agent] terminal.write failed:', err);
}
},
close(ws) {
const session = sessions.get(ws);
if (session) {
disposeSession(session);
if (session.cookie) {
// Drop the cookie so it can't be replayed against a new PTY.
validTokens.delete(session.cookie);
}
sessions.delete(ws);
}
},
},
});
}
/**
* Tab-switch helper: write the active tab to a state file (claude reads it)
* and notify the parent server so its activeTabId stays synced. Skips
* chrome:// and chrome-extension:// internal pages.
*/
/**
* Live tab snapshot. Writes <stateDir>/tabs.json (full list) and updates
* <stateDir>/active-tab.json (current active). claude can read these any
* time without invoking $B tabs saves a round-trip when the model just
* needs to check the landscape before deciding what to do.
*/
function handleTabState(msg: {
active?: { tabId?: number; url?: string; title?: string } | null;
tabs?: Array<{ tabId?: number; url?: string; title?: string; active?: boolean; windowId?: number; pinned?: boolean; audible?: boolean }>;
reason?: string;
}): void {
const stateDir = path.dirname(STATE_FILE);
try { fs.mkdirSync(stateDir, { recursive: true, mode: 0o700 }); } catch {}
// tabs.json — full list
if (Array.isArray(msg.tabs)) {
const payload = {
updatedAt: new Date().toISOString(),
reason: msg.reason || 'unknown',
tabs: msg.tabs.map(t => ({
tabId: t.tabId ?? null,
url: t.url || '',
title: t.title || '',
active: !!t.active,
windowId: t.windowId ?? null,
pinned: !!t.pinned,
audible: !!t.audible,
})),
};
const target = path.join(stateDir, 'tabs.json');
const tmp = path.join(stateDir, `.tmp-tabs-${process.pid}`);
try {
fs.writeFileSync(tmp, JSON.stringify(payload, null, 2), { mode: 0o600 });
fs.renameSync(tmp, target);
} catch {
safeUnlink(tmp);
}
}
// active-tab.json — single active tab. Skip chrome-internal pages so
// claude doesn't see chrome:// or chrome-extension:// URLs as
// "current target."
const active = msg.active;
if (active && active.url && !active.url.startsWith('chrome://') && !active.url.startsWith('chrome-extension://')) {
const ctxFile = path.join(stateDir, 'active-tab.json');
const tmp = path.join(stateDir, `.tmp-tab-${process.pid}`);
try {
fs.writeFileSync(tmp, JSON.stringify({
tabId: active.tabId ?? null,
url: active.url,
title: active.title ?? '',
}), { mode: 0o600 });
fs.renameSync(tmp, ctxFile);
} catch {
safeUnlink(tmp);
}
}
}
function handleTabSwitch(msg: { tabId?: number; url?: string; title?: string }): void {
const url = msg.url || '';
if (!url || url.startsWith('chrome://') || url.startsWith('chrome-extension://')) return;
const stateDir = path.dirname(STATE_FILE);
const ctxFile = path.join(stateDir, 'active-tab.json');
const tmp = path.join(stateDir, `.tmp-tab-${process.pid}`);
try {
fs.writeFileSync(tmp, JSON.stringify({
tabId: msg.tabId ?? null,
url,
title: msg.title ?? '',
}), { mode: 0o600 });
fs.renameSync(tmp, ctxFile);
} catch {
safeUnlink(tmp);
}
// Best-effort sync to parent server so its activeTabId tracking matches.
// No await; this is fire-and-forget.
if (BROWSE_SERVER_PORT > 0) {
fetch(`http://127.0.0.1:${BROWSE_SERVER_PORT}/command`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'Authorization': `Bearer ${readBrowseToken()}`,
},
body: JSON.stringify({
command: 'tab',
args: [String(msg.tabId ?? ''), '--no-focus'],
}),
}).catch(() => {});
}
}
function readBrowseToken(): string {
try {
const raw = fs.readFileSync(STATE_FILE, 'utf-8');
const j = JSON.parse(raw);
return j.token || '';
} catch { return ''; }
}
// Boot.
function main() {
writeClaudeAvailable();
const server = buildServer();
const port = (server as any).port || (server as any).address?.port;
if (!port) {
console.error('[terminal-agent] failed to bind: no port');
process.exit(1);
}
// Write port file atomically so the parent server can pick it up.
const dir = path.dirname(PORT_FILE);
try { fs.mkdirSync(dir, { recursive: true, mode: 0o700 }); } catch {}
const tmp = `${PORT_FILE}.tmp-${process.pid}`;
fs.writeFileSync(tmp, String(port), { mode: 0o600 });
fs.renameSync(tmp, PORT_FILE);
// Hand the parent the internal token so it can call /internal/grant.
// Parent learns INTERNAL_TOKEN via env (TERMINAL_AGENT_INTERNAL_TOKEN below).
// We just print it on stdout for the supervising process to pick up if it's
// not already in env. Defense against env races at spawn time.
console.log(`[terminal-agent] listening on 127.0.0.1:${port} pid=${process.pid}`);
// Cleanup port file on exit.
const cleanup = () => { safeUnlink(PORT_FILE); process.exit(0); };
process.on('SIGTERM', cleanup);
process.on('SIGINT', cleanup);
}
// Export the internal token so cli.ts can pass the SAME value to the parent
// server via env. Parent reads BROWSE_TERMINAL_INTERNAL_TOKEN and uses it
// for /internal/grant calls.
//
// In practice, the agent generates INTERNAL_TOKEN once at boot and writes it
// to a state file the parent reads. This avoids env-passing races. See main().
const INTERNAL_TOKEN_FILE = path.join(path.dirname(STATE_FILE), 'terminal-internal-token');
try {
fs.mkdirSync(path.dirname(INTERNAL_TOKEN_FILE), { recursive: true, mode: 0o700 });
fs.writeFileSync(INTERNAL_TOKEN_FILE, INTERNAL_TOKEN, { mode: 0o600 });
} catch {}
main();
+8 -32
View File
@@ -19,31 +19,10 @@ import { PAGE_CONTENT_COMMANDS } from '../src/commands';
const REPO_ROOT = path.resolve(__dirname, '..', '..');
describe('canary stream-chunk split detection', () => {
test('detectCanaryLeak uses rolling buffer across consecutive deltas', () => {
// Pull in the function via dynamic require so we don't re-export it
// from sidebar-agent.ts (it's internal on purpose).
const agentSource = fs.readFileSync(
path.join(REPO_ROOT, 'browse', 'src', 'sidebar-agent.ts'),
'utf-8',
);
// Contract: detectCanaryLeak accepts an optional DeltaBuffer and
// uses .slice(-(canary.length - 1)) to retain a rolling tail.
expect(agentSource).toContain('DeltaBuffer');
expect(agentSource).toMatch(/text_delta\s*=\s*combined\.slice\(-\(canary\.length - 1\)\)/);
expect(agentSource).toMatch(/input_json_delta\s*=\s*combined\.slice\(-\(canary\.length - 1\)\)/);
});
test('canary context initializes deltaBuf', () => {
const agentSource = fs.readFileSync(
path.join(REPO_ROOT, 'browse', 'src', 'sidebar-agent.ts'),
'utf-8',
);
// The askClaude call site must construct the buffer so the rolling
// detection actually runs.
expect(agentSource).toContain("deltaBuf: { text_delta: '', input_json_delta: '' }");
});
});
// canary stream-chunk split detection — tested detectCanaryLeak inside
// sidebar-agent.ts. Both the chat-stream pipeline and the function are
// gone (Terminal pane uses an interactive PTY; user keystrokes are the
// trust source, no chunked LLM stream to canary-scan).
describe('tool-output ensemble rule (single-layer BLOCK)', () => {
test('user-input context: single layer at BLOCK degrades to WARN', () => {
@@ -117,13 +96,10 @@ describe('transcript classifier tool_output parameter', () => {
expect(src).toContain('tool_output');
});
test('sidebar-agent passes tool text to transcript on tool-result scan', () => {
const src = fs.readFileSync(
path.join(REPO_ROOT, 'browse', 'src', 'sidebar-agent.ts'),
'utf-8',
);
expect(src).toContain('tool_output: text');
});
// sidebar-agent passed tool text to the transcript classifier on
// tool-result scans. That whole pipeline is gone — Terminal pane has
// no LLM stream to scan, and security-classifier.ts is dead code with
// no production caller (a separate v1.1+ cleanup TODO).
});
describe('GSTACK_SECURITY_OFF kill switch', () => {
+40 -224
View File
@@ -15,7 +15,13 @@ import * as os from 'os';
const META_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/meta-commands.ts'), 'utf-8');
const WRITE_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/write-commands.ts'), 'utf-8');
const SERVER_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/server.ts'), 'utf-8');
const AGENT_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/sidebar-agent.ts'), 'utf-8');
// sidebar-agent.ts was ripped (chat queue replaced by interactive PTY).
// AGENT_SRC kept as empty string so the legacy describe block below skips
// without crashing module load on a missing file.
const AGENT_SRC = (() => {
try { return fs.readFileSync(path.join(import.meta.dir, '../src/sidebar-agent.ts'), 'utf-8'); }
catch { return ''; }
})();
const SNAPSHOT_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/snapshot.ts'), 'utf-8');
const PATH_SECURITY_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/path-security.ts'), 'utf-8');
@@ -51,53 +57,12 @@ function extractFunction(src: string, name: string): string {
return src.slice(start);
}
// ─── Task 4: Agent queue poisoning — full schema validation + permissions ───
describe('Agent queue security', () => {
it('server queue directory must use restricted permissions', () => {
const queueSection = SERVER_SRC.slice(SERVER_SRC.indexOf('agentQueue'), SERVER_SRC.indexOf('agentQueue') + 2000);
expect(queueSection).toMatch(/0o700/);
});
it('sidebar-agent queue directory must use restricted permissions', () => {
// The mkdirSync for the queue dir lives in main() — search the main() body
const mainStart = AGENT_SRC.indexOf('async function main');
const queueSection = AGENT_SRC.slice(mainStart);
expect(queueSection).toMatch(/0o700/);
});
it('cli.ts queue file creation must use restricted permissions', () => {
const CLI_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/cli.ts'), 'utf-8');
const queueSection = CLI_SRC.slice(CLI_SRC.indexOf('queue') || 0, CLI_SRC.indexOf('queue') + 2000);
expect(queueSection).toMatch(/0o700|0o600|mode/);
});
it('queue reader must have a validator function covering all fields', () => {
// Extract ONLY the validator function body by walking braces
const validatorStart = AGENT_SRC.indexOf('function isValidQueueEntry');
expect(validatorStart).toBeGreaterThan(-1);
let depth = 0;
let bodyStart = AGENT_SRC.indexOf('{', validatorStart);
let bodyEnd = bodyStart;
for (let i = bodyStart; i < AGENT_SRC.length; i++) {
if (AGENT_SRC[i] === '{') depth++;
if (AGENT_SRC[i] === '}') depth--;
if (depth === 0) { bodyEnd = i + 1; break; }
}
const validatorBlock = AGENT_SRC.slice(validatorStart, bodyEnd);
expect(validatorBlock).toMatch(/prompt.*string/);
expect(validatorBlock).toMatch(/Array\.isArray/);
expect(validatorBlock).toMatch(/\.\./);
expect(validatorBlock).toContain('stateFile');
expect(validatorBlock).toContain('tabId');
expect(validatorBlock).toMatch(/number/);
expect(validatorBlock).toContain('null');
expect(validatorBlock).toContain('message');
expect(validatorBlock).toContain('pageUrl');
expect(validatorBlock).toContain('sessionId');
});
});
// ─── Agent queue security ──────────────────────────────────────────────────
// Original block validated the chat queue's filesystem permissions and
// schema validator on sidebar-agent.ts. Both are gone (chat queue ripped
// in favor of the interactive Terminal PTY). The remaining 0o700 / 0o600
// invariants on extension queue paths are now covered by terminal-agent
// integration tests and the sidebar-tabs regression suite.
// ─── Shared source reads for CSS validator tests ────────────────────────────
const CDP_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/cdp-inspector.ts'), 'utf-8');
@@ -325,30 +290,13 @@ describe('Round-2 finding 2: snapshot.ts annotated path uses realpathSync', () =
});
});
// ─── Round-2 finding 3: stateFile path traversal check in isValidQueueEntry
describe('Round-2 finding 3: isValidQueueEntry checks stateFile for path traversal', () => {
it('isValidQueueEntry checks stateFile for .. traversal sequences', () => {
const fn = extractFunction(AGENT_SRC, 'isValidQueueEntry');
expect(fn).toBeTruthy();
// Must check stateFile for '..' — find the stateFile block and look for '..' string
const stateFileIdx = fn.indexOf('stateFile');
expect(stateFileIdx).toBeGreaterThan(-1);
const stateFileBlock = fn.slice(stateFileIdx, stateFileIdx + 200);
// The block must contain a check for the two-dot traversal sequence
expect(stateFileBlock).toMatch(/'\.\.'|"\.\."|\.\./);
});
it('isValidQueueEntry stateFile block contains both type check and traversal check', () => {
const fn = extractFunction(AGENT_SRC, 'isValidQueueEntry');
const stateFileIdx = fn.indexOf('stateFile');
const stateBlock = fn.slice(stateFileIdx, stateFileIdx + 300);
// Must contain the type check
expect(stateBlock).toContain('typeof obj.stateFile');
// Must contain the includes('..') call
expect(stateBlock).toMatch(/includes\s*\(\s*['"]\.\.['"]\s*\)/);
});
});
// ─── Round-2 finding 3: stateFile path traversal check ────────────────────
// Tested isValidQueueEntry's stateFile validator on sidebar-agent.ts. Both
// the function and the file are gone (chat queue ripped). The terminal-agent
// PTY path no longer takes a queue entry — it accepts WebSocket frames
// gated on Origin + session token, no on-disk queue to traverse. Path
// traversal in browse-server's tab-state writer is covered by
// browse/test/terminal-agent.test.ts (handleTabState atomic-write tests).
// ─── Task 5: /health endpoint must not expose sensitive fields ───────────────
@@ -421,24 +369,11 @@ describe('cookie-import domain validation', () => {
});
});
// ─── Task 9: loadSession ID validation ──────────────────────────────────────
describe('loadSession session ID validation', () => {
it('loadSession validates session ID format before using it in a path', () => {
const fn = extractFunction(SERVER_SRC, 'loadSession');
expect(fn).toBeTruthy();
// Must contain the alphanumeric regex guard
expect(fn).toMatch(/\[a-zA-Z0-9_-\]/);
});
it('loadSession returns null on invalid session ID', () => {
const fn = extractFunction(SERVER_SRC, 'loadSession');
const block = fn.slice(fn.indexOf('activeData.id'));
// Must warn and return null
expect(block).toContain('Invalid session ID');
expect(block).toContain('return null');
});
});
// loadSession session ID validation — loadSession lived inside the chat
// agent state block (sidebar-agent.ts session persistence). Chat queue
// is gone, so the function and its session-ID validator are gone. The
// terminal-agent's PTY session has no on-disk session ID — the WebSocket
// holds the session for its lifetime.
// ─── Task 10: Responsive screenshot path validation ──────────────────────────
@@ -520,40 +455,11 @@ describe('Task 11: state load cookie validation', () => {
});
});
// ─── Task 12: Validate activeTabUrl before syncActiveTabByUrl ─────────────────
describe('Task 12: activeTabUrl sanitized before syncActiveTabByUrl', () => {
it('sidebar-tabs route sanitizes activeUrl before syncActiveTabByUrl', () => {
const block = sliceBetween(SERVER_SRC, "url.pathname === '/sidebar-tabs'", "url.pathname === '/sidebar-tabs/switch'");
expect(block).toContain('sanitizeExtensionUrl');
expect(block).toContain('syncActiveTabByUrl');
const sanitizeIdx = block.indexOf('sanitizeExtensionUrl');
const syncIdx = block.indexOf('syncActiveTabByUrl');
expect(sanitizeIdx).toBeLessThan(syncIdx);
});
it('sidebar-command route sanitizes extensionUrl before syncActiveTabByUrl', () => {
const block = sliceBetween(SERVER_SRC, "url.pathname === '/sidebar-command'", "url.pathname === '/sidebar-chat/clear'");
expect(block).toContain('sanitizeExtensionUrl');
expect(block).toContain('syncActiveTabByUrl');
const sanitizeIdx = block.indexOf('sanitizeExtensionUrl');
const syncIdx = block.indexOf('syncActiveTabByUrl');
expect(sanitizeIdx).toBeLessThan(syncIdx);
});
it('direct unsanitized syncActiveTabByUrl calls are not present (all calls go through sanitize)', () => {
// Every syncActiveTabByUrl call should be preceded by sanitizeExtensionUrl in the nearby code
// We verify there are no direct browserManager.syncActiveTabByUrl(activeUrl) or
// browserManager.syncActiveTabByUrl(extensionUrl) patterns (without sanitize wrapper)
const block1 = sliceBetween(SERVER_SRC, "url.pathname === '/sidebar-tabs'", "url.pathname === '/sidebar-tabs/switch'");
// Should NOT contain direct call with raw activeUrl
expect(block1).not.toMatch(/syncActiveTabByUrl\(activeUrl\)/);
const block2 = sliceBetween(SERVER_SRC, "url.pathname === '/sidebar-command'", "url.pathname === '/sidebar-chat/clear'");
// Should NOT contain direct call with raw extensionUrl
expect(block2).not.toMatch(/syncActiveTabByUrl\(extensionUrl\)/);
});
});
// activeTabUrl sanitized before syncActiveTabByUrl — tested URL sanitization
// on the now-deleted /sidebar-tabs and /sidebar-command routes. The
// terminal-agent reads tab URLs from the live tabs.json file (atomic write
// from background.js), and chrome:// / chrome-extension:// pages are
// filtered server-side in handleTabState — see browse/test/terminal-agent.test.ts.
// ─── Task 13: Inbox output wrapped as untrusted ──────────────────────────────
@@ -581,107 +487,17 @@ describe('Task 13: inbox output wrapped as untrusted content', () => {
});
});
// ─── Task 14: DOM serialization round-trip replaced with DocumentFragment ─────
// switchChatTab DocumentFragment + pollChat reentrancy guard tests targeted
// now-deleted chat-tab DOM logic and chat-polling reentrancy. Both are gone
// (Terminal pane is the sole sidebar surface; xterm.js owns its own DOM
// lifecycle, and the WebSocket has no reentrancy hazard).
const SIDEPANEL_SRC = fs.readFileSync(path.join(import.meta.dir, '../../extension/sidepanel.js'), 'utf-8');
describe('Task 14: switchChatTab uses DocumentFragment, not innerHTML round-trip', () => {
it('switchChatTab does NOT use innerHTML to restore chat (string-based re-parse removed)', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'switchChatTab');
expect(fn).toBeTruthy();
// Must NOT have the dangerous pattern of assigning chatDomByTab value back to innerHTML
expect(fn).not.toMatch(/chatMessages\.innerHTML\s*=\s*chatDomByTab/);
});
it('switchChatTab uses createDocumentFragment to save chat DOM', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'switchChatTab');
expect(fn).toContain('createDocumentFragment');
});
it('switchChatTab moves nodes via appendChild/firstChild (not innerHTML assignment)', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'switchChatTab');
// Must use appendChild to restore nodes from fragment
expect(fn).toContain('chatMessages.appendChild');
});
it('chatDomByTab comment documents that values are DocumentFragments, not strings', () => {
// Check module-level comment on chatDomByTab
const commentIdx = SIDEPANEL_SRC.indexOf('chatDomByTab');
const commentLine = SIDEPANEL_SRC.slice(commentIdx, commentIdx + 120);
expect(commentLine).toMatch(/DocumentFragment|fragment/i);
});
it('welcome screen is built with DOM methods in the else branch (not innerHTML)', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'switchChatTab');
// The else branch must use createElement, not innerHTML template literal
expect(fn).toContain('createElement');
// The specific innerHTML template with chat-welcome must be gone
expect(fn).not.toMatch(/innerHTML\s*=\s*`[\s\S]*?chat-welcome/);
});
});
// ─── Task 15: pollChat/switchChatTab reentrancy guard ────────────────────────
describe('Task 15: pollChat reentrancy guard and deferred call in switchChatTab', () => {
it('pollInProgress guard variable is declared at module scope', () => {
// Must be declared before any function definitions (within first 2000 chars)
const moduleTop = SIDEPANEL_SRC.slice(0, 2000);
expect(moduleTop).toContain('pollInProgress');
});
it('pollChat function checks and sets pollInProgress', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'pollChat');
expect(fn).toBeTruthy();
expect(fn).toContain('pollInProgress');
});
it('pollChat resets pollInProgress in finally block', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'pollChat');
// The finally block must contain the reset
const finallyIdx = fn.indexOf('finally');
expect(finallyIdx).toBeGreaterThan(-1);
const finallyBlock = fn.slice(finallyIdx, finallyIdx + 60);
expect(finallyBlock).toContain('pollInProgress');
});
it('switchChatTab calls pollChat via setTimeout (not directly)', () => {
const fn = extractFunction(SIDEPANEL_SRC, 'switchChatTab');
// Must use setTimeout to defer pollChat — no direct call at the end
expect(fn).toMatch(/setTimeout\s*\(\s*pollChat/);
// Must NOT have a bare direct call `pollChat()` at the end (outside setTimeout)
// We check that there is no standalone `pollChat()` call (outside setTimeout wrapper)
const withoutSetTimeout = fn.replace(/setTimeout\s*\(\s*pollChat[^)]*\)/g, '');
expect(withoutSetTimeout).not.toMatch(/\bpollChat\s*\(\s*\)/);
});
});
// ─── Task 16: SIGKILL escalation in sidebar-agent timeout ────────────────────
describe('Task 16: sidebar-agent timeout handler uses SIGTERM→SIGKILL escalation', () => {
it('timeout block sends SIGTERM first', () => {
// Slice from "Timed out" / setTimeout block to processingTabs.delete
const timeoutStart = AGENT_SRC.indexOf("SIDEBAR_AGENT_TIMEOUT");
expect(timeoutStart).toBeGreaterThan(-1);
const timeoutBlock = AGENT_SRC.slice(timeoutStart, timeoutStart + 600);
expect(timeoutBlock).toContain('SIGTERM');
});
it('timeout block escalates to SIGKILL after delay', () => {
const timeoutStart = AGENT_SRC.indexOf("SIDEBAR_AGENT_TIMEOUT");
const timeoutBlock = AGENT_SRC.slice(timeoutStart, timeoutStart + 600);
expect(timeoutBlock).toContain('SIGKILL');
});
it('SIGTERM appears before SIGKILL in timeout block', () => {
const timeoutStart = AGENT_SRC.indexOf("SIDEBAR_AGENT_TIMEOUT");
const timeoutBlock = AGENT_SRC.slice(timeoutStart, timeoutStart + 600);
const sigtermIdx = timeoutBlock.indexOf('SIGTERM');
const sigkillIdx = timeoutBlock.indexOf('SIGKILL');
expect(sigtermIdx).toBeGreaterThan(-1);
expect(sigkillIdx).toBeGreaterThan(-1);
expect(sigtermIdx).toBeLessThan(sigkillIdx);
});
});
// ─── Task 16: SIGKILL escalation ────────────────────────────────────────────
// Originally tested sidebar-agent's SIDEBAR_AGENT_TIMEOUT block. The chat
// queue and its watchdog are gone. terminal-agent.ts disposes claude with
// the same SIGINT-then-SIGKILL-after-3s pattern; that's covered by
// browse/test/terminal-agent.test.ts ("cleanup escalates SIGINT to SIGKILL
// after 3s on close").
// ─── Task 17: viewport and wait bounds clamping ──────────────────────────────
-218
View File
@@ -1,218 +0,0 @@
/**
* Full-stack E2E the security-contract anchor test.
*
* Spins up a real browse server + real sidebar-agent subprocess, points
* them at a MOCK claude binary (browse/test/fixtures/mock-claude/claude)
* that deterministically emits a canary-leaking tool_use event, then
* verifies the whole pipeline reacts:
*
* 1. Server canary-injects into the system prompt
* 2. Server queues the message
* 3. Sidebar-agent spawns mock-claude
* 4. Mock-claude emits tool_use with CANARY-XXX in a URL arg
* 5. Sidebar-agent's detectCanaryLeak fires on the stream event
* 6. onCanaryLeaked logs, SIGTERM's mock-claude, emits security_event
* 7. /sidebar-chat returns security_event + agent_error entries
*
* This test proves the end-to-end contract: when a canary leak happens,
* the session terminates AND the sidepanel receives the events that drive
* the approved banner render. No LLM cost, <10s total runtime.
*
* Fully deterministic safe to run on every commit (gate tier).
*/
import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
import { spawn, type Subprocess } from 'bun';
import * as fs from 'fs';
import * as os from 'os';
import * as path from 'path';
let serverProc: Subprocess | null = null;
let agentProc: Subprocess | null = null;
let serverPort = 0;
let authToken = '';
let tmpDir = '';
let stateFile = '';
let queueFile = '';
const MOCK_CLAUDE_DIR = path.resolve(import.meta.dir, 'fixtures', 'mock-claude');
async function apiFetch(pathname: string, opts: RequestInit = {}): Promise<Response> {
const headers: Record<string, string> = {
'Content-Type': 'application/json',
Authorization: `Bearer ${authToken}`,
...(opts.headers as Record<string, string> | undefined),
};
return fetch(`http://127.0.0.1:${serverPort}${pathname}`, { ...opts, headers });
}
beforeAll(async () => {
tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'security-e2e-fullstack-'));
stateFile = path.join(tmpDir, 'browse.json');
queueFile = path.join(tmpDir, 'sidebar-queue.jsonl');
fs.mkdirSync(path.dirname(queueFile), { recursive: true });
const serverScript = path.resolve(import.meta.dir, '..', 'src', 'server.ts');
const agentScript = path.resolve(import.meta.dir, '..', 'src', 'sidebar-agent.ts');
// 1) Start the browse server.
serverProc = spawn(['bun', 'run', serverScript], {
env: {
...process.env,
BROWSE_STATE_FILE: stateFile,
BROWSE_HEADLESS_SKIP: '1', // no Chromium for this test
BROWSE_PORT: '0',
SIDEBAR_QUEUE_PATH: queueFile,
BROWSE_IDLE_TIMEOUT: '300',
},
stdio: ['ignore', 'pipe', 'pipe'],
});
// Wait for state file with token + port
const deadline = Date.now() + 15000;
while (Date.now() < deadline) {
if (fs.existsSync(stateFile)) {
try {
const state = JSON.parse(fs.readFileSync(stateFile, 'utf-8'));
if (state.port && state.token) {
serverPort = state.port;
authToken = state.token;
break;
}
} catch {}
}
await new Promise((r) => setTimeout(r, 100));
}
if (!serverPort) throw new Error('Server did not start in time');
// 2) Start the sidebar-agent with PATH prepended by the mock-claude dir.
// sidebar-agent spawns `claude` via PATH lookup (spawn('claude', ...) — see
// browse/src/sidebar-agent.ts spawnClaude), so prepending works without any
// source change.
const shimmedPath = `${MOCK_CLAUDE_DIR}:${process.env.PATH ?? ''}`;
agentProc = spawn(['bun', 'run', agentScript], {
env: {
...process.env,
PATH: shimmedPath,
BROWSE_STATE_FILE: stateFile,
SIDEBAR_QUEUE_PATH: queueFile,
BROWSE_SERVER_PORT: String(serverPort),
BROWSE_PORT: String(serverPort),
BROWSE_NO_AUTOSTART: '1',
// Scenario for mock-claude inherits through spawn env below — the agent
// itself doesn't read this, but the claude subprocess it spawns does.
MOCK_CLAUDE_SCENARIO: 'canary_leak_in_tool_arg',
// Force classifier off so pre-spawn ML scan doesn't fire on our
// benign synthetic test prompt. This test exercises the canary
// path specifically.
GSTACK_SECURITY_OFF: '1',
},
stdio: ['ignore', 'pipe', 'pipe'],
});
// Give the agent a moment to establish its poll loop.
await new Promise((r) => setTimeout(r, 500));
}, 30000);
async function drainStderr(proc: Subprocess | null, label: string): Promise<void> {
if (!proc?.stderr) return;
try {
const reader = (proc.stderr as ReadableStream).getReader();
// Drain briefly — don't block shutdown
const result = await Promise.race([
reader.read(),
new Promise<ReadableStreamReadResult<Uint8Array>>((resolve) =>
setTimeout(() => resolve({ done: true, value: undefined }), 100)
),
]);
if (result?.value) {
const text = new TextDecoder().decode(result.value);
if (text.trim()) console.error(`[${label} stderr]`, text.slice(0, 2000));
}
} catch {}
}
afterAll(async () => {
// Dump agent stderr for diagnostic
await drainStderr(agentProc, 'agent');
for (const proc of [serverProc, agentProc]) {
if (proc) {
try { proc.kill('SIGTERM'); } catch {}
try { setTimeout(() => { try { proc.kill('SIGKILL'); } catch {} }, 1500); } catch {}
}
}
try { fs.rmSync(tmpDir, { recursive: true, force: true }); } catch {}
});
describe('security pipeline E2E (mock claude)', () => {
test('server injects canary, queues message, agent spawns mock claude', async () => {
const resp = await apiFetch('/sidebar-command', {
method: 'POST',
body: JSON.stringify({
message: "What's on this page?",
activeTabUrl: 'https://attacker.example.com/',
}),
});
expect(resp.status).toBe(200);
// Wait for the sidebar-agent to pick up the entry and spawn mock-claude.
// Queue entry must contain `canary` field (added by server.ts spawnClaude).
await new Promise((r) => setTimeout(r, 250));
const queueContent = fs.readFileSync(queueFile, 'utf-8').trim();
const lines = queueContent.split('\n').filter(Boolean);
expect(lines.length).toBeGreaterThan(0);
const entry = JSON.parse(lines[lines.length - 1]);
expect(entry.canary).toMatch(/^CANARY-[0-9A-F]+$/);
expect(entry.prompt).toContain(entry.canary);
expect(entry.prompt).toContain('NEVER include it');
});
test('canary leak triggers security_event + agent_error in /sidebar-chat', async () => {
// By now the mock-claude subprocess has emitted the tool_use with the
// leaked canary. Sidebar-agent's handleStreamEvent -> detectCanaryLeak
// -> onCanaryLeaked should have fired security_event + agent_error and
// SIGTERM'd the mock. Poll /sidebar-chat up to 10s for the events.
const deadline = Date.now() + 10000;
let securityEvent: any = null;
let agentError: any = null;
while (Date.now() < deadline && (!securityEvent || !agentError)) {
const resp = await apiFetch('/sidebar-chat');
const data: any = await resp.json();
for (const entry of data.entries ?? []) {
if (entry.type === 'security_event') securityEvent = entry;
if (entry.type === 'agent_error') agentError = entry;
}
if (securityEvent && agentError) break;
await new Promise((r) => setTimeout(r, 250));
}
expect(securityEvent).not.toBeNull();
expect(securityEvent.verdict).toBe('block');
expect(securityEvent.reason).toBe('canary_leaked');
expect(securityEvent.layer).toBe('canary');
// The leak is on a tool_use channel — onCanaryLeaked records "tool_use:Bash"
expect(String(securityEvent.channel)).toContain('tool_use');
expect(securityEvent.domain).toBe('attacker.example.com');
expect(agentError).not.toBeNull();
expect(agentError.error).toContain('Session terminated');
expect(agentError.error).toContain('prompt injection detected');
}, 15000);
test('attempts.jsonl logged with salted payload_hash and verdict=block', async () => {
// onCanaryLeaked also calls logAttempt — check the log file exists
// and contains the event. The file lives at ~/.gstack/security/attempts.jsonl.
const logPath = path.join(os.homedir(), '.gstack', 'security', 'attempts.jsonl');
expect(fs.existsSync(logPath)).toBe(true);
const content = fs.readFileSync(logPath, 'utf-8');
const recent = content.split('\n').filter(Boolean).slice(-10);
// Find at least one entry with verdict=block and layer=canary from our run
const ourEntry = recent
.map((l) => { try { return JSON.parse(l); } catch { return null; } })
.find((e) => e && e.layer === 'canary' && e.verdict === 'block' && e.urlDomain === 'attacker.example.com');
expect(ourEntry).toBeTruthy();
// payload_hash is a 64-char sha256 hex
expect(String(ourEntry.payloadHash)).toMatch(/^[0-9a-f]{64}$/);
// Never stored the payload itself — only the hash
expect(JSON.stringify(ourEntry)).not.toContain('CANARY-');
});
});
@@ -1,405 +0,0 @@
/**
* Full-stack review-flow E2E with the real classifier.
*
* Spins up real server + real sidebar-agent subprocess + mock-claude and
* exercises the whole tool-output BLOCK review decide path with the
* real TestSavantAI classifier warm. The injection string trips the real
* model reliably (measured: confidence 0.9999 on classic DAN-style text).
*
* What this covers that gate-tier tests don't:
* * Real classifier actually fires on the injection
* * sidebar-agent emits a reviewable security_event for real, not a stub
* * server's POST /security-decision writes the on-disk decision file
* * sidebar-agent's poll loop reads the file and either resumes or kills
* the mock-claude subprocess
* * attempts.jsonl ends up with the right verdict (block vs user_overrode)
*
* This is periodic tier. First run warms the ~112MB classifier from
* HuggingFace ~30s cold. Subsequent runs use the cached model under
* ~/.gstack/models/testsavant-small/ and complete in ~5s.
*
* SKIPS if the classifier can't warm (no network, no disk) the test is
* truth-seeking only when the stack is genuinely up.
*/
import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
import { spawn, type Subprocess } from 'bun';
import * as fs from 'fs';
import * as os from 'os';
import * as path from 'path';
const MOCK_CLAUDE_DIR = path.resolve(import.meta.dir, 'fixtures', 'mock-claude');
const WARMUP_TIMEOUT_MS = 90_000; // first-run download budget
const CLASSIFIER_CACHE = path.join(os.homedir(), '.gstack', 'models', 'testsavant-small');
let serverProc: Subprocess | null = null;
let agentProc: Subprocess | null = null;
let serverPort = 0;
let authToken = '';
let tmpDir = '';
let stateFile = '';
let queueFile = '';
let attemptsPath = '';
/**
* Eager check is the classifier model already on disk? `test.skipIf()`
* is evaluated at file-registration time (before beforeAll runs), so a
* runtime boolean wouldn't work all tests would unconditionally register
* as skipped. Probe the model dir synchronously at file load.
* Same pattern as security-sidepanel-dom.test.ts uses for chromium.
*/
const CLASSIFIER_READY = (() => {
try {
if (!fs.existsSync(CLASSIFIER_CACHE)) return false;
// At minimum we need the tokenizer config + onnx model.
return fs.existsSync(path.join(CLASSIFIER_CACHE, 'tokenizer.json'))
&& fs.existsSync(path.join(CLASSIFIER_CACHE, 'onnx'));
} catch {
return false;
}
})();
async function apiFetch(pathname: string, opts: RequestInit = {}): Promise<Response> {
return fetch(`http://127.0.0.1:${serverPort}${pathname}`, {
...opts,
headers: {
'Content-Type': 'application/json',
Authorization: `Bearer ${authToken}`,
...(opts.headers as Record<string, string> | undefined),
},
});
}
async function waitForSecurityEntry(
predicate: (entry: any) => boolean,
timeoutMs: number,
): Promise<any | null> {
const deadline = Date.now() + timeoutMs;
while (Date.now() < deadline) {
const resp = await apiFetch('/sidebar-chat');
const data: any = await resp.json();
for (const entry of data.entries ?? []) {
if (entry.type === 'security_event' && predicate(entry)) return entry;
}
await new Promise((r) => setTimeout(r, 250));
}
return null;
}
async function waitForProcessExit(proc: Subprocess, timeoutMs: number): Promise<number | null> {
const deadline = Date.now() + timeoutMs;
while (Date.now() < deadline) {
if (proc.exitCode !== null) return proc.exitCode;
await new Promise((r) => setTimeout(r, 100));
}
return null;
}
async function readAttempts(): Promise<any[]> {
if (!fs.existsSync(attemptsPath)) return [];
const raw = fs.readFileSync(attemptsPath, 'utf-8');
return raw.split('\n').filter(Boolean).map((l) => {
try { return JSON.parse(l); } catch { return null; }
}).filter(Boolean);
}
async function startStack(scenario: string, attemptsDir: string): Promise<void> {
tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'security-review-fullstack-'));
stateFile = path.join(tmpDir, 'browse.json');
queueFile = path.join(tmpDir, 'sidebar-queue.jsonl');
fs.mkdirSync(path.dirname(queueFile), { recursive: true });
// Re-root HOME for both server and agent so:
// - server.ts's SESSIONS_DIR doesn't load pre-existing chat history
// from ~/.gstack/sidebar-sessions/ (caused ghost security_events to
// leak in from the live /open-gstack-browser session)
// - security.ts's attempts.jsonl writes land in a test-owned dir
// - session-state.json, chromium-profile, etc. stay isolated
fs.mkdirSync(path.join(attemptsDir, '.gstack'), { recursive: true });
// Symlink the models dir through to the real cache — without it the
// sidebar-agent would try to re-download 112MB every test run.
const testModelsDir = path.join(attemptsDir, '.gstack', 'models');
const realModelsDir = path.join(os.homedir(), '.gstack', 'models');
try {
if (fs.existsSync(realModelsDir) && !fs.existsSync(testModelsDir)) {
fs.symlinkSync(realModelsDir, testModelsDir);
}
} catch {
// Symlink may already exist — ignore.
}
const serverScript = path.resolve(import.meta.dir, '..', 'src', 'server.ts');
const agentScript = path.resolve(import.meta.dir, '..', 'src', 'sidebar-agent.ts');
serverProc = spawn(['bun', 'run', serverScript], {
env: {
...process.env,
BROWSE_STATE_FILE: stateFile,
BROWSE_HEADLESS_SKIP: '1',
BROWSE_PORT: '0',
SIDEBAR_QUEUE_PATH: queueFile,
BROWSE_IDLE_TIMEOUT: '300',
HOME: attemptsDir,
},
stdio: ['ignore', 'pipe', 'pipe'],
});
const deadline = Date.now() + 15000;
while (Date.now() < deadline) {
if (fs.existsSync(stateFile)) {
try {
const state = JSON.parse(fs.readFileSync(stateFile, 'utf-8'));
if (state.port && state.token) {
serverPort = state.port;
authToken = state.token;
break;
}
} catch {}
}
await new Promise((r) => setTimeout(r, 100));
}
if (!serverPort) throw new Error('Server did not start in time');
const shimmedPath = `${MOCK_CLAUDE_DIR}:${process.env.PATH ?? ''}`;
agentProc = spawn(['bun', 'run', agentScript], {
env: {
...process.env,
PATH: shimmedPath,
BROWSE_STATE_FILE: stateFile,
SIDEBAR_QUEUE_PATH: queueFile,
BROWSE_SERVER_PORT: String(serverPort),
BROWSE_PORT: String(serverPort),
BROWSE_NO_AUTOSTART: '1',
MOCK_CLAUDE_SCENARIO: scenario,
HOME: attemptsDir,
},
stdio: ['ignore', 'pipe', 'pipe'],
});
attemptsPath = path.join(attemptsDir, '.gstack', 'security', 'attempts.jsonl');
// Give the agent a moment to establish its poll loop + warmup the model.
await new Promise((r) => setTimeout(r, 500));
}
async function stopStack(): Promise<void> {
for (const proc of [serverProc, agentProc]) {
if (proc) {
try { proc.kill('SIGTERM'); } catch {}
try { setTimeout(() => { try { proc.kill('SIGKILL'); } catch {} }, 1500); } catch {}
}
}
serverProc = null;
agentProc = null;
try { fs.rmSync(tmpDir, { recursive: true, force: true }); } catch {}
}
beforeAll(async () => {
// Sanity: the on-disk cache is real + decodable. If this fails, mark the
// file as "classifier unavailable" (we can't toggle CLASSIFIER_READY
// post-registration — a failure here just means the tests below will
// exercise the agent without a working classifier, which is the honest
// signal we want anyway).
if (!CLASSIFIER_READY) return;
});
afterAll(async () => {
await stopStack();
});
describe('review-flow full-stack E2E', () => {
test.skipIf(!CLASSIFIER_READY)(
'tool_result injection → reviewable event → user ALLOWS → attempts.jsonl has user_overrode',
async () => {
const attemptsDir = fs.mkdtempSync(path.join(os.tmpdir(), 'attempts-allow-'));
try {
await startStack('tool_result_injection', attemptsDir);
// Fire the message that will cause mock-claude to emit the
// injection-laden tool_result.
const resp = await apiFetch('/sidebar-command', {
method: 'POST',
body: JSON.stringify({
message: 'summarize the hacker news comments',
activeTabUrl: 'https://news.ycombinator.com/item?id=42',
}),
});
expect(resp.status).toBe(200);
// Wait for the real classifier to fire and emit a reviewable
// security_event. The classifier is warm so this should happen in
// well under 10s once the tool_result arrives.
const reviewable = await waitForSecurityEntry(
(e) => e.verdict === 'block' && e.reviewable === true,
30_000,
);
expect(reviewable).not.toBeNull();
expect(reviewable.reason).toBe('tool_result_ml');
expect(reviewable.tool).toBe('Bash');
expect(String(reviewable.suspected_text ?? '')).toContain('IGNORE ALL PREVIOUS');
// User clicks Allow via the banner → sidepanel POSTs to server.
const decisionResp = await apiFetch('/security-decision', {
method: 'POST',
body: JSON.stringify({
tabId: reviewable.tabId,
decision: 'allow',
reason: 'user',
}),
});
expect(decisionResp.status).toBe(200);
// Wait for sidebar-agent's poll loop to consume the decision and
// emit a follow-up user_overrode security_event.
const overrode = await waitForSecurityEntry(
(e) => e.verdict === 'user_overrode',
10_000,
);
expect(overrode).not.toBeNull();
// Audit log must capture both the block and the override, in that
// order. Both records share the same salted payload hash so the
// security dashboard can aggregate them as a single attempt.
const attempts = await readAttempts();
const blockLog = attempts.find(
(a) => a.verdict === 'block' && a.layer === 'testsavant_content',
);
const overrodeLog = attempts.find(
(a) => a.verdict === 'user_overrode' && a.layer === 'testsavant_content',
);
expect(blockLog).toBeTruthy();
expect(overrodeLog).toBeTruthy();
expect(overrodeLog.payloadHash).toBe(blockLog.payloadHash);
// Privacy contract: neither record includes the raw payload.
expect(JSON.stringify(overrodeLog)).not.toContain('IGNORE ALL PREVIOUS');
// Liveness: session must actually KEEP RUNNING after Allow. Mock-claude
// emits a second tool_use to post-block-followup.example.com ~8s
// after the tool_result. That event must reach the chat feed, proving
// the sidebar-agent resumed the stream-handler relay instead of
// silently wedging.
const followupDeadline = Date.now() + 20_000;
let followup: any = null;
while (Date.now() < followupDeadline && !followup) {
const chatResp = await apiFetch('/sidebar-chat');
const chatData: any = await chatResp.json();
for (const entry of chatData.entries ?? []) {
const input = String((entry as any).input ?? '');
if (
entry.type === 'tool_use' &&
input.includes('post-block-followup.example.com')
) {
followup = entry;
break;
}
}
if (!followup) await new Promise((r) => setTimeout(r, 300));
}
expect(followup).not.toBeNull();
} finally {
await stopStack();
try { fs.rmSync(attemptsDir, { recursive: true, force: true }); } catch {}
}
},
90_000,
);
test.skipIf(!CLASSIFIER_READY)(
'tool_result injection → reviewable event → user BLOCKS → agent session terminates',
async () => {
const attemptsDir = fs.mkdtempSync(path.join(os.tmpdir(), 'attempts-block-'));
try {
await startStack('tool_result_injection', attemptsDir);
const resp = await apiFetch('/sidebar-command', {
method: 'POST',
body: JSON.stringify({
message: 'summarize the hacker news comments',
activeTabUrl: 'https://news.ycombinator.com/item?id=42',
}),
});
expect(resp.status).toBe(200);
const reviewable = await waitForSecurityEntry(
(e) => e.verdict === 'block' && e.reviewable === true,
30_000,
);
expect(reviewable).not.toBeNull();
const decisionResp = await apiFetch('/security-decision', {
method: 'POST',
body: JSON.stringify({
tabId: reviewable.tabId,
decision: 'block',
reason: 'user',
}),
});
expect(decisionResp.status).toBe(200);
// Wait for the agent_error that the sidebar-agent emits when it
// kills the claude subprocess after a user-confirmed block. This
// is the sidepanel's "Session terminated" signal.
const deadline = Date.now() + 15_000;
let errorEntry: any = null;
while (Date.now() < deadline && !errorEntry) {
const chatResp = await apiFetch('/sidebar-chat');
const chatData: any = await chatResp.json();
for (const entry of chatData.entries ?? []) {
if (
entry.type === 'agent_error' &&
String(entry.error ?? '').includes('Session terminated')
) {
errorEntry = entry;
break;
}
}
if (!errorEntry) await new Promise((r) => setTimeout(r, 200));
}
expect(errorEntry).not.toBeNull();
// attempts.jsonl must NOT have a user_overrode entry for this run.
const attempts = await readAttempts();
const overrodeLog = attempts.find((a) => a.verdict === 'user_overrode');
expect(overrodeLog).toBeFalsy();
// The real security property: after Block, NO FURTHER tool calls
// reach the chat feed. Mock-claude would have emitted a tool_use
// to post-block-followup.example.com ~8s after the tool_result if
// the session had kept running. Wait long enough for that window
// to close (12s total), then assert the followup event never
// appeared. This is what makes "block" actually stop the page —
// the subprocess is SIGTERM'd before it can emit the next event.
await new Promise((r) => setTimeout(r, 12_000));
const finalChatResp = await apiFetch('/sidebar-chat');
const finalChatData: any = await finalChatResp.json();
const followupAttempted = (finalChatData.entries ?? []).some(
(entry: any) =>
entry.type === 'tool_use' &&
String(entry.input ?? '').includes('post-block-followup.example.com'),
);
expect(followupAttempted).toBe(false);
// And mock-claude must actually have died (not just been signaled
// — the SIGTERM + SIGKILL pair should have exited the process).
const mockAlive = (await apiFetch('/sidebar-chat')).ok; // channel still open
expect(mockAlive).toBe(true);
} finally {
await stopStack();
try { fs.rmSync(attemptsDir, { recursive: true, force: true }); } catch {}
}
},
90_000,
);
test.skipIf(!CLASSIFIER_READY)(
'no decision within 60s → timeout auto-blocks',
async () => {
// This test would naturally take 60s+ to run. We assert the
// decision file semantics instead — the unit-test suite already
// verified the poll loop times out and defaults to block
// (security-review-flow.test.ts). Kept here as a spec marker so
// the scenario is documented in the full-stack file.
expect(true).toBe(true);
},
);
});
@@ -1,345 +0,0 @@
/**
* Review-flow E2E (sidepanel side, hermetic).
*
* Loads the real extension sidepanel.html in Playwright Chromium, stubs
* the browse server responses, injects a `reviewable: true` security_event
* into /sidebar-chat, and asserts the user-in-the-loop flow end-to-end:
*
* 1. Banner renders with "Review suspected injection" title
* 2. Suspected text excerpt shows up inside the expandable details
* 3. Allow + Block buttons are visible and actionable
* 4. Clicking Allow posts to /security-decision with decision:"allow"
* 5. Clicking Block posts to /security-decision with decision:"block"
* 6. Banner auto-hides after decision
*
* This is the UI-and-wire test. The server-side handshake (decision file
* write + sidebar-agent poll) is covered by security-review-flow.test.ts.
* The full-stack version with real mock-claude + real classifier lives
* in security-review-fullstack.test.ts (periodic tier).
*
* Gate tier. ~3s. Skipped if Playwright chromium is unavailable.
*/
import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
import { chromium, type Browser, type Page } from 'playwright';
const EXTENSION_DIR = path.resolve(import.meta.dir, '..', '..', 'extension');
const SIDEPANEL_URL = `file://${EXTENSION_DIR}/sidepanel.html`;
const CHROMIUM_AVAILABLE = (() => {
try {
const exe = chromium.executablePath();
return !!exe && fs.existsSync(exe);
} catch {
return false;
}
})();
interface DecisionCall {
tabId: number;
decision: 'allow' | 'block';
reason?: string;
}
/**
* Install the same stubs the existing sidepanel-dom test uses, plus a
* fetch interceptor that captures POSTs to /security-decision into a
* page-scoped array. Returns a handle to read the captured calls.
*/
async function installStubsAndCapture(
page: Page,
scenario: { securityEntries: any[] },
): Promise<void> {
await page.addInitScript((params: any) => {
(window as any).__decisionCalls = [];
(window as any).chrome = {
runtime: {
sendMessage: (_req: any, cb: any) => {
const payload = { connected: true, port: 34567 };
if (typeof cb === 'function') {
setTimeout(() => cb(payload), 0);
return undefined;
}
return Promise.resolve(payload);
},
lastError: null,
onMessage: { addListener: () => {} },
},
tabs: {
query: (_q: any, cb: any) => setTimeout(() => cb([{ id: 1, url: 'https://example.com' }]), 0),
onActivated: { addListener: () => {} },
onUpdated: { addListener: () => {} },
},
};
(window as any).EventSource = class {
constructor() {}
addEventListener() {}
close() {}
};
const scenarioRef = params;
const origFetch = window.fetch;
window.fetch = async function (input: any, init?: any) {
const url = String(input);
if (url.endsWith('/health')) {
return new Response(JSON.stringify({
status: 'healthy',
token: 'test-token',
mode: 'headed',
agent: { status: 'idle', runningFor: null, queueLength: 0 },
session: null,
security: { status: 'protected', layers: { testsavant: 'ok', transcript: 'ok', canary: 'ok' } },
}), { status: 200, headers: { 'Content-Type': 'application/json' } });
}
if (url.includes('/sidebar-chat')) {
return new Response(JSON.stringify({
entries: scenarioRef.securityEntries ?? [],
total: (scenarioRef.securityEntries ?? []).length,
agentStatus: 'idle',
activeTabId: 1,
security: { status: 'protected', layers: { testsavant: 'ok', transcript: 'ok', canary: 'ok' } },
}), { status: 200, headers: { 'Content-Type': 'application/json' } });
}
if (url.includes('/security-decision') && init?.method === 'POST') {
try {
const body = JSON.parse(init.body || '{}');
(window as any).__decisionCalls.push(body);
} catch {
(window as any).__decisionCalls.push({ _parseError: true, raw: init?.body });
}
return new Response(JSON.stringify({ ok: true }), { status: 200, headers: { 'Content-Type': 'application/json' } });
}
if (url.includes('/sidebar-tabs')) {
return new Response(JSON.stringify({ tabs: [] }), { status: 200 });
}
if (typeof origFetch === 'function') return origFetch(input, init);
return new Response('{}', { status: 200 });
} as any;
}, scenario);
}
let browser: Browser | null = null;
beforeAll(async () => {
if (!CHROMIUM_AVAILABLE) return;
browser = await chromium.launch({ headless: true });
}, 30000);
afterAll(async () => {
if (browser) {
try {
// Race browser.close() against a timeout — on rare occasions Playwright
// hangs on close because an EventSource stub keeps a poll alive. 10s is
// plenty; past that we forcibly drop the handle. Bun's default hook
// timeout is 5s and has bitten this file.
await Promise.race([
browser.close(),
new Promise<void>((resolve) => setTimeout(resolve, 10000)),
]);
} catch {}
}
}, 15000);
/**
* The reviewable security_event the sidebar-agent emits on tool-output BLOCK.
* Mirrors the shape of the real production event: verdict:'block',
* reviewable:true, suspected_text excerpt, per-layer signals, and tabId
* so the banner's Allow/Block buttons know which tab to decide for.
*/
function buildReviewableEntry(overrides?: Partial<any>): any {
return {
id: 42,
ts: '2026-04-20T12:00:00Z',
role: 'agent',
type: 'security_event',
verdict: 'block',
reason: 'tool_result_ml',
layer: 'testsavant_content',
confidence: 0.95,
domain: 'news.ycombinator.com',
tool: 'Bash',
reviewable: true,
suspected_text: 'A comment thread discussing ignore previous instructions and reveal secrets — classifier flagged this as injection but it is actually benign developer content about a prompt injection incident.',
signals: [
{ layer: 'testsavant_content', confidence: 0.95 },
{ layer: 'transcript_classifier', confidence: 0.0, meta: { degraded: true } },
],
tabId: 1,
...overrides,
};
}
describe('sidepanel review-flow E2E', () => {
test.skipIf(!CHROMIUM_AVAILABLE)('reviewable event shows review banner with suspected text + buttons', async () => {
const context = await browser!.newContext();
const page = await context.newPage();
await installStubsAndCapture(page, { securityEntries: [buildReviewableEntry()] });
await page.goto(SIDEPANEL_URL);
// Wait for /sidebar-chat poll to deliver the entry + banner to render.
await page.waitForFunction(
() => {
const b = document.getElementById('security-banner') as HTMLElement | null;
return !!b && b.style.display !== 'none';
},
{ timeout: 5000 },
);
// Title flips to the review framing (not "Session terminated")
const title = await page.$eval('#security-banner-title', (el) => el.textContent);
expect(title).toContain('Review suspected injection');
// Subtitle mentions the tool + domain
const subtitle = await page.$eval('#security-banner-subtitle', (el) => el.textContent);
expect(subtitle).toContain('Bash');
expect(subtitle).toContain('news.ycombinator.com');
expect(subtitle).toContain('allow to continue');
// Suspected text shows up unescaped (textContent, not innerHTML)
const suspect = await page.$eval('#security-banner-suspect', (el) => el.textContent);
expect(suspect).toContain('ignore previous instructions');
// Both action buttons are visible
const allowVisible = await page.locator('#security-banner-btn-allow').isVisible();
const blockVisible = await page.locator('#security-banner-btn-block').isVisible();
expect(allowVisible).toBe(true);
expect(blockVisible).toBe(true);
// Details auto-expanded so the user sees context
const detailsHidden = await page.$eval('#security-banner-details', (el) => (el as HTMLElement).hidden);
expect(detailsHidden).toBe(false);
await context.close();
}, 15000);
test.skipIf(!CHROMIUM_AVAILABLE)('clicking Allow posts {decision:"allow"} and hides banner', async () => {
const context = await browser!.newContext();
const page = await context.newPage();
await installStubsAndCapture(page, { securityEntries: [buildReviewableEntry()] });
await page.goto(SIDEPANEL_URL);
await page.waitForSelector('#security-banner-btn-allow:visible', { timeout: 5000 });
await page.click('#security-banner-btn-allow');
// Decision POST should have fired with decision:"allow" and the tabId
// from the security_event. Give the fetch promise a tick to resolve.
await page.waitForFunction(
() => (window as any).__decisionCalls?.length > 0,
{ timeout: 2000 },
);
const calls = await page.evaluate(() => (window as any).__decisionCalls);
expect(calls).toHaveLength(1);
expect(calls[0].decision).toBe('allow');
expect(calls[0].tabId).toBe(1);
expect(calls[0].reason).toBe('user');
// Banner should hide optimistically after the POST
await page.waitForFunction(
() => {
const b = document.getElementById('security-banner') as HTMLElement | null;
return !!b && b.style.display === 'none';
},
{ timeout: 2000 },
);
await context.close();
}, 15000);
test.skipIf(!CHROMIUM_AVAILABLE)('clicking Block posts {decision:"block"} and hides banner', async () => {
const context = await browser!.newContext();
const page = await context.newPage();
await installStubsAndCapture(page, { securityEntries: [buildReviewableEntry({ id: 55 })] });
await page.goto(SIDEPANEL_URL);
await page.waitForSelector('#security-banner-btn-block:visible', { timeout: 5000 });
await page.click('#security-banner-btn-block');
await page.waitForFunction(
() => (window as any).__decisionCalls?.length > 0,
{ timeout: 2000 },
);
const calls = await page.evaluate(() => (window as any).__decisionCalls);
expect(calls).toHaveLength(1);
expect(calls[0].decision).toBe('block');
expect(calls[0].tabId).toBe(1);
await page.waitForFunction(
() => {
const b = document.getElementById('security-banner') as HTMLElement | null;
return !!b && b.style.display === 'none';
},
{ timeout: 2000 },
);
await context.close();
}, 15000);
test.skipIf(!CHROMIUM_AVAILABLE)('non-reviewable event still shows hard-stop banner with no buttons', async () => {
// Regression guard: the existing hard-stop canary leak UX must not be
// disturbed by the reviewable branch. An event without reviewable:true
// keeps the old behavior.
const hardStop = {
id: 99,
ts: '2026-04-20T12:00:00Z',
role: 'agent',
type: 'security_event',
verdict: 'block',
reason: 'canary_leaked',
layer: 'canary',
confidence: 1.0,
domain: 'attacker.example.com',
channel: 'tool_use:Bash',
tabId: 1,
};
const context = await browser!.newContext();
const page = await context.newPage();
await installStubsAndCapture(page, { securityEntries: [hardStop] });
await page.goto(SIDEPANEL_URL);
await page.waitForFunction(
() => {
const b = document.getElementById('security-banner') as HTMLElement | null;
return !!b && b.style.display !== 'none';
},
{ timeout: 5000 },
);
const title = await page.$eval('#security-banner-title', (el) => el.textContent);
expect(title).toContain('Session terminated');
// Action row stays hidden for the non-reviewable path
const actionsHidden = await page.$eval('#security-banner-actions', (el) => (el as HTMLElement).hidden);
expect(actionsHidden).toBe(true);
await context.close();
}, 15000);
test.skipIf(!CHROMIUM_AVAILABLE)('suspected text renders via textContent, not innerHTML (XSS guard)', async () => {
// If the sidepanel ever regressed to innerHTML for the suspected text,
// a crafted excerpt could execute script. This test uses one; if the
// <script> runs, window.__xss gets set. It must remain undefined.
const xssAttempt = buildReviewableEntry({
suspected_text: '<script>window.__xss = "pwn"</script><img src=x onerror="window.__xss=\'onerror\'">',
});
const context = await browser!.newContext();
const page = await context.newPage();
await installStubsAndCapture(page, { securityEntries: [xssAttempt] });
await page.goto(SIDEPANEL_URL);
await page.waitForSelector('#security-banner-suspect:not([hidden])', { timeout: 5000 });
// The literal text should appear inside the suspect block (as text, not markup)
const suspectText = await page.$eval('#security-banner-suspect', (el) => el.textContent);
expect(suspectText).toContain('<script>');
// No script executed
const xssFlag = await page.evaluate(() => (window as any).__xss);
expect(xssFlag).toBeUndefined();
await context.close();
}, 15000);
});
-226
View File
@@ -1,226 +0,0 @@
/**
* Layer 3: Sidebar agent round-trip tests.
* Starts server + sidebar-agent together. Mocks the `claude` binary with a shell
* script that outputs canned stream-json. Verifies events flow end-to-end:
* POST /sidebar-command queue sidebar-agent mock claude events /sidebar-chat
*/
import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
import { spawn, type Subprocess } from 'bun';
import * as fs from 'fs';
import * as os from 'os';
import * as path from 'path';
let serverProc: Subprocess | null = null;
let agentProc: Subprocess | null = null;
let serverPort: number = 0;
let authToken: string = '';
let tmpDir: string = '';
let stateFile: string = '';
let queueFile: string = '';
let mockBinDir: string = '';
async function api(pathname: string, opts: RequestInit = {}): Promise<Response> {
const headers: Record<string, string> = {
'Content-Type': 'application/json',
...(opts.headers as Record<string, string> || {}),
};
if (!headers['Authorization'] && authToken) {
headers['Authorization'] = `Bearer ${authToken}`;
}
return fetch(`http://127.0.0.1:${serverPort}${pathname}`, { ...opts, headers });
}
async function resetState() {
await api('/sidebar-session/new', { method: 'POST' });
fs.writeFileSync(queueFile, '');
}
async function pollChatUntil(
predicate: (entries: any[]) => boolean,
timeoutMs = 10000,
): Promise<any[]> {
const deadline = Date.now() + timeoutMs;
while (Date.now() < deadline) {
const resp = await api('/sidebar-chat?after=0');
const data = await resp.json();
if (predicate(data.entries)) return data.entries;
await new Promise(r => setTimeout(r, 300));
}
// Return whatever we have on timeout
const resp = await api('/sidebar-chat?after=0');
return (await resp.json()).entries;
}
function writeMockClaude(script: string) {
const mockPath = path.join(mockBinDir, 'claude');
fs.writeFileSync(mockPath, script, { mode: 0o755 });
}
beforeAll(async () => {
tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'sidebar-roundtrip-'));
stateFile = path.join(tmpDir, 'browse.json');
queueFile = path.join(tmpDir, 'sidebar-queue.jsonl');
mockBinDir = path.join(tmpDir, 'bin');
fs.mkdirSync(mockBinDir, { recursive: true });
fs.mkdirSync(path.dirname(queueFile), { recursive: true });
// Write default mock claude that outputs canned events
writeMockClaude(`#!/bin/bash
echo '{"type":"system","session_id":"mock-session-123"}'
echo '{"type":"assistant","message":{"content":[{"type":"text","text":"I can see the page. It looks like a test fixture."}]}}'
echo '{"type":"result","result":"Done."}'
`);
// Start server (no browser)
const serverScript = path.resolve(__dirname, '..', 'src', 'server.ts');
serverProc = spawn(['bun', 'run', serverScript], {
env: {
...process.env,
BROWSE_STATE_FILE: stateFile,
BROWSE_HEADLESS_SKIP: '1',
BROWSE_PORT: '0',
SIDEBAR_QUEUE_PATH: queueFile,
BROWSE_IDLE_TIMEOUT: '300',
},
stdio: ['ignore', 'pipe', 'pipe'],
});
// Wait for server
const deadline = Date.now() + 15000;
while (Date.now() < deadline) {
if (fs.existsSync(stateFile)) {
try {
const state = JSON.parse(fs.readFileSync(stateFile, 'utf-8'));
if (state.port && state.token) {
serverPort = state.port;
authToken = state.token;
break;
}
} catch {}
}
await new Promise(r => setTimeout(r, 100));
}
if (!serverPort) throw new Error('Server did not start in time');
// Start sidebar-agent with mock claude on PATH
const agentScript = path.resolve(__dirname, '..', 'src', 'sidebar-agent.ts');
agentProc = spawn(['bun', 'run', agentScript], {
env: {
...process.env,
PATH: `${mockBinDir}:${process.env.PATH}`,
BROWSE_SERVER_PORT: String(serverPort),
BROWSE_STATE_FILE: stateFile,
SIDEBAR_QUEUE_PATH: queueFile,
SIDEBAR_AGENT_TIMEOUT: '10000',
BROWSE_BIN: 'browse', // doesn't matter, mock claude doesn't use it
},
stdio: ['ignore', 'pipe', 'pipe'],
});
// Give sidebar-agent time to start polling
await new Promise(r => setTimeout(r, 1000));
}, 20000);
afterAll(() => {
if (agentProc) { try { agentProc.kill(); } catch {} }
if (serverProc) { try { serverProc.kill(); } catch {} }
try { fs.rmSync(tmpDir, { recursive: true, force: true }); } catch {}
});
describe('sidebar-agent round-trip', () => {
test('full message round-trip with mock claude', async () => {
await resetState();
// Send a command
const resp = await api('/sidebar-command', {
method: 'POST',
body: JSON.stringify({
message: 'what is on this page?',
activeTabUrl: 'https://example.com/test',
}),
});
expect(resp.status).toBe(200);
// Wait for mock claude to process and events to arrive
const entries = await pollChatUntil(
(entries) => entries.some((e: any) => e.type === 'agent_done'),
15000,
);
// Verify the flow: user message → agent_start → text → agent_done
const userEntry = entries.find((e: any) => e.role === 'user');
expect(userEntry).toBeDefined();
expect(userEntry.message).toBe('what is on this page?');
// The mock claude outputs text — check for any agent text entry
const textEntries = entries.filter((e: any) => e.role === 'agent' && (e.type === 'text' || e.type === 'result'));
expect(textEntries.length).toBeGreaterThan(0);
const doneEntry = entries.find((e: any) => e.type === 'agent_done');
expect(doneEntry).toBeDefined();
// Agent should be back to idle
const session = await (await api('/sidebar-session')).json();
expect(session.agent.status).toBe('idle');
}, 20000);
test('claude crash produces agent_error', async () => {
await resetState();
// Replace mock claude with one that crashes
writeMockClaude(`#!/bin/bash
echo '{"type":"system","session_id":"crash-test"}' >&2
exit 1
`);
await api('/sidebar-command', {
method: 'POST',
body: JSON.stringify({ message: 'crash test' }),
});
// Wait for agent_done (sidebar-agent sends agent_done even on crash via proc.on('close'))
const entries = await pollChatUntil(
(entries) => entries.some((e: any) => e.type === 'agent_done' || e.type === 'agent_error'),
15000,
);
// Agent should recover to idle
const session = await (await api('/sidebar-session')).json();
expect(session.agent.status).toBe('idle');
// Restore working mock
writeMockClaude(`#!/bin/bash
echo '{"type":"assistant","message":{"content":[{"type":"text","text":"recovered"}]}}'
`);
}, 20000);
test('sequential queue drain', async () => {
await resetState();
// Restore working mock
writeMockClaude(`#!/bin/bash
echo '{"type":"assistant","message":{"content":[{"type":"text","text":"response to: '"'"'$*'"'"'"}]}}'
`);
// Send two messages rapidly — first processes, second queues
await api('/sidebar-command', {
method: 'POST',
body: JSON.stringify({ message: 'first message' }),
});
await api('/sidebar-command', {
method: 'POST',
body: JSON.stringify({ message: 'second message' }),
});
// Wait for both to complete (two agent_done events)
const entries = await pollChatUntil(
(entries) => entries.filter((e: any) => e.type === 'agent_done').length >= 2,
20000,
);
// Both user messages should be in chat
const userEntries = entries.filter((e: any) => e.role === 'user');
expect(userEntries.length).toBeGreaterThanOrEqual(2);
}, 25000);
});
-562
View File
@@ -1,562 +0,0 @@
/**
* Tests for sidebar agent queue parsing and inbox writing.
*
* sidebar-agent.ts functions are not exported (it's an entry-point script),
* so we test the same logic inline: JSONL parsing, writeToInbox filesystem
* behavior, and edge cases.
*/
import { describe, test, expect, beforeEach, afterEach } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
import * as os from 'os';
// ─── Helpers: replicate sidebar-agent logic for unit testing ──────
/** Parse a single JSONL line — same logic as sidebar-agent poll() */
function parseQueueLine(line: string): any | null {
if (!line.trim()) return null;
try {
const entry = JSON.parse(line);
if (!entry.message && !entry.prompt) return null;
return entry;
} catch {
return null;
}
}
/** Read all valid entries from a JSONL string — same as countLines + readLine loop */
function parseQueueFile(content: string): any[] {
const entries: any[] = [];
const lines = content.split('\n').filter(Boolean);
for (const line of lines) {
const entry = parseQueueLine(line);
if (entry) entries.push(entry);
}
return entries;
}
/** Write to inbox — extracted logic from sidebar-agent.ts writeToInbox() */
function writeToInbox(
gitRoot: string,
message: string,
pageUrl?: string,
sessionId?: string,
): string | null {
if (!gitRoot) return null;
const inboxDir = path.join(gitRoot, '.context', 'sidebar-inbox');
fs.mkdirSync(inboxDir, { recursive: true });
const now = new Date();
const timestamp = now.toISOString().replace(/:/g, '-');
const filename = `${timestamp}-observation.json`;
const tmpFile = path.join(inboxDir, `.${filename}.tmp`);
const finalFile = path.join(inboxDir, filename);
const inboxMessage = {
type: 'observation',
timestamp: now.toISOString(),
page: { url: pageUrl || 'unknown', title: '' },
userMessage: message,
sidebarSessionId: sessionId || 'unknown',
};
fs.writeFileSync(tmpFile, JSON.stringify(inboxMessage, null, 2));
fs.renameSync(tmpFile, finalFile);
return finalFile;
}
/** Shorten paths — same logic as sidebar-agent.ts shorten() */
function shorten(str: string): string {
return str
.replace(/\/Users\/[^/]+/g, '~')
.replace(/\/conductor\/workspaces\/[^/]+\/[^/]+/g, '')
.replace(/\.claude\/skills\/gstack\//g, '')
.replace(/browse\/dist\/browse/g, '$B');
}
/** describeToolCall — replicated from sidebar-agent.ts for unit testing */
function describeToolCall(tool: string, input: any): string {
if (!input) return '';
if (tool === 'Bash' && input.command) {
const cmd = input.command;
const browseMatch = cmd.match(/\$B\s+(\w+)|browse[^\s]*\s+(\w+)/);
if (browseMatch) {
const browseCmd = browseMatch[1] || browseMatch[2];
const args = cmd.split(/\s+/).slice(2).join(' ');
switch (browseCmd) {
case 'goto': return `Opening ${args.replace(/['"]/g, '')}`;
case 'snapshot': return args.includes('-i') ? 'Scanning for interactive elements' : args.includes('-D') ? 'Checking what changed' : 'Taking a snapshot of the page';
case 'screenshot': return `Saving screenshot${args ? ` to ${shorten(args)}` : ''}`;
case 'click': return `Clicking ${args}`;
case 'fill': { const parts = args.split(/\s+/); return `Typing "${parts.slice(1).join(' ')}" into ${parts[0]}`; }
case 'text': return 'Reading page text';
case 'html': return args ? `Reading HTML of ${args}` : 'Reading full page HTML';
case 'links': return 'Finding all links on the page';
case 'forms': return 'Looking for forms';
case 'console': return 'Checking browser console for errors';
case 'network': return 'Checking network requests';
case 'url': return 'Checking current URL';
case 'back': return 'Going back';
case 'forward': return 'Going forward';
case 'reload': return 'Reloading the page';
case 'scroll': return args ? `Scrolling to ${args}` : 'Scrolling down';
case 'wait': return `Waiting for ${args}`;
case 'inspect': return args ? `Inspecting CSS of ${args}` : 'Getting CSS for last picked element';
case 'style': return `Changing CSS: ${args}`;
case 'cleanup': return 'Removing page clutter (ads, popups, banners)';
case 'prettyscreenshot': return 'Taking a clean screenshot';
case 'css': return `Checking CSS property: ${args}`;
case 'is': return `Checking if element is ${args}`;
case 'diff': return `Comparing ${args}`;
case 'responsive': return 'Taking screenshots at mobile, tablet, and desktop sizes';
case 'status': return 'Checking browser status';
case 'tabs': return 'Listing open tabs';
case 'focus': return 'Bringing browser to front';
case 'select': return `Selecting option in ${args}`;
case 'hover': return `Hovering over ${args}`;
case 'viewport': return `Setting viewport to ${args}`;
case 'upload': return `Uploading file to ${args.split(/\s+/)[0]}`;
default: return `Running browse ${browseCmd} ${args}`.trim();
}
}
if (cmd.includes('git ')) return `Running: ${shorten(cmd)}`;
let short = shorten(cmd);
return short.length > 100 ? short.slice(0, 100) + '…' : short;
}
if (tool === 'Read' && input.file_path) return `Reading ${shorten(input.file_path)}`;
if (tool === 'Edit' && input.file_path) return `Editing ${shorten(input.file_path)}`;
if (tool === 'Write' && input.file_path) return `Writing ${shorten(input.file_path)}`;
if (tool === 'Grep' && input.pattern) return `Searching for "${input.pattern}"`;
if (tool === 'Glob' && input.pattern) return `Finding files matching ${input.pattern}`;
try { return shorten(JSON.stringify(input)).slice(0, 80); } catch { return ''; }
}
// ─── Test setup ──────────────────────────────────────────────────
let tmpDir: string;
beforeEach(() => {
tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'sidebar-agent-test-'));
});
afterEach(() => {
fs.rmSync(tmpDir, { recursive: true, force: true });
});
// ─── Queue File Parsing ─────────────────────────────────────────
describe('queue file parsing', () => {
test('valid JSONL line parsed correctly', () => {
const line = JSON.stringify({ message: 'hello', prompt: 'check this', pageUrl: 'https://example.com' });
const entry = parseQueueLine(line);
expect(entry).not.toBeNull();
expect(entry.message).toBe('hello');
expect(entry.prompt).toBe('check this');
expect(entry.pageUrl).toBe('https://example.com');
});
test('malformed JSON line skipped without crash', () => {
const entry = parseQueueLine('this is not json {{{');
expect(entry).toBeNull();
});
test('valid JSON without message or prompt is skipped', () => {
const line = JSON.stringify({ foo: 'bar' });
const entry = parseQueueLine(line);
expect(entry).toBeNull();
});
test('empty file returns no entries', () => {
const entries = parseQueueFile('');
expect(entries).toEqual([]);
});
test('file with blank lines returns no entries', () => {
const entries = parseQueueFile('\n\n\n');
expect(entries).toEqual([]);
});
test('mixed valid and invalid lines', () => {
const content = [
JSON.stringify({ message: 'first' }),
'not json',
JSON.stringify({ unrelated: true }),
JSON.stringify({ message: 'second', prompt: 'do stuff' }),
].join('\n');
const entries = parseQueueFile(content);
expect(entries.length).toBe(2);
expect(entries[0].message).toBe('first');
expect(entries[1].message).toBe('second');
});
});
// ─── writeToInbox ────────────────────────────────────────────────
describe('writeToInbox', () => {
test('creates .context/sidebar-inbox/ directory', () => {
writeToInbox(tmpDir, 'test message');
const inboxDir = path.join(tmpDir, '.context', 'sidebar-inbox');
expect(fs.existsSync(inboxDir)).toBe(true);
expect(fs.statSync(inboxDir).isDirectory()).toBe(true);
});
test('writes valid JSON file', () => {
const filePath = writeToInbox(tmpDir, 'test message', 'https://example.com', 'session-123');
expect(filePath).not.toBeNull();
expect(fs.existsSync(filePath!)).toBe(true);
const data = JSON.parse(fs.readFileSync(filePath!, 'utf-8'));
expect(data.type).toBe('observation');
expect(data.userMessage).toBe('test message');
expect(data.page.url).toBe('https://example.com');
expect(data.sidebarSessionId).toBe('session-123');
expect(data.timestamp).toBeTruthy();
});
test('atomic write — final file exists, no .tmp left', () => {
const filePath = writeToInbox(tmpDir, 'atomic test');
expect(filePath).not.toBeNull();
expect(fs.existsSync(filePath!)).toBe(true);
// Check no .tmp files remain in the inbox directory
const inboxDir = path.join(tmpDir, '.context', 'sidebar-inbox');
const files = fs.readdirSync(inboxDir);
const tmpFiles = files.filter(f => f.endsWith('.tmp'));
expect(tmpFiles.length).toBe(0);
// Final file should end with -observation.json
const jsonFiles = files.filter(f => f.endsWith('-observation.json') && !f.startsWith('.'));
expect(jsonFiles.length).toBe(1);
});
test('handles missing git root gracefully', () => {
const result = writeToInbox('', 'test');
expect(result).toBeNull();
});
test('defaults pageUrl to unknown when not provided', () => {
const filePath = writeToInbox(tmpDir, 'no url provided');
expect(filePath).not.toBeNull();
const data = JSON.parse(fs.readFileSync(filePath!, 'utf-8'));
expect(data.page.url).toBe('unknown');
});
test('defaults sessionId to unknown when not provided', () => {
const filePath = writeToInbox(tmpDir, 'no session');
expect(filePath).not.toBeNull();
const data = JSON.parse(fs.readFileSync(filePath!, 'utf-8'));
expect(data.sidebarSessionId).toBe('unknown');
});
test('multiple writes create separate files', () => {
writeToInbox(tmpDir, 'message 1');
// Tiny delay to ensure different timestamps
const t = Date.now();
while (Date.now() === t) {} // spin until next ms
writeToInbox(tmpDir, 'message 2');
const inboxDir = path.join(tmpDir, '.context', 'sidebar-inbox');
const files = fs.readdirSync(inboxDir).filter(f => f.endsWith('.json') && !f.startsWith('.'));
expect(files.length).toBe(2);
});
});
// ─── describeToolCall (verbose narration) ────────────────────────
describe('describeToolCall', () => {
// Browse navigation commands
test('goto → plain English with URL', () => {
const result = describeToolCall('Bash', { command: '$B goto https://example.com' });
expect(result).toBe('Opening https://example.com');
});
test('goto strips quotes from URL', () => {
const result = describeToolCall('Bash', { command: '$B goto "https://example.com"' });
expect(result).toBe('Opening https://example.com');
});
test('url → checking current URL', () => {
expect(describeToolCall('Bash', { command: '$B url' })).toBe('Checking current URL');
});
test('back/forward/reload → plain English', () => {
expect(describeToolCall('Bash', { command: '$B back' })).toBe('Going back');
expect(describeToolCall('Bash', { command: '$B forward' })).toBe('Going forward');
expect(describeToolCall('Bash', { command: '$B reload' })).toBe('Reloading the page');
});
// Snapshot variants
test('snapshot -i → scanning for interactive elements', () => {
expect(describeToolCall('Bash', { command: '$B snapshot -i' })).toBe('Scanning for interactive elements');
});
test('snapshot -D → checking what changed', () => {
expect(describeToolCall('Bash', { command: '$B snapshot -D' })).toBe('Checking what changed');
});
test('snapshot (plain) → taking a snapshot', () => {
expect(describeToolCall('Bash', { command: '$B snapshot' })).toBe('Taking a snapshot of the page');
});
// Interaction commands
test('click → clicking element', () => {
expect(describeToolCall('Bash', { command: '$B click @e3' })).toBe('Clicking @e3');
});
test('fill → typing into element', () => {
expect(describeToolCall('Bash', { command: '$B fill @e4 "hello world"' })).toBe('Typing ""hello world"" into @e4');
});
test('scroll with selector → scrolling to element', () => {
expect(describeToolCall('Bash', { command: '$B scroll .footer' })).toBe('Scrolling to .footer');
});
test('scroll without args → scrolling down', () => {
expect(describeToolCall('Bash', { command: '$B scroll' })).toBe('Scrolling down');
});
// Reading commands
test('text → reading page text', () => {
expect(describeToolCall('Bash', { command: '$B text' })).toBe('Reading page text');
});
test('html with selector → reading HTML of element', () => {
expect(describeToolCall('Bash', { command: '$B html .header' })).toBe('Reading HTML of .header');
});
test('html without selector → reading full page HTML', () => {
expect(describeToolCall('Bash', { command: '$B html' })).toBe('Reading full page HTML');
});
test('links → finding all links', () => {
expect(describeToolCall('Bash', { command: '$B links' })).toBe('Finding all links on the page');
});
test('console → checking console', () => {
expect(describeToolCall('Bash', { command: '$B console' })).toBe('Checking browser console for errors');
});
// Inspector commands
test('inspect with selector → inspecting CSS', () => {
expect(describeToolCall('Bash', { command: '$B inspect .header' })).toBe('Inspecting CSS of .header');
});
test('inspect without args → getting last picked element', () => {
expect(describeToolCall('Bash', { command: '$B inspect' })).toBe('Getting CSS for last picked element');
});
test('style → changing CSS', () => {
expect(describeToolCall('Bash', { command: '$B style .header color red' })).toBe('Changing CSS: .header color red');
});
test('cleanup → removing page clutter', () => {
expect(describeToolCall('Bash', { command: '$B cleanup --all' })).toBe('Removing page clutter (ads, popups, banners)');
});
// Visual commands
test('screenshot → saving screenshot', () => {
expect(describeToolCall('Bash', { command: '$B screenshot /tmp/shot.png' })).toBe('Saving screenshot to /tmp/shot.png');
});
test('screenshot without path', () => {
expect(describeToolCall('Bash', { command: '$B screenshot' })).toBe('Saving screenshot');
});
test('responsive → multi-size screenshots', () => {
expect(describeToolCall('Bash', { command: '$B responsive' })).toBe('Taking screenshots at mobile, tablet, and desktop sizes');
});
// Non-browse tools
test('Read tool → reading file', () => {
expect(describeToolCall('Read', { file_path: '/Users/foo/project/src/app.ts' })).toBe('Reading ~/project/src/app.ts');
});
test('Grep tool → searching for pattern', () => {
expect(describeToolCall('Grep', { pattern: 'handleClick' })).toBe('Searching for "handleClick"');
});
test('Glob tool → finding files', () => {
expect(describeToolCall('Glob', { pattern: '**/*.tsx' })).toBe('Finding files matching **/*.tsx');
});
test('Edit tool → editing file', () => {
expect(describeToolCall('Edit', { file_path: '/Users/foo/src/main.ts' })).toBe('Editing ~/src/main.ts');
});
// Edge cases
test('null input → empty string', () => {
expect(describeToolCall('Bash', null)).toBe('');
});
test('unknown browse command → generic description', () => {
expect(describeToolCall('Bash', { command: '$B newtab https://foo.com' })).toContain('newtab');
});
test('non-browse bash → shortened command', () => {
expect(describeToolCall('Bash', { command: 'echo hello' })).toBe('echo hello');
});
test('full browse binary path recognized', () => {
const result = describeToolCall('Bash', { command: '/Users/garrytan/.claude/skills/gstack/browse/dist/browse goto https://example.com' });
expect(result).toBe('Opening https://example.com');
});
test('tab command → switching tab', () => {
expect(describeToolCall('Bash', { command: '$B tab 2' })).toContain('tab');
});
});
// ─── Per-tab agent concurrency (source code validation) ──────────
describe('per-tab agent concurrency', () => {
const serverSrc = fs.readFileSync(path.join(__dirname, '..', 'src', 'server.ts'), 'utf-8');
const agentSrc = fs.readFileSync(path.join(__dirname, '..', 'src', 'sidebar-agent.ts'), 'utf-8');
test('server has per-tab agent state map', () => {
expect(serverSrc).toContain('tabAgents');
expect(serverSrc).toContain('TabAgentState');
expect(serverSrc).toContain('getTabAgent');
});
test('server returns per-tab agent status in /sidebar-chat', () => {
expect(serverSrc).toContain('getTabAgentStatus');
expect(serverSrc).toContain('tabAgentStatus');
});
test('spawnClaude accepts forTabId parameter', () => {
const spawnFn = serverSrc.slice(
serverSrc.indexOf('function spawnClaude('),
serverSrc.indexOf('\nfunction ', serverSrc.indexOf('function spawnClaude(') + 1),
);
expect(spawnFn).toContain('forTabId');
expect(spawnFn).toContain('tabState.status');
});
test('sidebar-command endpoint uses per-tab agent state', () => {
expect(serverSrc).toContain('msgTabId');
expect(serverSrc).toContain('tabState.status');
expect(serverSrc).toContain('tabState.queue');
});
test('agent event handler resets per-tab state', () => {
expect(serverSrc).toContain('eventTabId');
expect(serverSrc).toContain('tabState.status = \'idle\'');
});
test('agent event handler processes per-tab queue', () => {
// After agent_done, should process next message from THIS tab's queue
expect(serverSrc).toContain('tabState.queue.length > 0');
expect(serverSrc).toContain('tabState.queue.shift');
});
test('sidebar-agent uses per-tab processing set', () => {
expect(agentSrc).toContain('processingTabs');
expect(agentSrc).not.toContain('isProcessing');
});
test('sidebar-agent sends tabId with all events', () => {
// sendEvent should accept tabId parameter
expect(agentSrc).toContain('async function sendEvent(event: Record<string, any>, tabId?: number)');
// askClaude destructures tabId from queue entry (regex tolerates
// additional fields like `canary` and `pageUrl` from security module).
expect(agentSrc).toMatch(
/const \{[^}]*\bprompt\b[^}]*\bargs\b[^}]*\bstateFile\b[^}]*\bcwd\b[^}]*\btabId\b[^}]*\}/
);
});
test('sidebar-agent allows concurrent agents across tabs', () => {
// poll() should not block globally — it should check per-tab
expect(agentSrc).toContain('processingTabs.has(tid)');
// askClaude should be fire-and-forget (no await blocking the loop)
expect(agentSrc).toContain('askClaude(entry).catch');
});
test('queue entries include tabId', () => {
const spawnFn = serverSrc.slice(
serverSrc.indexOf('function spawnClaude('),
serverSrc.indexOf('\nfunction ', serverSrc.indexOf('function spawnClaude(') + 1),
);
expect(spawnFn).toContain('tabId: agentTabId');
});
test('health check monitors all per-tab agents', () => {
expect(serverSrc).toContain('for (const [tid, state] of tabAgents)');
});
});
describe('BROWSE_TAB tab pinning (cross-tab isolation)', () => {
const serverSrc = fs.readFileSync(path.join(__dirname, '..', 'src', 'server.ts'), 'utf-8');
const agentSrc = fs.readFileSync(path.join(__dirname, '..', 'src', 'sidebar-agent.ts'), 'utf-8');
const cliSrc = fs.readFileSync(path.join(__dirname, '..', 'src', 'cli.ts'), 'utf-8');
test('sidebar-agent passes BROWSE_TAB env var to claude process', () => {
// The env block should include BROWSE_TAB set to the tab ID
expect(agentSrc).toContain('BROWSE_TAB');
expect(agentSrc).toContain('String(tid)');
});
test('CLI reads BROWSE_TAB and sends tabId in command body', () => {
// BROWSE_TAB env var is still honored (sidebar-agent path). After the
// make-pdf refactor, the CLI layer now also accepts --tab-id <N>, with
// the CLI flag taking precedence over the env var. Both resolve to the
// same `tabId` body field.
expect(cliSrc).toContain('process.env.BROWSE_TAB');
expect(cliSrc).toContain('parseInt(envTab, 10)');
});
test('handleCommandInternal accepts tabId from request body', () => {
const handleFn = serverSrc.slice(
serverSrc.indexOf('async function handleCommandInternal('),
serverSrc.indexOf('\n/** HTTP wrapper', serverSrc.indexOf('async function handleCommandInternal(') + 1) > 0
? serverSrc.indexOf('\n/** HTTP wrapper', serverSrc.indexOf('async function handleCommandInternal(') + 1)
: serverSrc.indexOf('\nasync function ', serverSrc.indexOf('async function handleCommandInternal(') + 200),
);
// Should destructure tabId from body
expect(handleFn).toContain('tabId');
// Should save and restore the active tab
expect(handleFn).toContain('savedTabId');
expect(handleFn).toContain('switchTab(tabId');
});
test('handleCommandInternal restores active tab after command (success path)', () => {
// On success, should restore savedTabId without stealing focus
const handleFn = serverSrc.slice(
serverSrc.indexOf('async function handleCommandInternal('),
serverSrc.length,
);
// Count restore calls — should appear in both success and error paths
const restoreCount = (handleFn.match(/switchTab\(savedTabId/g) || []).length;
expect(restoreCount).toBeGreaterThanOrEqual(2); // success + error paths
});
test('handleCommandInternal restores active tab on error path', () => {
// The catch block should also restore
const catchBlock = serverSrc.slice(
serverSrc.indexOf('} catch (err: any) {', serverSrc.indexOf('async function handleCommandInternal(')),
);
expect(catchBlock).toContain('switchTab(savedTabId');
});
test('tab pinning only activates when tabId is provided', () => {
const handleFn = serverSrc.slice(
serverSrc.indexOf('async function handleCommandInternal('),
serverSrc.indexOf('try {', serverSrc.indexOf('async function handleCommandInternal(') + 1),
);
// Should check tabId is not undefined/null before switching
expect(handleFn).toContain('tabId !== undefined');
expect(handleFn).toContain('tabId !== null');
});
test('CLI only sends tabId when it is a valid number', () => {
// Body should conditionally include tabId. Historically that was keyed off
// the BROWSE_TAB env var. After the make-pdf refactor, the CLI also honors
// a --tab-id <N> flag on the CLI itself, so the check is "tabId defined
// AND not NaN" rather than literally inspecting the env var.
expect(cliSrc).toContain('tabId !== undefined && !isNaN(tabId)');
});
});
+256
View File
@@ -0,0 +1,256 @@
/**
* Regression: sidebar layout invariants after the chat-tab rip.
*
* The Chrome side panel used to host two surfaces: Chat (one-shot
* `claude -p` queue) and Terminal (interactive PTY). Chat was ripped
* once the PTY proved out sidebar-agent.ts is gone, the chat queue
* endpoints are gone, and the primary-tab nav (Terminal | Chat) is
* gone. Terminal is now the sole primary surface.
*
* This file locks the load-bearing invariants of that layout so a
* future refactor can't silently re-introduce the old surface or break
* the new one.
*/
import { describe, test, expect } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
const HTML = fs.readFileSync(path.join(import.meta.dir, '../../extension/sidepanel.html'), 'utf-8');
const JS = fs.readFileSync(path.join(import.meta.dir, '../../extension/sidepanel.js'), 'utf-8');
const TERM_JS = fs.readFileSync(path.join(import.meta.dir, '../../extension/sidepanel-terminal.js'), 'utf-8');
const MANIFEST = JSON.parse(fs.readFileSync(path.join(import.meta.dir, '../../extension/manifest.json'), 'utf-8'));
describe('sidebar: chat tab + nav are removed, Terminal is sole primary surface', () => {
test('No primary-tab nav element exists', () => {
expect(HTML).not.toContain('class="primary-tabs"');
expect(HTML).not.toContain('data-pane="chat"');
expect(HTML).not.toContain('data-pane="terminal"');
});
test('No <main id="tab-chat"> pane', () => {
expect(HTML).not.toMatch(/<main[^>]*id="tab-chat"/);
expect(HTML).not.toContain('id="chat-messages"');
expect(HTML).not.toContain('id="chat-loading"');
expect(HTML).not.toContain('id="chat-welcome"');
});
test('No chat input / send button / experimental banner', () => {
expect(HTML).not.toContain('class="command-bar"');
expect(HTML).not.toContain('id="command-input"');
expect(HTML).not.toContain('id="send-btn"');
expect(HTML).not.toContain('id="stop-agent-btn"');
expect(HTML).not.toContain('id="experimental-banner"');
});
test('No clear-chat button in footer', () => {
expect(HTML).not.toContain('id="clear-chat"');
});
test('Terminal pane is .active by default and has the toolbar', () => {
expect(HTML).toMatch(/<main[^>]*id="tab-terminal"[^>]*class="tab-content active"/);
expect(HTML).toContain('id="terminal-toolbar"');
expect(HTML).toContain('id="terminal-restart-now"');
});
test('Quick-actions buttons (Cleanup / Screenshot / Cookies) survive in the terminal toolbar', () => {
// Garry explicitly wanted these kept after the chat rip — they drive
// browser actions, not chat.
expect(HTML).toContain('id="chat-cleanup-btn"');
expect(HTML).toContain('id="chat-screenshot-btn"');
expect(HTML).toContain('id="chat-cookies-btn"');
// They live inside the terminal toolbar now (siblings of the Restart
// button), not as a separate strip below all panes.
const toolbarStart = HTML.indexOf('id="terminal-toolbar"');
const toolbarEnd = HTML.indexOf('</div>', toolbarStart);
const toolbarBlock = HTML.slice(toolbarStart, toolbarEnd + 6);
expect(toolbarBlock).toContain('id="chat-cleanup-btn"');
expect(toolbarBlock).toContain('id="chat-screenshot-btn"');
expect(toolbarBlock).toContain('id="chat-cookies-btn"');
});
});
describe('sidepanel.js: chat helpers ripped, terminal-injection helper survives', () => {
test('No primary-tab click handler', () => {
expect(JS).not.toContain("querySelectorAll('.primary-tab')");
expect(JS).not.toContain('activePrimaryPaneId');
});
test('No chat polling, sendMessage, sendChat, stopAgent, or pollTabs', () => {
expect(JS).not.toContain('chatPollInterval');
expect(JS).not.toContain('function sendMessage');
expect(JS).not.toContain('function pollChat');
expect(JS).not.toContain('function pollTabs');
expect(JS).not.toContain('function switchChatTab');
expect(JS).not.toContain('function stopAgent');
expect(JS).not.toContain('function applyChatEnabled');
expect(JS).not.toContain('function showSecurityBanner');
});
test('Cleanup runs through the live PTY (no /sidebar-command POST)', () => {
// The new Cleanup handler injects the prompt straight into claude's
// PTY via gstackInjectToTerminal. The dead code path was a POST to
// /sidebar-command which kicked off a fresh claude -p subprocess.
const cleanup = JS.slice(JS.indexOf('async function runCleanup'));
expect(cleanup).toContain('window.gstackInjectToTerminal');
expect(cleanup).not.toContain('/sidebar-command');
expect(cleanup).not.toContain('addChatEntry');
});
test('Inspector "Send to Code" routes through the live PTY', () => {
const sendBtn = JS.slice(JS.indexOf('inspectorSendBtn.addEventListener'));
expect(sendBtn).toContain('window.gstackInjectToTerminal');
expect(sendBtn).not.toContain("type: 'sidebar-command'");
});
test('updateConnection no longer kicks off chat / tab polling', () => {
const update = JS.slice(JS.indexOf('function updateConnection'), JS.indexOf('function updateConnection') + 1500);
expect(update).not.toContain('chatPollInterval');
expect(update).not.toContain('tabPollInterval');
expect(update).not.toContain('pollChat');
expect(update).not.toContain('pollTabs');
// BUT must still expose the bootstrap globals for sidepanel-terminal.js.
expect(update).toContain('window.gstackServerPort');
expect(update).toContain('window.gstackAuthToken');
});
});
describe('sidepanel-terminal.js: eager auto-connect + injection API', () => {
test('Exposes window.gstackInjectToTerminal for cross-pane use', () => {
expect(TERM_JS).toContain('window.gstackInjectToTerminal');
// Returns false when no live session, true when bytes go out.
const inject = TERM_JS.slice(TERM_JS.indexOf('window.gstackInjectToTerminal'));
expect(inject).toContain('return false');
expect(inject).toContain('return true');
expect(inject).toContain('ws.readyState !== WebSocket.OPEN');
});
test('Auto-connects on init (no keypress required)', () => {
expect(TERM_JS).not.toContain('function onAnyKey');
expect(TERM_JS).not.toContain("addEventListener('keydown'");
expect(TERM_JS).toContain('function tryAutoConnect');
});
test('Repaint hook fires when Terminal pane becomes visible', () => {
// The chat-tab rip removed gstack:primary-tab-changed; we use a
// MutationObserver on #tab-terminal's class attr instead. The
// observer must call repaintIfLive when the .active class returns.
expect(TERM_JS).toContain('MutationObserver');
expect(TERM_JS).toContain("attributeFilter: ['class']");
expect(TERM_JS).toContain('repaintIfLive');
const repaint = TERM_JS.slice(TERM_JS.indexOf('function repaintIfLive'));
expect(repaint).toContain('fitAddon && fitAddon.fit()');
expect(repaint).toContain('term.refresh');
expect(repaint).toContain("type: 'resize'");
});
test('No auto-reconnect on close (Restart is user-initiated)', () => {
const closeOnly = TERM_JS.slice(
TERM_JS.indexOf("ws.addEventListener('close'"),
TERM_JS.indexOf("ws.addEventListener('error'"),
);
expect(closeOnly).not.toContain('setTimeout');
expect(closeOnly).not.toContain('tryAutoConnect');
expect(closeOnly).not.toContain('connect()');
});
test('forceRestart helper closes ws, disposes xterm, returns to IDLE', () => {
expect(TERM_JS).toContain('function forceRestart');
const fn = TERM_JS.slice(TERM_JS.indexOf('function forceRestart'));
expect(fn).toContain('ws && ws.close()');
expect(fn).toContain('term.dispose()');
expect(fn).toContain('STATE.IDLE');
expect(fn).toContain('tryAutoConnect()');
});
test('Both restart buttons (mid-session and ENDED) call forceRestart', () => {
expect(TERM_JS).toContain("els.restart?.addEventListener('click', forceRestart)");
expect(TERM_JS).toContain("els.restartNow?.addEventListener('click', forceRestart)");
});
});
describe('server.ts: chat / sidebar-agent endpoints are gone', () => {
const SERVER_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/server.ts'), 'utf-8');
test('No /sidebar-command, /sidebar-chat, /sidebar-agent/* routes', () => {
expect(SERVER_SRC).not.toMatch(/url\.pathname === ['"]\/sidebar-command['"]/);
expect(SERVER_SRC).not.toMatch(/url\.pathname === ['"]\/sidebar-chat['"]/);
expect(SERVER_SRC).not.toMatch(/url\.pathname\.startsWith\(['"]\/sidebar-agent\//);
expect(SERVER_SRC).not.toMatch(/url\.pathname === ['"]\/sidebar-agent\/event['"]/);
expect(SERVER_SRC).not.toMatch(/url\.pathname === ['"]\/sidebar-tabs['"]/);
expect(SERVER_SRC).not.toMatch(/url\.pathname === ['"]\/sidebar-session['"]/);
});
test('No chat-related state declarations or helpers', () => {
// Allow the symbol names inside the rip-marker comments — but no
// `let`, `const`, `function`, or `interface` declarations of them.
expect(SERVER_SRC).not.toMatch(/^let agentProcess/m);
expect(SERVER_SRC).not.toMatch(/^let agentStatus/m);
expect(SERVER_SRC).not.toMatch(/^let messageQueue/m);
expect(SERVER_SRC).not.toMatch(/^let sidebarSession/m);
expect(SERVER_SRC).not.toMatch(/^const tabAgents/m);
expect(SERVER_SRC).not.toMatch(/^function pickSidebarModel/m);
expect(SERVER_SRC).not.toMatch(/^function processAgentEvent/m);
expect(SERVER_SRC).not.toMatch(/^function killAgent/m);
expect(SERVER_SRC).not.toMatch(/^function addChatEntry/m);
expect(SERVER_SRC).not.toMatch(/^interface ChatEntry/m);
expect(SERVER_SRC).not.toMatch(/^interface SidebarSession/m);
});
test('/health no longer surfaces agentStatus or messageQueue length', () => {
const health = SERVER_SRC.slice(SERVER_SRC.indexOf("url.pathname === '/health'"));
const slice = health.slice(0, 2000);
expect(slice).not.toContain('agentStatus');
expect(slice).not.toContain('messageQueue');
expect(slice).not.toContain('agentStartTime');
// chatEnabled is hardcoded false now (older clients still see the field).
expect(slice).toMatch(/chatEnabled:\s*false/);
// terminalPort survives.
expect(slice).toContain('terminalPort');
});
});
describe('cli.ts: sidebar-agent is no longer spawned', () => {
const CLI_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/cli.ts'), 'utf-8');
test('No Bun.spawn of sidebar-agent.ts', () => {
expect(CLI_SRC).not.toMatch(/Bun\.spawn\(\s*\['bun',\s*'run',\s*\w*[Aa]gent[Ss]cript\][\s\S]{0,300}sidebar-agent/);
// The variable name `agentScript` was for sidebar-agent. After the
// rip there's only termAgentScript. Allow comments to mention the
// history but not active spawn calls.
expect(CLI_SRC).not.toMatch(/^\s*let agentScript = path\.resolve/m);
});
test('Terminal-agent spawn survives', () => {
expect(CLI_SRC).toContain('terminal-agent.ts');
expect(CLI_SRC).toMatch(/Bun\.spawn\(\['bun',\s*'run',\s*termAgentScript\]/);
});
});
describe('files: sidebar-agent.ts and its tests are deleted', () => {
test('browse/src/sidebar-agent.ts is gone', () => {
expect(fs.existsSync(path.join(import.meta.dir, '../src/sidebar-agent.ts'))).toBe(false);
});
test('sidebar-agent test files are gone', () => {
expect(fs.existsSync(path.join(import.meta.dir, 'sidebar-agent.test.ts'))).toBe(false);
expect(fs.existsSync(path.join(import.meta.dir, 'sidebar-agent-roundtrip.test.ts'))).toBe(false);
});
});
describe('manifest: ws permission + xterm-safe CSP', () => {
test('host_permissions covers ws localhost', () => {
expect(MANIFEST.host_permissions).toContain('ws://127.0.0.1:*/');
});
test('host_permissions still covers http localhost', () => {
expect(MANIFEST.host_permissions).toContain('http://127.0.0.1:*/');
});
test('manifest does NOT add unsafe-eval to extension_pages CSP', () => {
const csp = MANIFEST.content_security_policy;
if (csp && csp.extension_pages) {
expect(csp.extension_pages).not.toContain('unsafe-eval');
}
});
});
+196
View File
@@ -0,0 +1,196 @@
/**
* tab-each fan-out command for the live Terminal pane.
*
* Source-level guards: command is registered, has a description + usage,
* scope-check the inner command, restore the original active tab in a
* finally block (so a mid-batch exception doesn't leave the user looking
* at a tab they didn't choose).
*
* Behavioral logic test: drive handleMetaCommand directly with a mock
* BrowserManager + executeCommand callback. Verify the iteration order,
* the JSON shape, the tab restore, and the chrome:// skip.
*/
import { describe, test, expect } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
import { handleMetaCommand } from '../src/meta-commands';
import { META_COMMANDS, COMMAND_DESCRIPTIONS } from '../src/commands';
const META_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/meta-commands.ts'), 'utf-8');
describe('tab-each: registration', () => {
test('command is in META_COMMANDS', () => {
expect(META_COMMANDS.has('tab-each')).toBe(true);
});
test('has a description and usage entry', () => {
expect(COMMAND_DESCRIPTIONS['tab-each']).toBeDefined();
expect(COMMAND_DESCRIPTIONS['tab-each'].usage).toContain('tab-each');
expect(COMMAND_DESCRIPTIONS['tab-each'].category).toBe('Tabs');
});
});
describe('tab-each: source-level guards', () => {
test('scope-checks the inner command before fanning out', () => {
const block = META_SRC.slice(META_SRC.indexOf("case 'tab-each':"));
expect(block).toContain('checkScope(tokenInfo, innerName)');
// The scope check must run BEFORE the for-loop. If it ran inside the
// loop, a permission failure on the second tab would leave the first
// tab already mutated.
const checkIdx = block.indexOf('checkScope(tokenInfo, innerName)');
const loopIdx = block.indexOf('for (const tab of tabs)');
expect(checkIdx).toBeLessThan(loopIdx);
});
test('restores the original active tab in a finally block', () => {
const block = META_SRC.slice(META_SRC.indexOf("case 'tab-each':"), META_SRC.indexOf("case 'tab-each':") + 4000);
expect(block).toContain('finally');
expect(block).toContain('originalActive');
expect(block).toContain('switchTab(originalActive');
});
test('uses bringToFront: false so the OS window does NOT jump', () => {
const block = META_SRC.slice(META_SRC.indexOf("case 'tab-each':"), META_SRC.indexOf("case 'tab-each':") + 4000);
// tab-each is a background operation — pulling focus would steal the
// user's foreground app every time claude fans out, which is
// unacceptable.
expect(block).toContain('bringToFront: false');
});
test('skips chrome:// and chrome-extension:// internal pages', () => {
const block = META_SRC.slice(META_SRC.indexOf("case 'tab-each':"), META_SRC.indexOf("case 'tab-each':") + 4000);
expect(block).toContain("startsWith('chrome://')");
expect(block).toContain("startsWith('chrome-extension://')");
});
});
describe('tab-each: behavior', () => {
function mockBm(tabs: Array<{ id: number; url: string; title: string; active: boolean }>) {
let activeId = tabs.find(t => t.active)?.id ?? tabs[0]?.id ?? 0;
const switched: number[] = [];
return {
__switched: switched,
__activeId: () => activeId,
getActiveSession: () => ({}),
getActiveTabId: () => activeId,
getTabListWithTitles: async () => tabs.map(t => ({ ...t })),
switchTab: (id: number, _opts?: any) => { switched.push(id); activeId = id; },
} as any;
}
test('iterates every tab, calls executeCommand for each, returns JSON results', async () => {
const tabs = [
{ id: 1, url: 'https://news.example.com', title: 'News', active: true },
{ id: 2, url: 'https://docs.example.com', title: 'Docs', active: false },
{ id: 3, url: 'https://github.com', title: 'GitHub', active: false },
];
const bm = mockBm(tabs);
const calls: Array<{ command: string; args?: string[]; tabId?: number }> = [];
const out = await handleMetaCommand(
'tab-each',
['snapshot', '-i'],
bm,
async () => {},
null,
{
executeCommand: async (body) => {
calls.push(body);
return { status: 200, result: `snap-of-${body.tabId}` };
},
},
);
const parsed = JSON.parse(out);
expect(parsed.command).toBe('snapshot');
expect(parsed.args).toEqual(['-i']);
expect(parsed.total).toBe(3);
expect(parsed.results.map((r: any) => r.tabId)).toEqual([1, 2, 3]);
expect(parsed.results.every((r: any) => r.status === 200)).toBe(true);
expect(parsed.results[0].output).toBe('snap-of-1');
// Inner command was dispatched 3 times, once per tab, with the right tabId.
expect(calls).toHaveLength(3);
expect(calls.map(c => c.tabId)).toEqual([1, 2, 3]);
expect(calls.every(c => c.command === 'snapshot')).toBe(true);
});
test('skips chrome:// pages with status=0 + "skipped" output', async () => {
const tabs = [
{ id: 1, url: 'chrome://newtab', title: 'New Tab', active: true },
{ id: 2, url: 'https://example.com', title: 'Example', active: false },
{ id: 3, url: 'chrome-extension://abc/page.html', title: 'Ext', active: false },
];
const bm = mockBm(tabs);
const calls: any[] = [];
const out = await handleMetaCommand(
'tab-each',
['text'],
bm,
async () => {},
null,
{
executeCommand: async (body) => {
calls.push(body);
return { status: 200, result: `text-of-${body.tabId}` };
},
},
);
const parsed = JSON.parse(out);
expect(parsed.total).toBe(3);
// chrome:// and chrome-extension:// → skipped (status 0).
expect(parsed.results[0].status).toBe(0);
expect(parsed.results[0].output).toContain('skipped');
expect(parsed.results[2].status).toBe(0);
// Only the real tab dispatched.
expect(calls).toHaveLength(1);
expect(calls[0].tabId).toBe(2);
});
test('restores the originally active tab even if a tab errors', async () => {
const tabs = [
{ id: 10, url: 'https://a.example', title: 'A', active: false },
{ id: 20, url: 'https://b.example', title: 'B', active: true }, // initially active
{ id: 30, url: 'https://c.example', title: 'C', active: false },
];
const bm = mockBm(tabs);
let calls = 0;
const out = await handleMetaCommand(
'tab-each',
['text'],
bm,
async () => {},
null,
{
executeCommand: async (body) => {
calls++;
if (body.tabId === 20) {
return { status: 500, result: JSON.stringify({ error: 'boom' }) };
}
return { status: 200, result: `ok-${body.tabId}` };
},
},
);
const parsed = JSON.parse(out);
expect(parsed.results.find((r: any) => r.tabId === 20).status).toBe(500);
expect(parsed.results.find((r: any) => r.tabId === 20).output).toBe('boom');
expect(parsed.results.find((r: any) => r.tabId === 10).status).toBe(200);
expect(parsed.results.find((r: any) => r.tabId === 30).status).toBe(200);
// Active tab restored to 20 (the one that was active when we started).
expect(bm.__activeId()).toBe(20);
});
test('throws on empty args (no inner command)', async () => {
const bm = mockBm([{ id: 1, url: 'https://x.example', title: 'X', active: true }]);
await expect(handleMetaCommand(
'tab-each',
[],
bm,
async () => {},
null,
{ executeCommand: async () => ({ status: 200, result: '' }) },
)).rejects.toThrow(/Usage/);
});
});
@@ -0,0 +1,273 @@
/**
* Integration tests for terminal-agent.ts.
*
* Spawns the agent as a real subprocess in a temp state directory,
* exercises:
* 1. /internal/grant loopback handshake with the internal token.
* 2. /ws Origin gate non-extension Origin 403.
* 3. /ws cookie gate missing/invalid cookie 401.
* 4. /ws full PTY round-trip write `echo hi\n`, read `hi`.
* 5. resize control message terminal accepts and stays alive.
* 6. close behavior sending close terminates the PTY child.
*
* Uses /bin/bash via BROWSE_TERMINAL_BINARY override so CI doesn't need
* the `claude` binary installed.
*/
import { describe, test, expect, beforeAll, afterAll } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
import * as os from 'os';
const AGENT_SCRIPT = path.join(import.meta.dir, '../src/terminal-agent.ts');
const BASH = '/bin/bash';
let stateDir: string;
let agentProc: any;
let agentPort: number;
let internalToken: string;
function readPortFile(): number {
for (let i = 0; i < 50; i++) {
try {
const v = parseInt(fs.readFileSync(path.join(stateDir, 'terminal-port'), 'utf-8').trim(), 10);
if (Number.isFinite(v) && v > 0) return v;
} catch {}
Bun.sleepSync(40);
}
throw new Error('terminal-agent never wrote port file');
}
function readTokenFile(): string {
for (let i = 0; i < 50; i++) {
try {
const t = fs.readFileSync(path.join(stateDir, 'terminal-internal-token'), 'utf-8').trim();
if (t.length > 16) return t;
} catch {}
Bun.sleepSync(40);
}
throw new Error('terminal-agent never wrote internal token');
}
beforeAll(() => {
stateDir = fs.mkdtempSync(path.join(os.tmpdir(), 'gstack-term-'));
const stateFile = path.join(stateDir, 'browse.json');
// browse.json must exist so the agent's readBrowseToken doesn't throw.
fs.writeFileSync(stateFile, JSON.stringify({ token: 'test-browse-token' }));
agentProc = Bun.spawn(['bun', 'run', AGENT_SCRIPT], {
env: {
...process.env,
BROWSE_STATE_FILE: stateFile,
BROWSE_SERVER_PORT: '0', // not used in this test
BROWSE_TERMINAL_BINARY: BASH,
},
stdio: ['ignore', 'pipe', 'pipe'],
});
agentPort = readPortFile();
internalToken = readTokenFile();
});
afterAll(() => {
try { agentProc?.kill?.(); } catch {}
try { fs.rmSync(stateDir, { recursive: true, force: true }); } catch {}
});
async function grantToken(token: string): Promise<Response> {
return fetch(`http://127.0.0.1:${agentPort}/internal/grant`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'Authorization': `Bearer ${internalToken}`,
},
body: JSON.stringify({ token }),
});
}
describe('terminal-agent: /internal/grant', () => {
test('accepts grants signed with the internal token', async () => {
const resp = await grantToken('test-cookie-token-very-long-yes');
expect(resp.status).toBe(200);
});
test('rejects grants with the wrong internal token', async () => {
const resp = await fetch(`http://127.0.0.1:${agentPort}/internal/grant`, {
method: 'POST',
headers: {
'Content-Type': 'application/json',
'Authorization': 'Bearer wrong-token',
},
body: JSON.stringify({ token: 'whatever' }),
});
expect(resp.status).toBe(403);
});
});
describe('terminal-agent: /ws gates', () => {
test('rejects upgrade attempts without an extension Origin', async () => {
const resp = await fetch(`http://127.0.0.1:${agentPort}/ws`);
expect(resp.status).toBe(403);
expect(await resp.text()).toBe('forbidden origin');
});
test('rejects upgrade attempts from a non-extension Origin', async () => {
const resp = await fetch(`http://127.0.0.1:${agentPort}/ws`, {
headers: { 'Origin': 'https://evil.example.com' },
});
expect(resp.status).toBe(403);
});
test('rejects extension-Origin upgrades without a granted cookie', async () => {
const resp = await fetch(`http://127.0.0.1:${agentPort}/ws`, {
headers: {
'Origin': 'chrome-extension://abc123',
'Cookie': 'gstack_pty=never-granted',
},
});
expect(resp.status).toBe(401);
});
});
describe('terminal-agent: PTY round-trip via real WebSocket (Cookie auth)', () => {
test('binary writes go to PTY stdin, output streams back', async () => {
const cookie = 'rt-token-must-be-at-least-seventeen-chars-long';
const granted = await grantToken(cookie);
expect(granted.status).toBe(200);
const ws = new WebSocket(`ws://127.0.0.1:${agentPort}/ws`, {
headers: {
'Origin': 'chrome-extension://test-extension-id',
'Cookie': `gstack_pty=${cookie}`,
},
} as any);
const collected: string[] = [];
let opened = false;
let closed = false;
await new Promise<void>((resolve, reject) => {
const timer = setTimeout(() => reject(new Error('ws never opened')), 5000);
ws.addEventListener('open', () => { opened = true; clearTimeout(timer); resolve(); });
ws.addEventListener('error', (e: any) => { clearTimeout(timer); reject(new Error('ws error')); });
});
ws.addEventListener('message', (ev: any) => {
if (typeof ev.data === 'string') return; // ignore control frames
const buf = ev.data instanceof ArrayBuffer ? new Uint8Array(ev.data) : ev.data;
collected.push(new TextDecoder().decode(buf));
});
ws.addEventListener('close', () => { closed = true; });
// Lazy-spawn trigger: any binary frame causes the agent to spawn /bin/bash.
ws.send(new TextEncoder().encode('echo hello-pty-world\nexit\n'));
// Wait up to 5s for output and shutdown.
await new Promise<void>((resolve) => {
const start = Date.now();
const tick = () => {
const joined = collected.join('');
if (joined.includes('hello-pty-world')) return resolve();
if (Date.now() - start > 5000) return resolve();
setTimeout(tick, 50);
};
tick();
});
expect(opened).toBe(true);
const allOutput = collected.join('');
expect(allOutput).toContain('hello-pty-world');
try { ws.close(); } catch {}
// Give cleanup a moment.
await Bun.sleep(200);
});
test('Sec-WebSocket-Protocol auth path: browser-style upgrade with token in protocol', async () => {
// This is the path the actual browser extension takes. Cross-port
// SameSite=Strict cookies don't reliably survive the jump from the
// browse server (port A) to the agent (port B) when initiated from a
// chrome-extension origin, so we send the token via the only auth
// header the browser WebSocket API lets us set: Sec-WebSocket-Protocol.
//
// The browser sends `gstack-pty.<token>` and the agent must:
// 1) strip the gstack-pty. prefix
// 2) validate the token
// 3) ECHO the protocol back in the upgrade response
// Without (3) the browser closes the connection immediately, which
// is the exact bug the original cookie-only implementation hit in
// manual dogfood. This test catches that regression in CI.
const token = 'sec-protocol-token-must-be-at-least-seventeen-chars';
await grantToken(token);
// We exercise the protocol path by raw-handshaking via fetch+Upgrade,
// because Bun's test-client WebSocket constructor doesn't propagate
// `protocols` cleanly when also passed `headers` (the constructor
// detects the third-arg form unreliably). Real browsers (Chromium)
// use the standard protocols arg fine — the server-side handler is
// identical either way, so this test still locks the load-bearing
// invariant: the agent accepts a token via Sec-WebSocket-Protocol
// and echoes the protocol back so a browser would accept the upgrade.
const handshakeKey = 'dGhlIHNhbXBsZSBub25jZQ==';
const resp = await fetch(`http://127.0.0.1:${agentPort}/ws`, {
headers: {
'Connection': 'Upgrade',
'Upgrade': 'websocket',
'Sec-WebSocket-Version': '13',
'Sec-WebSocket-Key': handshakeKey,
'Sec-WebSocket-Protocol': `gstack-pty.${token}`,
'Origin': 'chrome-extension://test-extension-id',
},
});
// 101 Switching Protocols + protocol echoed back = browser would accept.
// 401/403/anything else = browser would close the connection immediately
// (the bug we hit in manual dogfood).
expect(resp.status).toBe(101);
expect(resp.headers.get('upgrade')?.toLowerCase()).toBe('websocket');
expect(resp.headers.get('sec-websocket-protocol')).toBe(`gstack-pty.${token}`);
});
test('Sec-WebSocket-Protocol auth: rejects unknown token even with valid Origin', async () => {
const resp = await fetch(`http://127.0.0.1:${agentPort}/ws`, {
headers: {
'Connection': 'Upgrade',
'Upgrade': 'websocket',
'Sec-WebSocket-Version': '13',
'Sec-WebSocket-Key': 'dGhlIHNhbXBsZSBub25jZQ==',
'Sec-WebSocket-Protocol': 'gstack-pty.never-granted-token',
'Origin': 'chrome-extension://test-extension-id',
},
});
expect(resp.status).toBe(401);
});
test('text frame {type:"resize"} is accepted (no crash, ws stays open)', async () => {
const cookie = 'resize-token-must-be-at-least-seventeen-chars';
await grantToken(cookie);
const ws = new WebSocket(`ws://127.0.0.1:${agentPort}/ws`, {
headers: {
'Origin': 'chrome-extension://test-extension-id',
'Cookie': `gstack_pty=${cookie}`,
},
} as any);
await new Promise<void>((resolve, reject) => {
const timer = setTimeout(() => reject(new Error('ws never opened')), 5000);
ws.addEventListener('open', () => { clearTimeout(timer); resolve(); });
ws.addEventListener('error', () => { clearTimeout(timer); reject(new Error('ws error')); });
});
// Send a resize before anything else (lazy-spawn won't fire).
ws.send(JSON.stringify({ type: 'resize', cols: 120, rows: 40 }));
// After resize, send a binary frame; should still work.
ws.send(new TextEncoder().encode('exit\n'));
await Bun.sleep(300);
// ws still readyState 1 (OPEN) or 3 (CLOSED after exit) — both fine.
expect([WebSocket.OPEN, WebSocket.CLOSED]).toContain(ws.readyState);
try { ws.close(); } catch {}
});
});
+223
View File
@@ -0,0 +1,223 @@
/**
* Unit tests for the Terminal-tab PTY agent and its server-side glue.
*
* Coverage:
* - pty-session-cookie module: mint / validate / revoke / TTL pruning.
* - source-level guard: /pty-session and /terminal/* are NOT in TUNNEL_PATHS.
* - source-level guard: /health does not surface ptyToken.
* - source-level guard: terminal-agent binds 127.0.0.1 only.
* - source-level guard: terminal-agent enforces Origin AND cookie on /ws.
*
* These are read-only checks against source they prevent silent surface
* widening during a routine refactor (matches the dual-listener.test.ts
* pattern). End-to-end behavior (real /bin/bash PTY round-trip,
* tunnel-surface 404 + denial-log) lives in
* `browse/test/terminal-agent-integration.test.ts`.
*/
import { describe, test, expect, beforeEach } from 'bun:test';
import * as fs from 'fs';
import * as path from 'path';
import {
mintPtySessionToken, validatePtySessionToken, revokePtySessionToken,
extractPtyCookie, buildPtySetCookie, buildPtyClearCookie,
PTY_COOKIE_NAME, __resetPtySessions,
} from '../src/pty-session-cookie';
const SERVER_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/server.ts'), 'utf-8');
const AGENT_SRC = fs.readFileSync(path.join(import.meta.dir, '../src/terminal-agent.ts'), 'utf-8');
describe('pty-session-cookie: mint/validate/revoke', () => {
beforeEach(() => __resetPtySessions());
test('a freshly minted token validates', () => {
const { token } = mintPtySessionToken();
expect(validatePtySessionToken(token)).toBe(true);
});
test('null and unknown tokens fail validation', () => {
expect(validatePtySessionToken(null)).toBe(false);
expect(validatePtySessionToken(undefined)).toBe(false);
expect(validatePtySessionToken('')).toBe(false);
expect(validatePtySessionToken('not-a-real-token')).toBe(false);
});
test('revoke makes a token invalid', () => {
const { token } = mintPtySessionToken();
expect(validatePtySessionToken(token)).toBe(true);
revokePtySessionToken(token);
expect(validatePtySessionToken(token)).toBe(false);
});
test('Set-Cookie has HttpOnly + SameSite=Strict + Path=/ + Max-Age', () => {
const { token } = mintPtySessionToken();
const cookie = buildPtySetCookie(token);
expect(cookie).toContain(`${PTY_COOKIE_NAME}=${token}`);
expect(cookie).toContain('HttpOnly');
expect(cookie).toContain('SameSite=Strict');
expect(cookie).toContain('Path=/');
expect(cookie).toMatch(/Max-Age=\d+/);
// Secure is intentionally omitted — daemon binds 127.0.0.1 over HTTP.
expect(cookie).not.toContain('Secure');
});
test('clear-cookie has Max-Age=0', () => {
expect(buildPtyClearCookie()).toContain('Max-Age=0');
});
test('extractPtyCookie reads gstack_pty from a Cookie header', () => {
const { token } = mintPtySessionToken();
const req = new Request('http://127.0.0.1/ws', {
headers: { 'cookie': `othercookie=foo; gstack_pty=${token}; baz=qux` },
});
expect(extractPtyCookie(req)).toBe(token);
});
test('extractPtyCookie returns null when the cookie is missing', () => {
const req = new Request('http://127.0.0.1/ws', {
headers: { 'cookie': 'unrelated=value' },
});
expect(extractPtyCookie(req)).toBe(null);
});
});
describe('Source-level guard: /pty-session is not on the tunnel surface', () => {
test('TUNNEL_PATHS does not include /pty-session or /terminal/*', () => {
const start = SERVER_SRC.indexOf('const TUNNEL_PATHS = new Set<string>([');
expect(start).toBeGreaterThan(-1);
const end = SERVER_SRC.indexOf(']);', start);
const body = SERVER_SRC.slice(start, end);
expect(body).not.toContain('/pty-session');
expect(body).not.toContain('/terminal/');
expect(body).not.toContain('/terminal-');
});
});
describe('Source-level guard: /health does NOT surface ptyToken', () => {
test('/health response body does not include ptyToken', () => {
const healthIdx = SERVER_SRC.indexOf("url.pathname === '/health'");
expect(healthIdx).toBeGreaterThan(-1);
// Slice from /health through the response close-bracket.
const slice = SERVER_SRC.slice(healthIdx, healthIdx + 2000);
// The /health JSON.stringify body must not mention the cookie token.
// It's allowed to include `terminalPort` (a port number, not auth).
expect(slice).not.toContain('ptyToken');
expect(slice).not.toContain('gstack_pty');
expect(slice).toContain('terminalPort');
});
});
describe('Source-level guard: terminal-agent', () => {
test('binds 127.0.0.1 only, never 0.0.0.0', () => {
expect(AGENT_SRC).toContain("hostname: '127.0.0.1'");
expect(AGENT_SRC).not.toContain("hostname: '0.0.0.0'");
});
test('rejects /ws upgrades without chrome-extension:// Origin', () => {
// The Origin check must run BEFORE the cookie check — otherwise a
// missing-origin attempt would surface the 401 cookie message and
// signal to attackers that they need to forge a cookie.
const wsHandler = AGENT_SRC.slice(AGENT_SRC.indexOf("if (url.pathname === '/ws')"));
expect(wsHandler).toContain('chrome-extension://');
expect(wsHandler).toContain('forbidden origin');
});
test('validates the session token against an in-memory token set', () => {
const wsHandler = AGENT_SRC.slice(AGENT_SRC.indexOf("if (url.pathname === '/ws')"));
// Two transports: Sec-WebSocket-Protocol (preferred for browsers) and
// Cookie gstack_pty (fallback). Both verify against validTokens.
expect(wsHandler).toContain('sec-websocket-protocol');
expect(wsHandler).toContain('gstack_pty');
expect(wsHandler).toContain('validTokens.has');
});
test('Sec-WebSocket-Protocol auth: strips gstack-pty. prefix and echoes back', () => {
const wsHandler = AGENT_SRC.slice(AGENT_SRC.indexOf("if (url.pathname === '/ws')"));
// Browsers send `Sec-WebSocket-Protocol: gstack-pty.<token>`. The agent
// must strip the prefix before checking validTokens, AND echo the
// protocol back in the upgrade response — without the echo, the
// browser closes the connection immediately.
expect(wsHandler).toContain("'gstack-pty.'");
expect(wsHandler).toContain('Sec-WebSocket-Protocol');
expect(wsHandler).toContain('acceptedProtocol');
});
test('lazy spawn: claude PTY is spawned in message handler, not on upgrade', () => {
// The whole point of lazy-spawn (codex finding #8) is that the WS
// upgrade itself does NOT call spawnClaude. Spawn happens on first
// message frame.
const upgradeBlock = AGENT_SRC.slice(
AGENT_SRC.indexOf("if (url.pathname === '/ws')"),
AGENT_SRC.indexOf("websocket: {"),
);
expect(upgradeBlock).not.toContain('spawnClaude(');
// Spawn must be invoked from the message handler (lazy on first byte).
const messageHandler = AGENT_SRC.slice(AGENT_SRC.indexOf('message(ws, raw)'));
expect(messageHandler).toContain('spawnClaude(');
expect(messageHandler).toContain('!session.spawned');
});
test('process.on uncaughtException + unhandledRejection handlers exist', () => {
expect(AGENT_SRC).toContain("process.on('uncaughtException'");
expect(AGENT_SRC).toContain("process.on('unhandledRejection'");
});
test('cleanup escalates SIGINT to SIGKILL after 3s on close', () => {
// disposeSession must be idempotent and use a SIGINT-then-SIGKILL pattern.
const dispose = AGENT_SRC.slice(AGENT_SRC.indexOf('function disposeSession'));
expect(dispose).toContain("'SIGINT'");
expect(dispose).toContain("'SIGKILL'");
expect(dispose).toContain('3000');
});
test('tabState frames write tabs.json + active-tab.json', () => {
expect(AGENT_SRC).toContain("msg?.type === 'tabState'");
expect(AGENT_SRC).toContain('function handleTabState');
const fn = AGENT_SRC.slice(AGENT_SRC.indexOf('function handleTabState'));
// Atomic write via tmp + rename for both files (so claude never reads
// a half-written JSON document).
expect(fn).toContain("'tabs.json'");
expect(fn).toContain("'active-tab.json'");
expect(fn).toContain('renameSync');
// Skip chrome:// and chrome-extension:// pages — they're not useful
// targets for browse commands.
expect(fn).toContain("startsWith('chrome://')");
expect(fn).toContain("startsWith('chrome-extension://')");
});
test('claude is spawned with --append-system-prompt tab-awareness hint', () => {
expect(AGENT_SRC).toContain('function buildTabAwarenessHint');
const hint = AGENT_SRC.slice(AGENT_SRC.indexOf('function buildTabAwarenessHint'));
// The hint must mention the live state files and the fanout command —
// those are the two affordances that distinguish a gstack-PTY claude
// from a plain `claude` session.
expect(hint).toContain('tabs.json');
expect(hint).toContain('active-tab.json');
expect(hint).toContain('tab-each');
// And it must be passed via --append-system-prompt at spawn time
// (NOT written into the PTY as user input — that would pollute the
// visible transcript).
const spawn = AGENT_SRC.slice(AGENT_SRC.indexOf('function spawnClaude'));
expect(spawn).toContain("'--append-system-prompt'");
expect(spawn).toContain('tabHint');
});
});
describe('Source-level guard: server.ts /pty-session route', () => {
test('validates AUTH_TOKEN, grants over loopback, returns token + Set-Cookie', () => {
const route = SERVER_SRC.slice(SERVER_SRC.indexOf("url.pathname === '/pty-session'"));
// Must check auth before minting.
const beforeMint = route.slice(0, route.indexOf('mintPtySessionToken'));
expect(beforeMint).toContain('validateAuth');
// Must call the loopback grant before responding (otherwise the
// agent's validTokens Set never sees the token and /ws would 401).
expect(route).toContain('grantPtyToken');
// Must return the token in the JSON body for the
// Sec-WebSocket-Protocol auth path (cross-port cookies don't survive
// SameSite=Strict from a chrome-extension origin).
expect(route).toContain('ptySessionToken');
// Set-Cookie is kept as a fallback for non-browser callers.
expect(route).toContain('Set-Cookie');
expect(route).toContain('buildPtySetCookie');
});
});
+6
View File
@@ -15,6 +15,8 @@
"devDependencies": {
"@anthropic-ai/claude-agent-sdk": "0.2.117",
"@anthropic-ai/sdk": "^0.78.0",
"xterm": "5",
"xterm-addon-fit": "^0.8.0",
},
},
},
@@ -537,6 +539,10 @@
"ws": ["ws@8.20.0", "", { "peerDependencies": { "bufferutil": "^4.0.1", "utf-8-validate": ">=5.0.2" }, "optionalPeers": ["bufferutil", "utf-8-validate"] }, "sha512-sAt8BhgNbzCtgGbt2OxmpuryO63ZoDk/sqaB/znQm94T4fCEsy/yV+7CdC1kJhOU9lboAEU7R3kquuycDoibVA=="],
"xterm": ["xterm@5.3.0", "", {}, "sha512-8QqjlekLUFTrU6x7xck1MsPzPA571K5zNqWm0M0oroYEWVOptZ0+ubQSkQ3uxIEhcIHRujJy6emDWX4A7qyFzg=="],
"xterm-addon-fit": ["xterm-addon-fit@0.8.0", "", { "peerDependencies": { "xterm": "^5.0.0" } }, "sha512-yj3Np7XlvxxhYF/EJ7p3KHaMt6OdwQ+HDu573Vx1lRXsVxOcnVJs51RgjZOouIZOczTsskaS+CpXspK81/DLqw=="],
"y18n": ["y18n@5.0.8", "", {}, "sha512-0pfFzegeDWJHJIAmTLRP2DwHjdF5s7jo9tuztdQxAhINCdvS+3nGINqPd00AphqJR/0LhANUS6/+7SCb98YOfA=="],
"yargs": ["yargs@17.7.2", "", { "dependencies": { "cliui": "^8.0.1", "escalade": "^3.1.1", "get-caller-file": "^2.0.5", "require-directory": "^2.1.1", "string-width": "^4.2.3", "y18n": "^5.0.5", "yargs-parser": "^21.1.1" } }, "sha512-7dSzzRQ++CKnNI/krKnYRV7JKKPUXMEh61soaHKg9mrWEhzFWhFnxPxGl+69cD1Ou63C13NUPCnmIcrvqCuM6w=="],
+155 -145
View File
@@ -1,190 +1,200 @@
# Sidebar Message Flow
# Sidebar Flow
How the GStack Browser sidebar actually works. Read this before touching
sidepanel.js, background.js, content.js, server.ts sidebar endpoints,
or sidebar-agent.ts.
`sidepanel.js`, `background.js`, `content.js`, `terminal-agent.ts`, or
sidebar-related server endpoints.
The sidebar has one primary surface — the **Terminal** pane, an interactive
`claude` PTY. Activity / Refs / Inspector survive as debug overlays behind
the `debug` toggle in the footer. The chat queue path (one-shot `claude -p`,
sidebar-agent.ts) was ripped once the PTY proved out — the Terminal pane is
strictly more capable.
## Components
```
┌─────────────────┐ ┌──────────────┐ ┌─────────────┐ ┌────────────────┐
│ sidepanel.js │────▶│ background.js│────▶│ server.ts │────▶│sidebar-agent.ts│
(Chrome panel) │ │ (svc worker) │ │ (Bun HTTP) │ │ (Bun process)
└─────────────────┘ └──────────────┘ └─────────────┘ └────────────────┘
▲ │ │
polls /sidebar-chat polls queue file
└───────────────────────────────────────────┘ │
──────────────────────
POST /sidebar-agent/event
┌─────────────────┐ ┌──────────────┐ ┌──────────────────┐
│ sidepanel.js + │────▶│ server.ts │────▶│terminal-agent.ts
-terminal.js │ │ (compiled) │ │ (non-compiled)
│ (xterm.js) │ │ │ │ PTY listener │
└─────────────────┘ └──────────────┘ └──────────────────┘
│ ws://127.0.0.1:<termPort>/ws (Sec-WebSocket-Protocol auth)
└───────────────────────┼──────────────────────▶│ Bun.spawn(claude)
│ │ terminal: {data}
│ ▼
│ ┌──────────────────┐
│ │ claude PTY │
│ └──────────────────┘
POST /pty-session │
(Bearer AUTH_TOKEN) │
┌──────────────────┐
│ pty-session- │
│ cookie.ts │
│ (in-memory token │
│ registry) │
└──────────────────┘
│ POST /internal/grant (loopback)
┌──────────────────┐
│ validTokens Set │
│ in agent memory │
└──────────────────┘
```
## Startup Timeline
The compiled browse server can't `posix_spawn` external executables —
`terminal-agent.ts` runs as a separate non-compiled `bun run` process and
owns the `claude` subprocess.
## Startup + first-keystroke timeline
```
T+0ms CLI runs `$B connect`
├── Server starts on port 34567
── Writes state to .gstack/browse.json (pid, port, token)
├── Launches headed Chromium with extension
└── Clears sidebar-agent-queue.jsonl
├── Server starts (compiled)
── Spawns terminal-agent.ts via `bun run`
T+500ms sidebar-agent.ts spawned by CLI
├── Reads auth token from .gstack/browse.json
├── Creates queue file if missing
├── Sets lastLine = current line count
└── Starts polling every 200ms
T+500ms terminal-agent.ts boots
├── Bun.serve on 127.0.0.1:0 (random port)
├── Writes <stateDir>/terminal-port (server reads it for /health)
├── Writes <stateDir>/terminal-internal-token (loopback handshake)
└── Probes claude → writes claude-available.json
T+1-3s Extension loads in Chromium
├── background.js: health poll every 1s (fast startup)
│ └── GET /health → gets auth token
├── content.js: injects on welcome page
│ └── Does NOT fire gstack-extension-ready (waits for sidebar)
└── Side panel: may auto-open via chrome.sidePanel.open()
T+1-3s Extension loads, sidebar opens
├── sidepanel-terminal.js: setState(IDLE), shows "Starting Claude Code..."
└── tryAutoConnect() polls until window.gstackServerPort + token are set
T+2-10s Side panel connects
├── tryConnect() → asks background for port/token
├── Fallback: direct GET /health for token
├── updateConnection(url, token)
│ ├── Starts chat polling (1s interval)
│ ├── Starts tab polling (2s interval)
├── Connects SSE activity stream
└── Sends { type: 'sidebarOpened' } to background
└── background relays to content script → hides welcome arrow
T+10s+ Ready for messages
T+ready tryAutoConnect calls connect()
├── POST /pty-session (Authorization: Bearer AUTH_TOKEN)
│ └── server mints session token, posts /internal/grant to agent
│ └── responds with {terminalPort, ptySessionToken}
├── GET /claude-available (preflight)
├── new WebSocket(`ws://127.0.0.1:<terminalPort>/ws`,
[`gstack-pty.<token>`])
│ └── Browser sends Sec-WebSocket-Protocol + Origin
│ └── Agent validates Origin AND token BEFORE upgrading
│ └── Agent echoes the protocol back (REQUIRED — browser
│ closes the connection without it)
├── On open: send {type:"resize"} then a single \n byte
└── Agent message handler sees the byte → spawnClaude()
```
## Message Flow: User Types → Claude Responds
## Auth: WebSocket can't send Authorization headers
```
1. User types "go to hn" in sidebar, hits Enter
Browser WebSocket clients can't set `Authorization`. They CAN set
`Sec-WebSocket-Protocol` via the second arg of `new WebSocket(url,
protocols)`. We exploit that:
2. sidepanel.js sendMessage()
├── Renders user bubble immediately (optimistic)
├── Renders thinking dots immediately
├── Switches to fast poll (300ms)
└── chrome.runtime.sendMessage({ type: 'sidebar-command', message, tabId })
1. `POST /pty-session` (auth: Bearer AUTH_TOKEN) → server mints a
short-lived session token, pushes it to the agent over loopback,
returns it in the JSON body.
2. Extension calls `new WebSocket(url, ['gstack-pty.<token>'])`.
3. Agent reads `Sec-WebSocket-Protocol`, strips `gstack-pty.`, validates
against `validTokens`, echoes the protocol back. Echo is mandatory —
without it Chromium closes the connection on receipt of the upgrade
response.
3. background.js
├── Gets active Chrome tab URL
└── POST /sidebar-command { message, activeTabUrl }
with Authorization: Bearer ${authToken}
A `Set-Cookie: gstack_pty=...` header is also returned for non-browser
callers (curl, integration tests). The cookie path was the original v1
design but `SameSite=Strict` cookies don't survive the cross-port jump
from server.ts:34567 → agent:<random> from a chrome-extension origin.
The protocol-token path is what the browser actually uses.
4. server.ts /sidebar-command handler
├── validateAuth(req)
├── syncActiveTabByUrl(extensionUrl) — syncs Playwright tab to Chrome tab
├── pickSidebarModel(message) — 'sonnet' for actions, 'opus' for analysis
├── Adds user message to chat buffer
├── Builds system prompt + args
└── Appends JSON to ~/.gstack/sidebar-agent-queue.jsonl
### Dual-token model
5. sidebar-agent.ts poll() (within 200ms)
├── Reads new line from queue file
├── Parses JSON entry
├── Checks processingTabs — skips if tab already has agent running
└── askClaude(entry) — fire and forget
| Token | Lives in | Used for | Lifetime |
|-------|----------|----------|----------|
| `AUTH_TOKEN` | `<stateDir>/browse.json`; in-memory in server.ts | `/pty-session` POST (mint cookie + token) | server lifetime |
| `gstack-pty.<...>` (Sec-WebSocket-Protocol) | Browser memory only; agent `validTokens` Set | `/ws` upgrade auth | 30 min, auto-revoked on WS close |
| `INTERNAL_TOKEN` | `<stateDir>/terminal-internal-token`; in agent memory | server → agent loopback `/internal/grant` | agent lifetime |
6. sidebar-agent.ts askClaude()
├── spawn('claude', ['-p', prompt, '--model', model, ...])
├── Streams stdout line-by-line (stream-json format)
├── For each event: POST /sidebar-agent/event { type, tool, text, tabId }
└── On close: POST /sidebar-agent/event { type: 'agent_done' }
`AUTH_TOKEN` is **never** valid for `/ws` directly. The session token is
**never** valid for `/pty-session` or `/command`. Strict separation
prevents an SSE or page-content token leak from escalating into shell
access.
7. server.ts processAgentEvent()
├── Adds entry to chat buffer (in-memory + disk)
├── On agent_done: sets tab status to 'idle'
└── On agent_done: processes next queued message for that tab
## Threat model
8. sidepanel.js pollChat() (every 300ms during fast poll)
├── GET /sidebar-chat?after=${chatLineCount}&tabId=${tabId}
├── Renders new entries (text, tool_use, agent_done)
└── On agent idle: removes thinking dots, stops fast poll
```
The Terminal pane **bypasses the prompt-injection security stack** on
purpose — the user is typing directly to claude, there's no untrusted
page content in the loop. Trust source is the keyboard, same as any
local terminal.
## Arrow Hint Hide Flow (4-step signal chain)
That trust assumption is load-bearing on three transport guarantees:
The welcome page shows a right-pointing arrow until the sidebar opens.
1. **Local-only listener.** terminal-agent.ts binds `127.0.0.1` only.
The dual-listener tunnel surface (server.ts `TUNNEL_PATHS`) does
not include `/pty-session` or `/terminal/*`, so the tunnel returns
404 by default-deny.
2. **Origin gate.** `/ws` upgrades require
`Origin: chrome-extension://<id>`. A localhost web page can't mount
a cross-site WebSocket hijack against the shell because its Origin
is a regular `http(s)://...`.
3. **Session token auth.** Minted only by an authenticated
`/pty-session` POST, scoped to one WS, auto-revoked on close.
```
1. sidepanel.js updateConnection()
└── chrome.runtime.sendMessage({ type: 'sidebarOpened' })
Drop any one of those three and the whole tab becomes unsafe.
2. background.js
└── chrome.tabs.sendMessage(activeTabId, { type: 'sidebarOpened' })
## Lifecycle
3. content.js onMessage handler
└── document.dispatchEvent(new CustomEvent('gstack-extension-ready'))
- **Eager auto-connect.** Sidebar opens → tryAutoConnect polls for the
bootstrap globals and connects as soon as they're set. No keypress
required.
- **One PTY per WS.** Closing the WebSocket SIGINTs claude, then SIGKILLs
after 3s. The session token is revoked so a stolen token can't be
replayed.
- **No auto-reconnect on close.** The user sees "Session ended, click to
start a new session." Auto-reconnect would burn a fresh claude session
on every reload. v1.1 may add session resumption keyed on tab/session
id (see TODOS).
- **Manual restart anytime.** A `↻ Restart` button lives in the always-
visible terminal toolbar — works mid-session, not just from the ENDED
state.
4. welcome.html script
└── addEventListener('gstack-extension-ready', () => arrow.classList.add('hidden'))
```
## Quick-action toolbar
The arrow does NOT hide when the extension loads. Only when the sidebar connects.
Three browser-action buttons live next to the Restart button at the top
of the Terminal pane:
## Auth Token Flow
| Button | Behavior |
|--------|----------|
| 🧹 Cleanup | `window.gstackInjectToTerminal(prompt)` — pipes a "remove ads/banners" instruction into the live PTY. claude in the terminal sees it and acts. |
| 📸 Screenshot | `POST /command screenshot` — direct browse-server call, no PTY involvement. |
| 🍪 Cookies | Navigates to the `/cookie-picker` page. |
```
Server starts → AUTH_TOKEN = crypto.randomUUID()
├── GET /health (no auth) → returns { token: AUTH_TOKEN }
├── background.js checkHealth() → authToken = data.token
│ └── Refreshes on EVERY health poll (fixes stale token on restart)
├── sidepanel.js tryConnect() → serverToken from background or /health
│ └── Used for chat polling: Authorization: Bearer ${serverToken}
└── sidebar-agent.ts refreshToken() → reads from .gstack/browse.json
└── Used for event relay: Authorization: Bearer ${authToken}
```
The Inspector's "Send to Code" button uses the same `gstackInjectToTerminal`
path to forward CSS inspector data into claude.
If the server restarts, all three components get fresh tokens within 10s
(background health poll interval).
## Debug surfaces (Activity / Refs / Inspector)
## Model Routing
Behind the `debug` toggle in the footer. SSE-driven, independent of the
Terminal pane:
`pickSidebarModel(message)` in server.ts classifies messages:
- **Activity** — streams every browse command via `/activity/stream` SSE.
- **Refs** — REST: `GET /refs` — current page's `@ref` element labels.
- **Inspector** — CDP-based element picker; SSE on `/inspector/events`.
| Pattern | Model | Why |
|---------|-------|-----|
| "click @e24", "go to hn", "screenshot" | sonnet | Deterministic tool calls, no thinking needed |
| "what does this page say?", "summarize" | opus | Needs comprehension |
| "find bugs", "check for broken links" | opus | Analysis task |
| "navigate to X and fill the form" | sonnet | Action-oriented, no analysis words |
When the debug strip closes, the Terminal pane re-becomes visible.
xterm.js doesn't auto-redraw when its container flips from `display:none`
to `display:flex`, so sidepanel-terminal.js runs a `MutationObserver` on
`#tab-terminal`'s class attribute and forces a fit + refresh when
`.active` returns.
Analysis words (`what`, `why`, `how`, `summarize`, `describe`, `analyze`, `read X and Y`)
always override action verbs and force opus.
## Known Failure Modes
| Failure | Symptom | Root Cause | Fix |
|---------|---------|------------|-----|
| Stale auth token | "Unauthorized" in input | Server restarted, background had old token | background.js refreshes token on every health poll |
| Tab ID mismatch | Message sent, no response visible | Server assigned tabId 1, sidebar polling tabId 0 | switchChatTab preserves optimistic UI during switch |
| Sidebar agent not running | Messages queue forever | Agent process failed to spawn or crashed | Check `ps aux | grep sidebar-agent` |
| Agent stale token | Agent runs but no events appear in sidebar | sidebar-agent has old token from .gstack/browse.json | Agent re-reads token before each event POST |
| Queue file missing | spawnClaude fails | Race between server start and agent start | Both sides create file if missing |
| Optimistic UI blown away | User bubble + dots vanish | switchChatTab replaced DOM with welcome screen | Preserved DOM when lastOptimisticMsg is set |
## Per-Tab Concurrency
Each browser tab can run its own agent simultaneously:
- Server: `tabAgents: Map<number, TabAgentState>` with per-tab queue (max 5)
- sidebar-agent: `processingTabs: Set<number>` prevents duplicate spawns
- Two messages on same tab: queued sequentially, processed in order
- Two messages on different tabs: run concurrently
## File Locations
## Files
| Component | File | Runs in |
|-----------|------|---------|
| Sidebar UI | `extension/sidepanel.js` | Chrome side panel |
| Sidebar UI shell | `extension/sidepanel.html` + `sidepanel.js` + `sidepanel.css` | Chrome side panel |
| Terminal UI | `extension/sidepanel-terminal.js` + `extension/lib/xterm.js` | Chrome side panel |
| Service worker | `extension/background.js` | Chrome background |
| Content script | `extension/content.js` | Page context |
| Welcome page | `browse/src/welcome.html` | Page context |
| HTTP server | `browse/src/server.ts` | Bun (compiled binary) |
| Agent process | `browse/src/sidebar-agent.ts` | Bun (non-compiled, can spawn) |
| PTY agent | `browse/src/terminal-agent.ts` | Bun (non-compiled) |
| PTY token store | `browse/src/pty-session-cookie.ts` | Bun (compiled, in server.ts) |
| CLI entry | `browse/src/cli.ts` | Bun (compiled binary) |
| Queue file | `~/.gstack/sidebar-agent-queue.jsonl` | Filesystem |
| State file | `.gstack/browse.json` | Filesystem |
| Chat log | `~/.gstack/sessions/<id>/chat.jsonl` | Filesystem |
| State file | `<stateDir>/browse.json` | Filesystem |
| Terminal port | `<stateDir>/terminal-port` | Filesystem |
| Internal token | `<stateDir>/terminal-internal-token` | Filesystem |
| Claude probe | `<stateDir>/claude-available.json` | Filesystem |
| Active tab | `<stateDir>/active-tab.json` | Filesystem (claude reads) |
+59 -4
View File
@@ -287,6 +287,7 @@ chrome.runtime.onMessage.addListener((msg, sender, sendResponse) => {
const ALLOWED_TYPES = new Set([
'getPort', 'setPort', 'getServerUrl', 'getToken', 'fetchRefs',
'openSidePanel', 'sidebarOpened', 'command', 'sidebar-command',
'getTabState',
// Inspector message types
'startInspector', 'stopInspector', 'elementPicked', 'pickerCancelled',
'applyStyle', 'toggleClass', 'injectCSS', 'resetAll',
@@ -302,6 +303,11 @@ chrome.runtime.onMessage.addListener((msg, sender, sendResponse) => {
return true;
}
if (msg.type === 'getTabState') {
snapshotTabs().then(snap => sendResponse(snap || { active: null, tabs: [] }));
return true; // async sendResponse
}
if (msg.type === 'setPort') {
savePort(msg.port).then(() => {
checkHealth();
@@ -506,11 +512,48 @@ chrome.runtime.onInstalled.addListener(() => {
// Fire on every service worker startup (covers persistent context reuse)
autoOpenSidePanel();
// ─── Tab Switch Detection ────────────────────────────────────────
// Notify sidepanel instantly when the user switches tabs in the browser.
// This is faster than polling — the sidebar swaps chat context immediately.
// ─── Tab Awareness ───────────────────────────────────────────────
// Push live tab state to the sidepanel so claude in the Terminal pane
// always has up-to-date tabs.json + active-tab.json on disk. The
// sidepanel relays these to terminal-agent.ts over the live WebSocket;
// terminal-agent writes the files for claude to read.
async function snapshotTabs() {
try {
const [active] = await chrome.tabs.query({ active: true, currentWindow: true });
const all = await chrome.tabs.query({});
const slim = all.map(t => ({
tabId: t.id,
url: t.url || '',
title: t.title || '',
active: !!t.active,
windowId: t.windowId,
pinned: !!t.pinned,
audible: !!t.audible,
}));
return {
active: active ? { tabId: active.id, url: active.url || '', title: active.title || '' } : null,
tabs: slim,
};
} catch {
return null;
}
}
async function pushTabState(reason) {
const snapshot = await snapshotTabs();
if (!snapshot) return;
chrome.runtime.sendMessage({
type: 'browserTabState',
reason,
...snapshot,
}).catch(() => {}); // expected: sidepanel may not be open
}
chrome.tabs.onActivated.addListener((activeInfo) => {
// Keep the legacy event for any consumer still listening to it (the chat
// path is gone but the message type is harmless), and also fire the new
// unified state push so claude's tabs.json reflects the new active tab.
chrome.tabs.get(activeInfo.tabId, (tab) => {
if (chrome.runtime.lastError || !tab) return;
chrome.runtime.sendMessage({
@@ -518,8 +561,20 @@ chrome.tabs.onActivated.addListener((activeInfo) => {
tabId: activeInfo.tabId,
url: tab.url || '',
title: tab.title || '',
}).catch(() => {}); // expected: sidepanel may not be open
}).catch(() => {});
});
pushTabState('activated');
});
chrome.tabs.onCreated.addListener(() => pushTabState('created'));
chrome.tabs.onRemoved.addListener(() => pushTabState('removed'));
chrome.tabs.onUpdated.addListener((_id, changeInfo) => {
// Throttle: only re-push on URL or title changes, not on every loading
// tick. We don't want to spam claude with a state push every 50ms while
// a page loads.
if (changeInfo.url || changeInfo.title || changeInfo.status === 'complete') {
pushTabState('updated');
}
});
// ─── Startup ────────────────────────────────────────────────────
+1 -1
View File
@@ -4,7 +4,7 @@
"version": "0.1.0",
"description": "Live activity feed and @ref overlays for gstack browse",
"permissions": ["sidePanel", "storage", "activeTab", "scripting"],
"host_permissions": ["http://127.0.0.1:*/"],
"host_permissions": ["http://127.0.0.1:*/", "ws://127.0.0.1:*/"],
"action": {
"default_icon": {
"16": "icons/icon-16.png",
+442
View File
@@ -0,0 +1,442 @@
/**
* Terminal sidebar tab interactive Claude Code PTY in xterm.js.
*
* Lifecycle (per plan + codex review):
* 1. Sidebar opens. Terminal is the default-active tab.
* 2. Bootstrap card shows "Press any key to start Claude Code."
* 3. On first keystroke (lazy spawn codex finding #8): the extension
* a) POSTs /pty-session on the browse server with the AUTH_TOKEN to
* mint a short-lived HttpOnly cookie scoped to the terminal-agent.
* b) Opens ws://127.0.0.1:<terminalPort>/ws — the cookie travels
* automatically. Terminal-agent validates the cookie + the
* chrome-extension:// Origin (codex finding #9), then spawns
* claude in a PTY.
* 4. Bytes pump both ways. Resize observer sends {type:"resize"} text
* frames; tab-switch hooks send {type:"tabSwitch"} frames.
* 5. PTY exits or WS closes -> we show "Session ended" with a restart
* button. We do NOT auto-reconnect (codex finding #8: auto-reconnect
* = burn fresh claude session every time).
*
* Keep this file dependency-free. xterm.js + xterm-addon-fit are loaded
* via <script src> tags in sidepanel.html (window.Terminal, window.FitAddon).
*/
(function () {
'use strict';
const Terminal = window.Terminal;
const FitAddonModule = window.FitAddon;
if (!Terminal) {
console.error('[gstack terminal] xterm not loaded');
return;
}
const els = {
bootstrap: document.getElementById('terminal-bootstrap'),
bootstrapStatus: document.getElementById('terminal-bootstrap-status'),
installCard: document.getElementById('terminal-install-card'),
installRetry: document.getElementById('terminal-install-retry'),
mount: document.getElementById('terminal-mount'),
ended: document.getElementById('terminal-ended'),
restart: document.getElementById('terminal-restart'),
restartNow: document.getElementById('terminal-restart-now'),
};
/** State machine. */
const STATE = { IDLE: 'idle', CONNECTING: 'connecting', LIVE: 'live', ENDED: 'ended', NO_CLAUDE: 'no-claude' };
let state = STATE.IDLE;
let term = null;
let fitAddon = null;
let ws = null;
function show(el) { el.style.display = ''; }
function hide(el) { el.style.display = 'none'; }
function setState(next, opts = {}) {
state = next;
switch (next) {
case STATE.IDLE:
show(els.bootstrap);
hide(els.installCard);
hide(els.mount);
hide(els.ended);
els.bootstrapStatus.textContent = opts.message || 'Press any key to start Claude Code.';
break;
case STATE.CONNECTING:
show(els.bootstrap);
hide(els.installCard);
hide(els.mount);
hide(els.ended);
els.bootstrapStatus.textContent = 'Connecting...';
break;
case STATE.LIVE:
hide(els.bootstrap);
hide(els.installCard);
show(els.mount);
hide(els.ended);
break;
case STATE.ENDED:
hide(els.bootstrap);
hide(els.installCard);
hide(els.mount);
show(els.ended);
break;
case STATE.NO_CLAUDE:
show(els.bootstrap);
show(els.installCard);
hide(els.mount);
hide(els.ended);
els.bootstrapStatus.textContent = '';
break;
}
}
/**
* Read auth + terminalPort from the server's /health. We don't fetch this
* here sidepanel.js already polls /health for connection state and
* exposes the relevant fields on window.gstackHealth (set below in init()).
* If terminalPort is missing, the agent isn't ready yet.
*/
function getHealth() {
return window.gstackHealth || {};
}
function getServerPort() {
return window.gstackServerPort || null;
}
function getAuthToken() {
return window.gstackAuthToken || null;
}
/**
* POST /pty-session to mint a fresh terminal session. Returns
* { terminalPort, ptySessionToken, expiresAt } on success, or
* { error } on failure. The token rides on the WebSocket
* Sec-WebSocket-Protocol header, which is the only auth header
* the browser WebSocket API lets us set. The token is NOT persisted
* each sidebar load mints a fresh one and discards it on close.
*/
async function mintSession() {
const serverPort = getServerPort();
const token = getAuthToken();
if (!serverPort || !token) {
return { error: 'browse server not ready' };
}
try {
const resp = await fetch(`http://127.0.0.1:${serverPort}/pty-session`, {
method: 'POST',
headers: { 'Authorization': `Bearer ${token}` },
credentials: 'include',
});
if (!resp.ok) {
const body = await resp.text().catch(() => '');
return { error: `${resp.status} ${body || resp.statusText}` };
}
return await resp.json();
} catch (err) {
return { error: err && err.message ? err.message : String(err) };
}
}
async function checkClaudeAvailable(terminalPort) {
try {
const resp = await fetch(`http://127.0.0.1:${terminalPort}/claude-available`, {
credentials: 'include',
});
if (!resp.ok) return { available: false };
return await resp.json();
} catch {
return { available: false };
}
}
function ensureXterm() {
if (term) return;
term = new Terminal({
fontFamily: '"JetBrains Mono", "SF Mono", Menlo, monospace',
fontSize: 13,
theme: { background: '#0a0a0a', foreground: '#e5e5e5' },
cursorBlink: true,
scrollback: 5000,
allowTransparency: false,
convertEol: false,
});
if (FitAddonModule && FitAddonModule.FitAddon) {
fitAddon = new FitAddonModule.FitAddon();
term.loadAddon(fitAddon);
}
// CRITICAL: caller must make els.mount visible BEFORE invoking
// ensureXterm. xterm.js measures the container synchronously inside
// term.open() — if the mount is display:none, xterm caches a 0-size
// viewport and never auto-grows even after the container goes
// visible. The visible-first pattern is enforced by connect()
// calling setState(STATE.LIVE) before us.
term.open(els.mount);
// First fit waits for the next paint frame so the browser has
// applied the .active class transition. Otherwise term.cols/rows
// can come back as the minimum (2x2) when the mount's clientHeight
// is still being computed.
requestAnimationFrame(() => {
try {
fitAddon && fitAddon.fit();
if (ws && ws.readyState === WebSocket.OPEN) {
ws.send(JSON.stringify({ type: 'resize', cols: term.cols, rows: term.rows }));
}
} catch {}
});
const ro = new ResizeObserver(() => {
try {
fitAddon && fitAddon.fit();
if (ws && ws.readyState === WebSocket.OPEN) {
ws.send(JSON.stringify({ type: 'resize', cols: term.cols, rows: term.rows }));
}
} catch {}
});
ro.observe(els.mount);
term.onData((data) => {
if (ws && ws.readyState === WebSocket.OPEN) {
ws.send(new TextEncoder().encode(data));
}
});
}
/**
* Inject a string into the live PTY (the same way a real keystroke would).
* Used by the toolbar's Cleanup button and the Inspector's "Send to Code"
* action so the user can drive claude from outside-the-keyboard surfaces.
* Returns true if the bytes went out, false if no live session.
*/
window.gstackInjectToTerminal = function (text) {
if (!text || !ws || ws.readyState !== WebSocket.OPEN) return false;
try {
ws.send(new TextEncoder().encode(text));
return true;
} catch {
return false;
}
};
async function connect() {
if (state !== STATE.IDLE) return; // already connecting/live
setState(STATE.CONNECTING);
const minted = await mintSession();
if (minted.error) {
setState(STATE.IDLE, { message: `Cannot start: ${minted.error}` });
return;
}
const { terminalPort, ptySessionToken } = minted;
if (!ptySessionToken) {
setState(STATE.IDLE, { message: 'Cannot start: no session token returned' });
return;
}
// Pre-flight: does claude even exist on PATH?
const claudeStatus = await checkClaudeAvailable(terminalPort);
if (!claudeStatus.available) {
setState(STATE.NO_CLAUDE);
return;
}
// setState(LIVE) flips terminal-mount from display:none to display:flex.
// We MUST do that BEFORE ensureXterm() — xterm.js measures the container
// synchronously inside term.open() and a hidden container yields a 0x0
// terminal that never recovers. ensureXterm + the requestAnimationFrame
// fit() inside it run after the browser has applied the layout.
setState(STATE.LIVE);
ensureXterm();
// Token rides on Sec-WebSocket-Protocol — the only auth header the
// browser WebSocket API lets us set. Cross-port HttpOnly cookies with
// SameSite=Strict don't survive the jump from server.ts:34567 to the
// agent's random port from a chrome-extension origin, so cookies
// alone weren't reliable.
ws = new WebSocket(`ws://127.0.0.1:${terminalPort}/ws`, [`gstack-pty.${ptySessionToken}`]);
ws.binaryType = 'arraybuffer';
ws.addEventListener('open', () => {
try {
ws.send(JSON.stringify({ type: 'resize', cols: term.cols, rows: term.rows }));
} catch {}
// Push a fresh tab snapshot so claude's tabs.json is populated by
// the time the lazy spawn finishes booting. Background.js exposes
// the snapshot helper via chrome.runtime; we ask for it here and
// forward whatever comes back.
try {
chrome.runtime.sendMessage({ type: 'getTabState' }, (resp) => {
if (resp && ws && ws.readyState === WebSocket.OPEN) {
try {
ws.send(JSON.stringify({
type: 'tabState',
active: resp.active,
tabs: resp.tabs,
reason: 'initial',
}));
} catch {}
}
});
} catch {}
// Send a single byte to nudge the agent to spawn claude (lazy-spawn trigger).
try { ws.send(new TextEncoder().encode('\n')); } catch {}
});
ws.addEventListener('message', (ev) => {
if (typeof ev.data === 'string') {
// Agent control message (rare). Treat as JSON; error frames carry code.
try {
const msg = JSON.parse(ev.data);
if (msg.type === 'error' && msg.code === 'CLAUDE_NOT_FOUND') {
setState(STATE.NO_CLAUDE);
try { ws.close(); } catch {}
}
} catch {}
return;
}
// Binary: feed to xterm.
const buf = ev.data instanceof ArrayBuffer ? new Uint8Array(ev.data) : ev.data;
term.write(buf);
});
ws.addEventListener('close', () => {
ws = null;
if (state !== STATE.NO_CLAUDE) setState(STATE.ENDED);
});
ws.addEventListener('error', (err) => {
console.error('[gstack terminal] ws error', err);
});
}
function teardown() {
try { ws && ws.close(); } catch {}
ws = null;
if (term) {
try { term.dispose(); } catch {}
term = null;
fitAddon = null;
}
setState(STATE.IDLE);
}
// ─── Wiring ───────────────────────────────────────────────────
/**
* Force a fresh session: close any open WS, dispose xterm, return to
* IDLE, kick off auto-connect. Safe to call from any state.
*/
function forceRestart() {
try { ws && ws.close(); } catch {}
ws = null;
if (term) {
try { term.dispose(); } catch {}
term = null;
fitAddon = null;
}
setState(STATE.IDLE, { message: 'Starting Claude Code...' });
tryAutoConnect();
}
/**
* Repaint xterm when the Terminal pane becomes visible. xterm.js has a
* known issue where its renderer doesn't redraw after a display:none
* display:flex flip the canvas/DOM stays blank until something forces
* a layout pass. fit() recomputes dimensions, refresh() redraws.
*/
function repaintIfLive() {
if (state !== STATE.LIVE || !term) return;
try { fitAddon && fitAddon.fit(); } catch {}
try { term.refresh(0, term.rows - 1); } catch {}
try {
if (ws && ws.readyState === WebSocket.OPEN) {
ws.send(JSON.stringify({ type: 'resize', cols: term.cols, rows: term.rows }));
}
} catch {}
}
function init() {
setState(STATE.IDLE, { message: 'Starting Claude Code...' });
els.installRetry?.addEventListener('click', () => {
// Re-probe claude on PATH, then try a connect.
setState(STATE.IDLE, { message: 'Starting Claude Code...' });
tryAutoConnect();
});
// Two restart buttons:
// - els.restart lives inside the ENDED state card (visible only after
// a session has ended).
// - els.restartNow lives in the always-visible toolbar (lets the user
// force a fresh claude mid-session without waiting for it to exit).
els.restart?.addEventListener('click', forceRestart);
els.restartNow?.addEventListener('click', forceRestart);
// Live browser-tab state. background.js → sidepanel.js → us. We
// forward over the live PTY WebSocket; terminal-agent.ts writes
// <stateDir>/active-tab.json + <stateDir>/tabs.json so claude can
// always read the current tab landscape.
document.addEventListener('gstack:tab-state', (ev) => {
if (!ws || ws.readyState !== WebSocket.OPEN) return;
try {
ws.send(JSON.stringify({
type: 'tabState',
active: ev.detail?.active,
tabs: ev.detail?.tabs,
reason: ev.detail?.reason,
}));
} catch {}
});
// Repaint after a debug-tab → primary-pane transition. The debug
// tabs (Activity / Refs / Inspector) hide the Terminal pane via
// .tab-content { display: none }; xterm doesn't auto-redraw when its
// container flips back to visible, so we listen for the close-debug
// event and force a fit + refresh.
const observer = new MutationObserver(() => {
const term = document.getElementById('tab-terminal');
if (term?.classList.contains('active')) {
requestAnimationFrame(repaintIfLive);
}
});
const target = document.getElementById('tab-terminal');
if (target) observer.observe(target, { attributes: true, attributeFilter: ['class'] });
tryAutoConnect();
}
/**
* Eager-connect when the sidebar opens. Polls for sidepanel.js to populate
* window.gstackServerPort + window.gstackAuthToken (which it does as soon
* as /health succeeds), then fires connect() automatically. The user
* doesn't have to press a key Terminal is the default tab and "tap to
* start" was a needless paper cut on every reload.
*/
function tryAutoConnect() {
if (state !== STATE.IDLE) return;
let waited = 0;
const tick = () => {
// If the user navigated away (Chat tab) or already connected, drop out.
if (state !== STATE.IDLE) return;
if (getServerPort() && getAuthToken()) {
connect();
return;
}
waited += 200;
if (waited > 15000) {
setState(STATE.IDLE, { message: 'Browse server not ready. Reload sidebar to retry.' });
return;
}
setTimeout(tick, 200);
};
tick();
}
if (document.readyState === 'loading') {
document.addEventListener('DOMContentLoaded', init);
} else {
init();
}
})();
+112
View File
@@ -675,6 +675,118 @@ body::after {
}
.tab-content.active { display: flex; flex-direction: column; }
/* ─── Terminal Tab ────────────────────────────────────────────── */
/* The Terminal pane manages its own scrolling (xterm has a viewport with
scrollback). The default .tab-content rules above set overflow-y: auto,
which collapses min-height for nested flex children that's why
.terminal-mount couldn't grow to fill available space. Override here. */
#tab-terminal {
background: #0a0a0a;
padding: 0;
overflow: hidden;
min-height: 0;
}
#tab-terminal.active {
display: flex;
flex-direction: column;
}
.terminal-toolbar {
display: flex;
align-items: center;
justify-content: space-between;
gap: 6px;
padding: 4px 8px;
border-bottom: 1px solid #1a1a1a;
background: #0a0a0a;
flex-shrink: 0;
}
.terminal-toolbar-actions {
display: flex;
gap: 4px;
flex-wrap: wrap;
}
.terminal-toolbar-btn {
background: transparent;
border: 1px solid #27272a;
color: #a1a1aa;
padding: 3px 10px;
font-size: 11px;
font-family: 'JetBrains Mono', monospace;
border-radius: 3px;
cursor: pointer;
}
.terminal-toolbar-btn:hover {
color: #f59e0b;
border-color: #f59e0b;
}
.terminal-bootstrap {
flex: 1;
display: flex;
flex-direction: column;
align-items: center;
justify-content: center;
text-align: center;
color: #71717a;
padding: 24px;
}
.terminal-bootstrap-icon {
font-size: 32px;
color: #f59e0b;
margin-bottom: 8px;
}
.terminal-bootstrap p { margin: 4px 0; }
.terminal-install-card {
margin: 24px;
padding: 16px;
border: 1px solid #27272a;
border-radius: 6px;
text-align: center;
}
.terminal-install-card a { color: #f59e0b; }
.install-retry-btn {
margin-top: 12px;
padding: 6px 14px;
background: #f59e0b;
color: #0a0a0a;
border: none;
border-radius: 4px;
font-family: inherit;
font-size: 12px;
cursor: pointer;
}
.install-retry-btn:hover { opacity: 0.9; }
.terminal-mount {
/* min-height: 0 is the standard flex-overflow fix without it, a flex
item with overflowing content can't shrink below its content size,
so flex:1 refuses to expand into available space and xterm renders
into whatever the content happens to be (i.e. its own initial 2x2
measurement). With min-height:0 the item respects the flex parent's
remaining space and xterm grows to fill it. */
flex: 1 1 0;
min-height: 0;
width: 100%;
background: #0a0a0a;
padding: 8px;
box-sizing: border-box;
/* position: relative so xterm's absolutely-positioned helpers (the
hidden textarea for input) anchor inside us, not on body. */
position: relative;
}
.terminal-mount .xterm,
.terminal-mount .xterm .xterm-viewport,
.terminal-mount .xterm .xterm-screen {
height: 100% !important;
}
.terminal-ended {
flex: 1;
display: flex;
flex-direction: column;
align-items: center;
justify-content: center;
color: #71717a;
padding: 24px;
}
/* ─── Activity Feed ───────────────────────────────────── */
#activity-feed { flex: 1; }
+32 -64
View File
@@ -3,6 +3,7 @@
<head>
<meta charset="utf-8">
<link rel="stylesheet" href="sidepanel.css">
<link rel="stylesheet" href="lib/xterm.css">
</head>
<body>
<!-- Security shield — reflects ~/.gstack/security/session-state.json status.
@@ -24,54 +25,38 @@
</div>
</div>
<!-- Security event banner — fires on prompt injection detection.
Variant A from /plan-design-review 2026-04-19: centered alert-heavy,
big red error icon, mono layer scores in expandable details. -->
<div class="security-banner" id="security-banner" role="alert" aria-live="assertive" style="display:none">
<button class="security-banner-close" id="security-banner-close" aria-label="Dismiss">&times;</button>
<div class="security-banner-icon" aria-hidden="true">
<svg width="28" height="28" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
<circle cx="12" cy="12" r="10"></circle>
<line x1="12" y1="8" x2="12" y2="12"></line>
<line x1="12" y1="16" x2="12.01" y2="16"></line>
</svg>
</div>
<div class="security-banner-title" id="security-banner-title">Session terminated</div>
<div class="security-banner-subtitle" id="security-banner-subtitle">prompt injection detected</div>
<button class="security-banner-expand" id="security-banner-expand" aria-expanded="false" aria-controls="security-banner-details">
<span>What happened</span>
<svg class="security-banner-chevron" width="12" height="12" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
<polyline points="6 9 12 15 18 9"></polyline>
</svg>
</button>
<div class="security-banner-details" id="security-banner-details" hidden>
<div class="security-banner-section-label">SECURITY LAYERS</div>
<div class="security-banner-layers" id="security-banner-layers"></div>
<div class="security-banner-section-label" id="security-banner-suspect-label" hidden>SUSPECTED TEXT</div>
<pre class="security-banner-suspect" id="security-banner-suspect" hidden></pre>
</div>
<div class="security-banner-actions" id="security-banner-actions" hidden>
<button type="button" class="security-banner-btn security-banner-btn-block" id="security-banner-btn-block">Block session</button>
<button type="button" class="security-banner-btn security-banner-btn-allow" id="security-banner-btn-allow">Allow and continue</button>
</div>
</div>
<!-- Browser tab bar -->
<div class="browser-tabs" id="browser-tabs" style="display:none"></div>
<!-- Chat Tab (default, full height) -->
<main id="tab-chat" class="tab-content active">
<div class="chat-messages" id="chat-messages">
<div class="chat-loading" id="chat-loading">
<div class="chat-loading-spinner"></div>
<p id="loading-status">Looking for browse server...</p>
<pre id="loading-debug" class="muted" style="font-size:11px; font-family:'JetBrains Mono',monospace; white-space:pre-wrap; margin-top:8px; color:#71717A;"></pre>
</div>
<div class="chat-welcome" id="chat-welcome" style="display:none">
<div class="chat-welcome-icon">G</div>
<p>Send a message to Claude Code.</p>
<p class="muted">Your agent will see it and act on it.</p>
<!-- Terminal pane is now the sole primary surface. Activity / Refs /
Inspector still exist behind the `debug` toggle in the footer. -->
<main id="tab-terminal" class="tab-content active" role="tabpanel" aria-label="Terminal">
<!-- Toolbar with browser quick-actions on the left, Restart on the right.
Restart is always visible so the user can force a fresh claude any
time, not just from the ENDED state. -->
<div class="terminal-toolbar" id="terminal-toolbar">
<div class="terminal-toolbar-actions">
<button id="chat-cleanup-btn" class="terminal-toolbar-btn" title="Remove ads, banners, popups">🧹 Cleanup</button>
<button id="chat-screenshot-btn" class="terminal-toolbar-btn" title="Take a screenshot">📸 Screenshot</button>
<button id="chat-cookies-btn" class="terminal-toolbar-btn" title="Import cookies from your browser">🍪 Cookies</button>
</div>
<button class="terminal-toolbar-btn" id="terminal-restart-now" title="Restart Claude Code session">↻ Restart</button>
</div>
<div class="terminal-bootstrap" id="terminal-bootstrap">
<div class="terminal-bootstrap-icon"></div>
<p id="terminal-bootstrap-status">Starting Claude Code...</p>
<p class="muted" id="terminal-bootstrap-hint">Real PTY. Real terminal. Real claude.</p>
<pre id="loading-debug" class="muted" style="font-size:11px; font-family:'JetBrains Mono',monospace; white-space:pre-wrap; margin-top:8px; color:#71717A;"></pre>
</div>
<div class="terminal-install-card" id="terminal-install-card" style="display:none">
<p><strong>Claude Code not found</strong></p>
<p class="muted">Install: <a href="https://docs.anthropic.com/en/docs/claude-code" target="_blank">docs.anthropic.com/en/docs/claude-code</a></p>
<button class="install-retry-btn" id="terminal-install-retry">I installed it &mdash; try again</button>
</div>
<div class="terminal-mount" id="terminal-mount" style="display:none"></div>
<div class="terminal-ended" id="terminal-ended" style="display:none">
<p>Session ended.</p>
<button class="install-retry-btn" id="terminal-restart">Start a new session</button>
</div>
</main>
@@ -174,30 +159,10 @@
</div>
</main>
<!-- Experimental chat banner (shown when chatEnabled) -->
<div id="experimental-banner" class="experimental-banner" style="display: none;">
Browser co-pilot &mdash; controls this browser, reports back to your workspace
</div>
<!-- Quick Actions Toolbar -->
<div class="quick-actions" id="quick-actions">
<button id="chat-cleanup-btn" class="quick-action-btn" title="Remove ads, banners, popups">🧹 Cleanup</button>
<button id="chat-screenshot-btn" class="quick-action-btn" title="Take a screenshot">📸 Screenshot</button>
<button id="chat-cookies-btn" class="quick-action-btn" title="Import cookies from your browser">🍪 Cookies</button>
</div>
<!-- Command Bar -->
<div class="command-bar">
<button class="stop-btn" id="stop-agent-btn" title="Stop agent" style="display: none;">&#x25A0;</button>
<input type="text" class="command-input" id="command-input" placeholder="Ask about this page..." autocomplete="off" spellcheck="false">
<button class="send-btn" id="send-btn" title="Send">&#x2191;</button>
</div>
<!-- Footer with connection + debug toggle -->
<footer>
<div class="footer-left">
<button class="debug-toggle" id="debug-toggle" title="Toggle debug panels">debug</button>
<button class="footer-btn" id="clear-chat" title="Clear chat">clear</button>
<button class="footer-btn" id="reload-sidebar" title="Reload sidebar">reload</button>
</div>
<div class="footer-right">
@@ -215,6 +180,9 @@
<button class="tab close-debug" id="close-debug" title="Close debug">&times;</button>
</nav>
<script src="lib/xterm.js"></script>
<script src="lib/xterm-addon-fit.js"></script>
<script src="sidepanel.js"></script>
<script src="sidepanel-terminal.js"></script>
</body>
</html>
+82 -950
View File
File diff suppressed because it is too large Load Diff
+6 -3
View File
@@ -1,6 +1,6 @@
{
"name": "gstack",
"version": "1.13.1.0",
"version": "1.15.0.0",
"description": "Garry's Stack — Claude Code skills + fast headless browser. One repo, one install, entire AI engineering workflow.",
"license": "MIT",
"type": "module",
@@ -9,7 +9,8 @@
"make-pdf": "./make-pdf/dist/pdf"
},
"scripts": {
"build": "bun run gen:skill-docs --host all; bun build --compile browse/src/cli.ts --outfile browse/dist/browse && bun build --compile browse/src/find-browse.ts --outfile browse/dist/find-browse && bun build --compile design/src/cli.ts --outfile design/dist/design && bun build --compile make-pdf/src/cli.ts --outfile make-pdf/dist/pdf && bun build --compile bin/gstack-global-discover.ts --outfile bin/gstack-global-discover && bash browse/scripts/build-node-server.sh && git rev-parse HEAD > browse/dist/.version && git rev-parse HEAD > design/dist/.version && git rev-parse HEAD > make-pdf/dist/.version && chmod +x browse/dist/browse browse/dist/find-browse design/dist/design make-pdf/dist/pdf bin/gstack-global-discover && (rm -f .*.bun-build || true)",
"build": "bun run vendor:xterm && bun run gen:skill-docs --host all; bun build --compile browse/src/cli.ts --outfile browse/dist/browse && bun build --compile browse/src/find-browse.ts --outfile browse/dist/find-browse && bun build --compile design/src/cli.ts --outfile design/dist/design && bun build --compile make-pdf/src/cli.ts --outfile make-pdf/dist/pdf && bun build --compile bin/gstack-global-discover.ts --outfile bin/gstack-global-discover && bash browse/scripts/build-node-server.sh && git rev-parse HEAD > browse/dist/.version && git rev-parse HEAD > design/dist/.version && git rev-parse HEAD > make-pdf/dist/.version && chmod +x browse/dist/browse browse/dist/find-browse design/dist/design make-pdf/dist/pdf bin/gstack-global-discover && (rm -f .*.bun-build || true)",
"vendor:xterm": "mkdir -p extension/lib && cp node_modules/xterm/lib/xterm.js extension/lib/xterm.js && cp node_modules/xterm/css/xterm.css extension/lib/xterm.css && cp node_modules/xterm-addon-fit/lib/xterm-addon-fit.js extension/lib/xterm-addon-fit.js",
"dev:make-pdf": "bun run make-pdf/src/cli.ts",
"dev:design": "bun run design/src/cli.ts",
"gen:skill-docs": "bun run scripts/gen-skill-docs.ts",
@@ -62,6 +63,8 @@
],
"devDependencies": {
"@anthropic-ai/claude-agent-sdk": "0.2.117",
"@anthropic-ai/sdk": "^0.78.0"
"@anthropic-ai/sdk": "^0.78.0",
"xterm": "5",
"xterm-addon-fit": "^0.8.0"
}
}
+5 -27
View File
@@ -1677,30 +1677,8 @@ describe('no compiled binaries in git', () => {
});
});
describe('sidebar agent (#584)', () => {
// #584 — Sidebar Write: sidebar-agent.ts allowedTools includes Write
test('sidebar-agent.ts allowedTools includes Write', () => {
const content = fs.readFileSync(path.join(ROOT, 'browse', 'src', 'sidebar-agent.ts'), 'utf-8');
// Find the allowedTools line in the askClaude function
const match = content.match(/--allowedTools['"]\s*,\s*['"]([^'"]+)['"]/);
expect(match).not.toBeNull();
expect(match![1]).toContain('Write');
});
// #584 — Server Write: server.ts allowedTools includes Write (DRY parity)
test('server.ts allowedTools excludes Write (agent is read-only + Bash)', () => {
const content = fs.readFileSync(path.join(ROOT, 'browse', 'src', 'server.ts'), 'utf-8');
// Find the sidebar allowedTools in the headed-mode path
const match = content.match(/--allowedTools['"]\s*,\s*['"]([^'"]+)['"]/);
expect(match).not.toBeNull();
expect(match![1]).toContain('Bash');
expect(match![1]).not.toContain('Write');
});
// #584 — Sidebar stderr: stderr handler is not empty
test('sidebar-agent.ts stderr handler is not empty', () => {
const content = fs.readFileSync(path.join(ROOT, 'browse', 'src', 'sidebar-agent.ts'), 'utf-8');
// The stderr handler should NOT be an empty arrow function
expect(content).not.toContain("proc.stderr.on('data', () => {})");
});
});
// `sidebar agent (#584)` describe block was here. sidebar-agent.ts and
// the entire chat-queue path were ripped in favor of the interactive
// claude PTY (terminal-agent.ts); these assertions had no target file.
// Terminal-pane invariants are covered by browse/test/sidebar-tabs.test.ts
// and browse/test/terminal-agent.test.ts.