mirror of
https://github.com/garrytan/gstack.git
synced 2026-05-02 03:35:09 +02:00
934b270960e9ba72aea9488cedd85eb63df12fc5
277 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Garry Tan
|
934b270960 |
test(browser-skills-e2e): exercise dispatch with bundled hackernews-frontpage
Covers the full \$B skill list/show/test pipeline against the real bundled reference skill (defaultTierPaths picks up <repo>/browser-skills/). Verifies frontmatter shape, the three-tier walk surfaces the bundled entry, and \$B skill test successfully runs the bundled script.test.ts in a child bun process. \$B skill run end-to-end against the live network is intentionally NOT covered here (would be flaky against news.ycombinator.com); the spawn lifecycle is exercised in browser-skill-commands.test.ts using inline synthetic skills. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
288442ccc0 |
release: v1.9.0.0 — browser-skills runtime
VERSION 1.8.0.0 → 1.9.0.0. CHANGELOG entry leads with what humans can do today (hand-write deterministic browser scripts, run them in 200ms via \$B skill run). Notes explicitly that agent authoring lands in next release; no fabricated perf numbers. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
3378bfe5b6 |
docs(todos): replace self-authoring-\$B P1 with browser-skills phases
Phase 1 of the browser-skills design shipped on this branch (sidesteps the in-daemon isolation problem the original P1 was blocked on). The new entries enumerate the work that remains: P1: Phase 2 (/scrape + /automate skill templates) P2: Phase 3 (resolver injection at session start) P2: Phase 4 (eval infra + fixture staleness + OS sandbox) Cross-references docs/designs/BROWSER_SKILLS_V1.md for the full architecture and the 8 Codex review findings + responses. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
063a99809a |
docs(designs): add BROWSER_SKILLS_V1 design doc
Captures the 13 locked decisions, two-axis trust model (daemon-side scoped tokens + process-side env access), 3-tier lookup, file layout, and full responses to all 8 Codex outside-voice findings. Includes Phase 2-4 sketches for future branches. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
225305416e |
test(skill-validation): cover bundled browser-skills
Adds 7 assertions per bundled skill at <root>/browser-skills/<name>/: - SKILL.md exists - frontmatter parses with required fields (name/host/triggers/args) - script.ts exists - _lib/browse-client.ts exists and matches the canonical SDK byte-for-byte - script.test.ts exists - script.ts imports browse from ./_lib/browse-client The byte-identical SDK check enforces the version-pinning contract: when the canonical SDK at browse/src/browse-client.ts changes, every bundled skill's _lib/ copy must be re-synced or this test fails. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
8a60d99c74 |
feat(browser-skills): bundled hackernews-frontpage reference skill
Smallest interesting browser-skill: scrapes HN front page, returns
30 stories as JSON. No auth, stable HTML, fully fixture-tested.
Files:
SKILL.md frontmatter + prose
script.ts exports parseStoriesFromHtml(html)
main: goto + html + parse + JSON.stringify
_lib/browse-client.ts vendored copy of the SDK
fixtures/hn-2026-04-26.html captured front page (5 stories)
script.test.ts 13 assertions against the fixture
The parser is a pure function over HTML so script.test.ts runs
without a daemon (just imports parseStoriesFromHtml and asserts).
This exercises every Phase 1 component end-to-end:
- browse-client SDK (script imports browse from ./_lib/)
- 3-tier lookup (hackernews-frontpage lives in the bundled tier)
- scoped tokens (read+write is enough for goto + html)
- spawn lifecycle (\$B skill run hackernews-frontpage)
- file-fixture testing (\$B skill test hackernews-frontpage)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
e6a27993fe |
feat(browser-skills): \$B skill list/show/run/test/rm subcommands
handleSkillCommand dispatches to per-subcommand handlers; spawnSkill is
the load-bearing function that:
1. Mints a per-spawn scoped token (read+write only) bound to the
skill name + spawn-id.
2. Builds the spawn env:
- trusted: passes process.env minus GSTACK_TOKEN (defense in depth).
- untrusted: minimal allowlist (LANG, LC_ALL, TERM, TZ) + locked
PATH; explicitly drops anything matching TOKEN/KEY/SECRET/etc.
Also drops AWS_/AZURE_/GCP_/GOOGLE_APPLICATION_/ANTHROPIC_/OPENAI_/
GITHUB_/GH_/SSH_/GPG_/NPM_TOKEN/PYPI_ patterns.
3. Always injects GSTACK_PORT + GSTACK_SKILL_TOKEN last (cannot be
overridden by parent env).
4. Spawns bun run script.ts -- <args> with cwd=skillDir, captures
stdout (1MB cap), stderr, and timeout-kills past the deadline.
5. Revokes the token in finally{}, always.
list output prints the resolved tier inline so "why did it run that
one?" never becomes a debugging mystery (Codex finding #4 mitigation).
server.ts threads the listen port to meta-commands via MetaCommandOpts.daemonPort.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
faf663b22c |
feat(browser-skills): 3-tier storage helpers
listBrowserSkills() walks project > global > bundled (first-wins),
parses SKILL.md frontmatter, no INDEX.json. readBrowserSkill() does
the same for a single name. tombstoneBrowserSkill() moves a skill
into .tombstones/<name>-<ts>/ for recoverability.
Frontmatter parser handles the subset browser-skills need: scalars
(host, description, trusted, version, source), string lists
(triggers), and arg-mapping lists ([{name, description}, ...]).
Quoted values handle colons; trusted defaults to false.
Bundled tier path is auto-detected from the binary install location;
project tier comes from git rev-parse; global is ~/.gstack/. All tier
paths are overridable for hermetic tests.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
c0dff84647 |
feat(browse-client): SDK for browser-skill scripts
Thin wrapper over POST /command with bearer auth. Resolves daemon port + token from GSTACK_PORT + GSTACK_SKILL_TOKEN env vars first (set by $B skill run when spawning), falls back to .gstack/browse.json for standalone debug runs. Convenience methods cover the read+write surface skills typically need: goto, click, fill, text, html, snapshot, links, forms, accessibility, attrs, media, data, scroll, press, type, select, wait, hover, screenshot. Low-level command(cmd, args) escape hatch for anything else. This is the canonical SDK source. Each browser-skill ships a sibling copy at <skill>/_lib/browse-client.ts so each skill is fully portable and version-pinned. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
8d7fcd44c2 |
feat(skill-token): mint scoped tokens per skill spawn
Wraps token-registry.createToken/revokeToken with skill-specific clientId encoding (skill:<name>:<spawn-id>) and read+write defaults. Skill scripts get a per-spawn capability token bound to browser-driving commands; the daemon root token never leaves the harness. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
b3d5a6b0b7 |
test(skill-validation): exempt deliberate large fixtures from 2MB limit
Pre-existing failure: the "git tracks no files larger than 2MB" test caught browse/test/fixtures/security-bench-haiku-responses.json (28.8MB of replay data committed in v1.6.4.0 for security benchmark gate tests). The test exists to catch accidentally-committed binaries (Mach-O dist binaries, etc), not to forbid all large files. Add an explicit LARGE_FIXTURE_EXEMPTIONS allowlist so deliberate replay fixtures pass the gate while accidental binaries still fail. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
472e79a28b |
test(brain-sync): GSTACK_HOME isolation test compares mtime, not content
Pre-existing flaky test: the GSTACK_HOME-overrides-real-config test asserted the real ~/.gstack/config.yaml does NOT contain "gbrain_sync_mode: full" after the test. That fails for any user whose real config legitimately has that key set from prior usage — the test's invariant is "the command did not modify the real file," not "the real file lacks any specific value." Switch to mtime + content snapshot: capture both BEFORE running the command, then verify both are unchanged after. Also add a positive assertion that the tmpHome config DID get the new key. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
c72c56dde9 |
fix(brain-sync): bearer-token regex catches values with leading space
Pre-existing bug from v1.7.0.0: the bearer-token-json secret pattern
required values matching [A-Za-z0-9_./+=-]{16,}, which rejected the
"Bearer <token>" form because the literal space after "Bearer" wasn't
in the character class. Real Authorization headers use "Bearer <token>"
syntax, and the test fixture
'"authorization":"Bearer abcdef1234567890abcdef1234567890"'
sat unscanned despite being a leak-class secret.
One-character fix: add space to the value character class. Test
'gstack-brain-sync secret scan > blocks bearer-json' now passes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
4c3360f067 |
test(fixtures): regenerate ship SKILL.md golden baselines for v1.7.0.0
Pre-existing failures inherited from garrytan/gbrain-support: the GBrain Sync preamble block (added in v1.7.0.0) appears in regenerated SKILL.md output but the golden baselines in test/fixtures/golden/ were never updated. Three failures fixed: golden-file regression > Claude ship skill matches golden baseline golden-file regression > Codex ship skill matches golden baseline golden-file regression > Factory ship skill matches golden baseline Goldens regenerated by copying the current ship/SKILL.md, codex .agents/skills/gstack-ship/SKILL.md, and .factory/skills/gstack-ship/SKILL.md files. Diff is the v1.7.0.0 GBrain Sync preamble block + privacy stop-gate (no behavioral changes — just preamble text). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
e1d1029a22 |
docs: regenerate SKILL.md files with new $B commands
bun run gen:skill-docs picks up the domain-skill and cdp META_COMMANDS entries added in commands.ts. Both top-level SKILL.md and browse/SKILL.md now list the new commands in their Meta and Inspection tables. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
27962738db |
test(browse): E2E gate-tier tests for domain-skills + CDP
domain-skills-e2e.test.ts (4 tests): - save derives host from active tab top-level origin (T3) - save lands quarantined; list surfaces it - readSkill returns null until 3 uses without flag promote to active (T6) - save without an active page errors with structured guidance cdp-e2e.test.ts (8 tests): - Accessibility.getFullAXTree returns wrapped JSON (allowed, untrusted-output) - Performance.getMetrics returns plain JSON (allowed, trusted-output) - Runtime.evaluate DENIED with structured guidance (T2 RCE block) - Page.navigate DENIED (must use $B goto for blocklist routing) - Network.getResponseBody DENIED (exfil block) - malformed JSON params surfaces clear error - non Domain.method format surfaces clear error - $B cdp help returns help text Both files boot a real Chromium via BrowserManager.launch() and exercise the dispatch handlers end-to-end. Total 12 E2E tests in <2s. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
cc90f6761d |
fix(browse): lazy GSTACK_HOME resolution in domain-skills
Module-level constants (GLOBAL_FILE, derived path) were evaluated at module-load and cached. When E2E and unit tests run in the same Bun test pass and set GSTACK_HOME differently, the second test sees the first test's path. Switch to lazy gstackHome() / globalFile() / projectFile() helpers so process.env mutations take effect. Mirrors the pattern already used in telemetry.ts. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
bee6ceb4f3 |
todos: 7 follow-ups from v1.8.0.0 review trail
P1: Self-authoring $B commands with out-of-process worker isolation
(Codex T1 deferred from v1.8.0.0 — needs real isolation design)
P2: Migrate /learn to SQLite (Codex T5 long-term primitive fix)
P2: Remove plan-mode handshake from /plan-devex-review (skill bug)
P3: GBrain skillpack publishing for domain-skills
P3: Replay/record demonstrated flows to domain-skills
P3: $B commands review batch-mode UX (alternative to inline approval)
P3: Heuristic command-gap watcher (DX D4 alternative C)
Each entry has the standard What/Why/Pros/Cons/Context/Effort/Priority/
Depends-on shape so anyone picking these up later has full context.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
c8d62612b3 |
chore: bump version and changelog (v1.8.0.0)
Branch-scoped bump on top of merged 1.7.0.0 base. CHANGELOG entry covers the full v1.8.0.0 scope: $B domain-skill, $B cdp escape hatch, two-tier mutex, telemetry signals, sidebar prompt-context injection. Includes Codex outside-voice trail (7 of 20 findings resolved, 12 mooted by T1 scope drop). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
4dea0846ed |
docs(readme): browser-harness-js plug + domain-skills section
New "Domain skills + raw CDP escape hatch" section under "The sprint" covering both v1.8.0.0 features. Plugs browser-use/browser-harness-js as the no-rails alternative for users who want raw CDP without gstack's security stack. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
ce9b277581 |
docs: domain-skills reference + error lookup table
docs/domain-skills.md mirrors the layered shape of docs/gbrain-sync.md (DX D8): how agents use it, state machine, storage layout, security model (L1-L3 + L4 layered defense), error reference table. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
6c51bf9b1b |
test(browse): telemetry module
3 tests covering: - logTelemetry writes JSONL with ts injected - GSTACK_TELEMETRY_OFF=1 silences all events - logTelemetry never throws on disk failures Uses GSTACK_HOME env var to redirect writes to a tmp dir; the telemetry module reads HOME lazily so test mutations take effect. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
c2074f4d59 |
feat(browse): sidebar prompt-context injection + CDP telemetry
server.ts spawnClaude now: - Imports per-project domain skill matching the active tab's hostname via readDomainSkill() - Wraps the body in UNTRUSTED EXTERNAL CONTENT envelope (so the L4 classifier in sidebar-agent sees it at load time per Eng D4) - Appends as <domain-skill source="..." host="..." version="..."> block - Fires domain_skill_fired telemetry (host, source, version) - Calls recordSkillUse fire-and-forget so the auto-promote-after-N=3 state machine advances on each successful prompt injection System prompt also gets a one-liner introducing $B domain-skill commands to agents (DX D4 start-of-task discoverability hint). cdp-bridge.ts fires: - cdp_method_denied (drives next allow-list growth) - cdp_method_lock_acquire_ms (P50/P99 quantile observability) - cdp_method_called (allowed methods) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
b0d1a9b2e9 |
feat(browse): telemetry signals + project-slug helper
Lightweight telemetry per DX D9: piggybacks on ~/.gstack/analytics/ pattern.
Hostname + aggregate counters only, no body content. GSTACK_TELEMETRY_OFF=1
silences. Fire-and-forget — never blocks calling path.
Signals fired so far:
- domain_skill_saved {host, scope, state, bytes}
- domain_skill_save_blocked {host, reason}
(domain_skill_fired and cdp_method_* fired in subsequent commits.)
Also extracts project-slug resolution into project-slug.ts so server.ts
and domain-skill-commands.ts share one cached lookup.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
d2a4ea0b6a |
test(browse): CDP allowlist + two-tier mutex
13 tests: - Allowlist linter: every entry has 4 required fields, no duplicates, justification length > 20 chars - Deny-list verification: dangerous methods (Runtime.evaluate, Page.navigate, Network.getResponseBody, Browser.close, Target.attachToTarget, etc.) are NOT allowed (Codex T2 categories 4-7) - Per-tab mutex serializes ops on same tab - Per-tab mutex allows parallel ops across different tabs - Global lock blocks tab locks; tab locks block global lock - Acquire timeout yields CDPMutexAcquireTimeout (no silent hang) - Timeout error names the tab id and the timeout budget Also extends Network.disable justification to satisfy linter. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
78c1f5b33c |
feat(browse): $B cdp escape hatch — deny-default allowlist + two-tier mutex
Codex T2: flip CDP posture to deny-default. Allowed methods enumerated in cdp-allowlist.ts with (scope: tab|browser, output: trusted|untrusted, justification) per entry. Initial allowlist (~25 methods) covers: - Accessibility tree extraction (read-only) - DOM/CSS inspection (read-only) - Performance metrics - Tracing - Emulation viewport/UA override - Page screenshot/PDF capture (output is binary, no marker injection vector) - Network.enable/disable (no bodies/cookies — those are exfil surfaces) - Runtime.getProperties (NO evaluate/callFunctionOn — those would be RCE) Page.navigate is INTENTIONALLY NOT allowed; agents use $B goto which goes through the URL blocklist. Codex T7: two-tier mutex. tab-scoped methods take per-tab lock; browser- scoped take global lock that blocks all tab locks. 5s acquire timeout yields CDPMutexAcquireTimeout (no silent hangs). All lock acquires use try/finally so errors don't leak the lock. Path A from spike: uses Playwright's newCDPSession() per page. No second WebSocket, no need for --remote-debugging-port. CDPSession is cached per page in a WeakMap and cleared on page close. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
41f008debd |
feat(browse): $B domain-skill subcommands
Wire the domain-skills storage layer into the browse CLI as a META command:
$B domain-skill save save body from stdin or --from-file
(host derived from active tab — T3)
$B domain-skill list list all skills visible to current project
$B domain-skill show <host> print skill body
$B domain-skill edit <host> open in $EDITOR
$B domain-skill promote-to-global <host> cross-project promotion (T4)
$B domain-skill rollback <host> [--global] restore prior version
$B domain-skill rm <host> [--global] tombstone
Save path runs L1-L3 content filters from content-security.ts (importable
in compiled binary, unlike L4 ML classifier — see CLAUDE.md). The L4
classifier scan happens in sidebar-agent at prompt-injection load time.
Output is structured (problem + cause + suggested-action) per DX D7.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
58b50da77d |
test(browse): domain-skills storage + state machine
14 tests covering: - T3 hostname normalization (lowercase, www. strip, port/path/query strip, subdomain-exact preserved) - T4 scope shadowing (per-project active shadows global for same host) - T5 persistence (version monotonicity, tolerant parser drops partial line) - T6 state machine (quarantined → active after N=3 uses, classifier-flag blocks promotion, save-time score >= 0.85 rejected) - Rollback by version log (restore prior body, advance version counter) - Tombstone deletion (read returns null after delete) All 14 pass in 27ms via bun test. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
db38fd2c07 |
feat(browse): domain-skills storage + state machine
New module browse/src/domain-skills.ts implements the per-site notes
the agent writes for itself, persisted as type:"domain" rows alongside
/learn's per-project learnings.
Three scopes layered: per-project default, global by explicit promotion.
Project-active shadows global for the same host.
State machine (T6 — codex outside-voice):
quarantined --3 uses w/o flag--> active(project) --promote--> global
^ |
+----- classifier flag during use
- Append-only JSONL with O_APPEND for atomic small writes
- Tolerant parser drops partial trailing line on read
- Tombstone for deletes (compactor cleans up later)
- Version log per (host, scope) enables rollback
- Hostname derived from active tab top-level origin (T3 confused-deputy fix)
- writeSkill rejects classifier_score >= 0.85 with structured error
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
54d3ad923d |
Merge branch 'main' into garrytan/gbrain-support
Main added v1.6.4.0 (Haiku classifier FP reduction, security ensemble tuning). My branch is already at v1.7.0.0 so no re-bump needed. Kept v1.7.0.0 across VERSION + package.json. CHANGELOG now reads 1.7.0.0 → 1.6.4.0 → 1.6.3.0 → 1.6.2.0 → 1.6.1.0, contiguous. |
||
|
Garry Tan
|
d75402bbd2 |
v1.6.4.0: cut Haiku classifier FP from 44% to 23%, gate now enforced (#1135)
* feat(security): v2 ensemble tuning — label-first voting + SOLO_CONTENT_BLOCK Cuts Haiku classifier false-positive rate from 44.1% → 22.9% on BrowseSafe-Bench smoke. Detection trades from 67.3% → 56.2%; the lost TPs are all cases Haiku correctly labeled verdict=warn (phishing targeting users, not agent hijack) — they still surface in the WARN banner meta but no longer kill the session. Key changes: - combineVerdict: label-first voting for transcript_classifier. Only meta.verdict==='block' block-votes; verdict==='warn' is a soft signal. Missing meta.verdict never block-votes (backward-compat). - Hallucination guard: verdict='block' at confidence < LOG_ONLY (0.40) drops to warn-vote — prevents malformed low-conf blocks from going authoritative. - New THRESHOLDS.SOLO_CONTENT_BLOCK = 0.92 decoupled from BLOCK (0.85). Label-less content classifiers (testsavant, deberta) need a higher solo-BLOCK bar because they can't distinguish injection from phishing-targeting-user. Transcript keeps label-gated solo path (verdict=block AND conf >= BLOCK). - THRESHOLDS.WARN bumped 0.60 → 0.75 — borderline fires drop out of the 2-of-N ensemble pool. - Haiku model pinned (claude-haiku-4-5-20251001). `claude -p` spawns from os.tmpdir() so project CLAUDE.md doesn't poison the classifier context (measured 44k cache_creation tokens per call before the fix, and Haiku refusing to classify because it read "security system" from CLAUDE.md and went meta). - Haiku timeout 15s → 45s. Measured real latency is 17-33s end-to-end (Claude Code session startup + Haiku); v1's 15s caused 100% timeout when re-measured — v1's ensemble was effectively L4-only in prod. - Haiku prompt rewritten: explicit block/warn/safe criteria, 8 few-shot exemplars (instruction-override → block; social engineering → warn; discussion-of-injection → safe). Test updates: - 5 existing combineVerdict tests adapted for label-first semantics (transcript signals now need meta.verdict to block-vote). - 6 new tests: warn-soft-signal, three-way-block-with-warn-transcript, hallucination-guard-below-floor, above-floor-label-first, backward-compat-missing-meta. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(security): live + fixture-replay bench harness with 500-case capture Adds two new benches that permanently guard the v2 tuning: - security-bench-ensemble-live.test.ts (opt-in via GSTACK_BENCH_ENSEMBLE=1). Runs full ensemble on BrowseSafe-Bench smoke with real Haiku calls. Worker-pool concurrency (default 8, tunable via GSTACK_BENCH_ENSEMBLE_CONCURRENCY) cuts wall clock from ~2hr to ~25min on 500 cases. Captures Haiku responses to fixture for replay. Subsampling via GSTACK_BENCH_ENSEMBLE_CASES for faster iteration. Stop-loss iterations write to ~/.gstack-dev/evals/stop-loss-iter-N-* WITHOUT overwriting canonical fixture. - security-bench-ensemble.test.ts (CI gate, deterministic replay). Replays captured fixture through combineVerdict, asserts detection >= 55% AND FP <= 25%. Fail-closed when fixture is missing AND security-layer files changed in branch diff. Uses `git diff --name-only base` (two-dot) to catch both committed and working-tree changes — `git diff base...HEAD` would silently skip in CI after fixture lands. - browse/test/fixtures/security-bench-haiku-responses.json — 500 cases × 3 classifier signals each. Header includes schema_version, pinned model, component hashes (prompt, exemplars, thresholds, combiner, dataset version). Any change invalidates the fixture and forces fresh live capture. - docs/evals/security-bench-ensemble-v2.json — durable PR artifact with measured TP/FN/FP/TN, 95% CIs, knob state, v1 baseline delta. Checked in so reviewers can see the numbers that justified the ship. Measured baseline on the new harness: TP=146 FN=114 FP=55 TN=185 → 56.2% / 22.9% → GATE PASS Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore(release): v1.5.1.0 — cut Haiku FP 44% → 23% - VERSION: 1.5.0.0 → 1.5.1.0 (TUNING bump) - CHANGELOG: [1.5.1.0] entry with measured numbers, knob list, and stop-loss rule spec - TODOS: mark "Cut Haiku FP 44% → ~15%" P0 as SHIPPED with pointer to CHANGELOG and v1 plan Measured: 56.2% detection (CI 50.1-62.1) / 22.9% FP (CI 18.1-28.6) on 500-case BrowseSafe-Bench smoke. Gate passes (floor 55%, ceiling 25%). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(changelog): add v1.6.4.0 placeholder entry at top Per CLAUDE.md branch-scoped discipline, our VERSION 1.6.4.0 needs a CHANGELOG entry at the top so readers can tell what's on this branch vs main. Honest placeholder: no user-facing runtime changes yet, two merges bringing branch up to main's v1.6.3.0, and the approved injection-tuning plan is queued but unimplemented. Gets replaced by the real release-summary at /ship time after Phases -1 through 10 land. * docs(changelog): strip process minutiae from entries; rewrite v1.6.4.0 CLAUDE.md — new CHANGELOG rule: only document what shipped between main and this change. Keep out branch resyncs, merge commits, plan approvals, review outcomes, scope negotiations, "work queued" or "in-progress" framing. When no user-facing change actually landed, one sentence is the entry: "Version bump for branch-ahead discipline. No user-facing changes yet." CHANGELOG.md — v1.6.4.0 entry rewritten to match. Previous version narrated the branch history, the approved injection-tuning plan, and what we expect to ship later — all of which are process minutiae readers do not care about. * docs(changelog): rewrite v1.6.4.0; strip process minutiae Rewrote v1.6.4.0 entry to follow the new CLAUDE.md rule: only document what shipped between main and this change. Previous entry narrated the branch history, the approved injection-tuning plan, and what we expect to ship later, all process minutiae readers do not care about. v1.6.4.0 now reads: what the detection tuning did for users, the before/after numbers, the stop-loss rule, and the itemized changes for contributors. CLAUDE.md — new rule: only document what shipped between main and this change. Keep out branch resyncs, merge commits, plan approvals, review outcomes, scope negotiations, "work queued" / "in-progress" framing. If nothing user-facing landed, one sentence: "Version bump for branch-ahead discipline. No user-facing changes yet." --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
c7d6add473 |
fix(test): session-awareness reads AskUserQuestion Format from a Tier 2+ SKILL.md
The test was reading ROOT/SKILL.md (browse skill, Tier 1) which never contained '## AskUserQuestion Format' — that section is only emitted for Tier 2+ skills by scripts/resolvers/preamble.ts. As a result the agent was prompted with an empty format guide and only emitted 'RECOMMENDATION' intermittently, making the test flaky. Pre-existing on main (same ROOT/SKILL.md shape there) — surfaced now because the agent run didn't hit the RECOMMENDATION/recommend/option a fallback strings in this particular attempt. Fix: read from office-hours/SKILL.md (Tier 3, always has the section) with a fallback that scans for the first top-level skill dir whose SKILL.md contains the header. Future template moves won't break this test again. |
||
|
Garry Tan
|
8407b4930b |
Merge branch 'main' into garrytan/gbrain-support
Resolves: - VERSION: keep 1.7.0.0 (my branch's bump is higher than main's 1.6.3.0) - package.json: keep 1.7.0.0 (same logic) - CHANGELOG.md: keep 1.7.0.0 entry on top, preserve main's 1.6.2.0 + 1.6.3.0 entries chronologically between 1.7.0.0 and the shared 1.6.1.0 tail - context-save/SKILL.md.tmpl: accept main's deletion of the "Resume flow" section (logic moved to the separate /context-restore skill) - Regenerated all SKILL.md files via bun run gen:skill-docs so they match both branches' template state post-merge |
||
|
Garry Tan
|
69733e2622 |
fix(plan-reviews): restore RECOMMENDATION + Completeness split + Codex ELI10 (v1.6.3.0) (#1149)
* test: add AskUserQuestion format regression eval for plan reviews Four-case periodic-tier eval that captures the verbatim AskUserQuestion text /plan-ceo-review and /plan-eng-review produce, then asserts the format rule is honored: RECOMMENDATION always, Completeness: N/10 only on coverage-differentiated options, and an explicit "options differ in kind" note on kind-differentiated options. Cases: - plan-ceo-review mode selection (kind-differentiated) - plan-ceo-review approach menu (coverage-differentiated) - plan-eng-review per-issue coverage decision - plan-eng-review per-issue architectural choice (kind-differentiated) Classified periodic because behavior depends on Opus non-determinism — gate-tier would flake and block merges. Test harness instructs the agent to write its would-be AskUserQuestion text to $OUT_FILE rather than invoke a real tool (MCP AskUserQuestion isn't wired in the test subprocess). Regex predicates then validate the captured content. Cost: ~$2 per full run. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(plan-reviews): restore RECOMMENDATION + split Completeness by question type Opus 4.7 users reported /plan-ceo-review and /plan-eng-review stopped emitting the RECOMMENDATION line and per-option Completeness: X/10 scores. E2E capture showed the real failure mode: on kind-differentiated questions (mode selection, architectural A-vs-B, cherry-pick), Opus 4.7 either fabricated filler scores (10/10 on every option — conveys nothing) or dropped the format entirely when the metric didn't fit. Fix is at two layers: 1. scripts/resolvers/preamble/generate-ask-user-format.ts splits the old run-on step 3 into: - Step 3 "Recommend (ALWAYS)": RECOMMENDATION is required on every question, coverage- or kind-differentiated. - Step 4 "Score completeness (when meaningful)": emit Completeness: N/10 only when options differ in coverage. When options differ in kind, skip the score and include a one-line explanatory note. Do not fabricate scores. 2. scripts/resolvers/preamble/generate-completeness-section.ts updates the Completeness Principle tail to match. Without this, the preamble contained two rules (one conditional, one unconditional) and the model hedged. Template anchors reinforce the distinction where agent judgment is most likely to drift: - plan-ceo-review Section 0C-bis (approach menu) gets the coverage-differentiated anchor. - plan-ceo-review Section 0F (mode selection) gets the kind-differentiated anchor. - plan-eng-review CRITICAL RULE section gets the coverage-vs-kind rule for every per-issue AskUserQuestion raised during the review. Regenerated SKILL.md for all T2 skills + golden fixtures refreshed. Every skill using the T2 preamble now has the same conditional scoring rule. Verified via new periodic-tier eval (test/skill-e2e-plan-format.test.ts): all 4 cases fail on prior behavior, all 4 pass with this fix. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: bump version and changelog (v1.6.2.0) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * test: add Codex eval for AskUserQuestion format compliance Four-case periodic-tier eval mirrors test/skill-e2e-plan-format.test.ts but drives the plan review skills via codex exec instead of claude -p. Context: Codex under the gpt.md "No preamble / Prefer doing over listing" overlay tends to skip the Simplify/ELI10 paragraph and the RECOMMENDATION line on AskUserQuestion calls. Users have to manually re-prompt "ELI10 and don't forget to recommend" almost every time. This test pins the behavior so regressions surface. Cases: - plan-ceo-review mode selection (kind-differentiated) - plan-ceo-review approach menu (coverage-differentiated) - plan-eng-review per-issue coverage decision - plan-eng-review per-issue architectural choice (kind-differentiated) Assertions on captured AskUserQuestion text: - RECOMMENDATION: Choose present (all cases) - Completeness: N/10 present on coverage, absent on kind - "options differ in kind" note present on kind - ELI10 length floor (>400 chars) — catches bare options-only output Cost: ~\$2-4 per full run. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * fix(preamble): harden AskUserQuestion Format + Codex ELI10 carve-out Follow-up to v1.6.2.0. Codex (GPT-5.4) under the gpt.md overlay treated "No preamble / Prefer doing over listing" as license to skip the Simplify paragraph and the RECOMMENDATION line on AskUserQuestion calls. Users had to manually re-prompt "ELI10 and don't forget to recommend" almost every time. Two layers: 1. model-overlays/gpt.md — adds an explicit "AskUserQuestion is NOT preamble" carve-out. The "No preamble" rule applies to direct answers; AskUserQuestion content must emit the full format (Re-ground, Simplify/ELI10, Recommend, Options). Tells the model: if you find yourself about to skip any of these, back up and emit them — the user will ask anyway, so do it the first time. 2. scripts/resolvers/preamble/generate-ask-user-format.ts — step 2 renamed to "Simplify (ELI10, ALWAYS)" with explicit "not optional verbosity, not preamble" framing. Step 3 "Recommend (ALWAYS)" hardened: "Never omit, never collapse into the options list." All T2 skills regenerated across all hosts. Golden fixtures refreshed (claude-ship, codex-ship, factory-ship). Updated the ELI10 assertion in test/gen-skill-docs.test.ts to match the new wording. Codex compliance to be verified empirically via test/codex-e2e-plan-format.test.ts. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * test: fix Codex eval sandbox + collector API Two test infrastructure bugs in the initial Codex eval landed in the prior commit: 1. sandbox: 'read-only' (the default) blocked Codex from writing $OUT_FILE. Test reported "STATUS: BLOCKED" and exited 0 without a capture file. Fixed: sandbox: 'workspace-write' for all 4 cases, allowing writes inside the tempdir. 2. recordCodexResult called a non-existent evalCollector.record() API (I invented it). The real surface is addTest() with a different field schema. Aligned with test/codex-e2e.test.ts pattern. With both fixed, the eval now actually measures Codex AskUserQuestion format compliance. All 4 cases pass on v1.6.2.0 with the gpt.md carve-out: RECOMMENDATION always, Completeness: N/10 only on coverage, "options differ in kind" note on kind, ELI10 explanation present. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * chore: bump version and changelog (v1.6.3.0) Adds the Codex ELI10 + RECOMMENDATION carve-out scope landed after v1.6.2.0's Claude-verified fix. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> |
||
|
Garry Tan
|
756525100c |
chore: regenerate SKILL.md files for gbrain-sync preamble block
Re-runs bun run gen:skill-docs after adding generateBrainSyncBlock
to scripts/resolvers/preamble.ts in
|
||
|
Garry Tan
|
c3f73f91d4 |
chore: bump version and changelog (v1.7.0.0)
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> |
||
|
Garry Tan
|
91c734a6af |
docs(gbrain-sync): user guide + error lookup + README section
docs/gbrain-sync.md: setup walkthrough, privacy modes, cross-machine workflow, secret protection, two-machine conflict handling, uninstall, troubleshooting reference. docs/gbrain-sync-errors.md: problem/cause/fix index for every user-visible error. Patterned on Rust's error docs + Stripe's API error reference. README.md: 'Cross-machine memory with GBrain sync' section near the top (discovery moment), plus docs-table entry. |
||
|
Garry Tan
|
c064743eda |
test(gbrain-sync): 27-test consolidated suite
test/brain-sync.test.ts covers: - Config: validation, defaults, GSTACK_HOME env isolation - Enqueue: no-op gates, skip list, concurrent atomicity, JSON escape - JSONL merge driver: 3-way + ts-sort + SHA-256 fallback - Init + sync: canonical file creation, merge driver registration, push-reject + fetch+merge retry path - Init refuses different remote (idempotency) - Cross-machine restore round-trip (machine A write → machine B sees) - Secret scan across all 6 regex families (AWS, GH, OpenAI, PEM, JWT, bearer-JSON). --skip-file unblock remediation - Uninstall removes sync config, preserves user data - --discover-new idempotence via mtime+size cursor Behaviors verified via integration smokes during implementation. Known follow-up: bun-test 5s default timeout needs 30s wrapper for spawnSync-heavy tests. |
||
|
Garry Tan
|
a2aa8a07d4 |
feat(gbrain-sync): preamble block — privacy gate + boundary sync
scripts/resolvers/preamble/generate-brain-sync-block.ts emits bash that runs at every skill invocation: - Detects ~/.gstack-brain-remote.txt on machines without local .git and surfaces a restore-available hint (does NOT auto-run restore). - Runs gstack-brain-sync --once at skill start to drain any pending writes (and at skill end via prose instruction). - Once-per-day auto-pull (cached via .brain-last-pull) for append-only JSONL files. - Emits BRAIN_SYNC: status line every skill run. Also emits prose for the host LLM to fire the one-time privacy stop-gate (full / artifacts-only / off) when gbrain is detected and gbrain_sync_mode_prompted is false. Wired into preamble.ts composition. |
||
|
Garry Tan
|
f088fe96f8 |
feat(gbrain-sync): init, restore, uninstall, consumer registry
bin/gstack-brain-init: idempotent first-run. git init ~/.gstack/, .gitignore=*, canonical .brain-allowlist + .brain-privacy-map.json, pre-commit secret-scan hook (defense-in-depth), merge driver registration via git config, gh repo create --private OR arbitrary --remote <url>, initial push, ~/.gstack-brain-remote.txt for new-machine discovery, GBrain consumer registration via HTTP POST. bin/gstack-brain-restore: safe new-machine bootstrap. Refuses clobber of existing allowlisted files, clones to staging, rsync-copies tracked files, re-registers merge drivers (required — not cloned from remote), rehydrates consumers.json, prompts for per-consumer tokens. bin/gstack-brain-uninstall: clean off-ramp. Removes .git + .brain-* files + consumers.json + config keys. Preserves user data (learnings, plans, retros, profile). Optional --delete-remote for GitHub repos. bin/gstack-brain-consumer + bin/gstack-brain-reader (symlink alias): registry management. Internal 'consumer' term; user-facing 'reader' per DX review decision. |
||
|
Garry Tan
|
97cbacf409 |
feat(gbrain-sync): --once drain + secret scan + push
bin/gstack-brain-sync is the core sync binary. Subcommands: --once (drain queue, allowlist-filter, privacy-class-filter, secret-scan staged diff, commit with template, push with fetch+merge retry), --status, --skip-file <path>, --drop-queue --yes, --discover-new (cursor-based detection of artifact writes that skip the shim). Secret regex families: AWS keys, GitHub tokens (ghp_/gho_/ghu_/ghs_/ ghr_/github_pat_), OpenAI sk-, PEM blocks, JWTs, bearer-token-in-JSON. On hit: unstage, preserve queue, print remediation hint (--skip-file or edit), exit clean. No daemon — invoked by preamble at skill boundaries. |
||
|
Garry Tan
|
45638297ba |
feat(gbrain-sync): queue primitives + writer shims
Adds bin/gstack-brain-enqueue (atomic append to sync queue) and bin/gstack-jsonl-merge (git merge driver, ts-sort with SHA-256 fallback). Wires one backgrounded enqueue call into learnings-log, timeline-log, review-log, and developer-profile --migrate. question-log and question-preferences stay local per Codex v2 decision. gstack-config gains gbrain_sync_mode (off/artifacts-only/full) and gbrain_sync_mode_prompted keys, plus GSTACK_HOME env alignment so tests don't leak into real ~/.gstack/config.yaml. |
||
|
Garry Tan
|
656df0e37e |
feat(v1.5.2.0): Opus 4.7 migration — model overlay, voice, routing (#1117)
* feat(v1.5.2.0): Opus 4.7 migration — model overlay, voice, routing
Adapts GStack skill text for Claude Opus 4.7's behavioral changes per
Anthropic's migration guide and community findings.
Key changes:
model-overlays/claude.md:
- Fan out explicitly (4.7 spawns fewer subagents by default)
- Effort-match the step (avoid overthinking simple tasks at max)
- Batch questions in one AskUserQuestion turn
- Literal interpretation awareness (deliver full scope)
hosts/claude.ts:
- coAuthorTrailer updated to Claude Opus 4.7
SKILL.md.tmpl:
- Expanded routing triggers with colloquial variants ("wtf",
"this doesn't work", "send it", "where was I") — 4.7 won't
generalize from sparse trigger patterns like 4.6 did
- Added missing routes: /context-save, /context-restore, /cso, /make-pdf
- Changed routing fallback from strict "do NOT answer directly" to
"when in doubt, invoke the skill" — false positives are cheaper
than false negatives on 4.7's literal interpreter
generate-voice-directive.ts:
- Added concrete good/bad voice example — 4.7 needs shown examples,
not just described tone. "auth.ts:47 returns undefined..." vs
"I've identified a potential issue..."
Regenerated all 38 SKILL.md files. All tests pass.
* refactor(opus-4.7): split overlay, align routing, fix trailer fallback
Follow-up to wintermute's initial Opus 4.7 migration commit (addresses
ship-quality review findings before v1.6.1.0 release).
Overlay split (model-overlays/):
- Move 4 Opus-4.7-specific nudges (Fan out, Effort-match, Batch your
questions, Literal interpretation) from claude.md into new
opus-4-7.md with {{INHERIT:claude}}
- claude.md now holds only model-agnostic nudges (Todo discipline,
Think before heavy, Dedicated tools over Bash)
- Prevents Opus-4.7-specific guidance leaking onto Sonnet/Haiku
- Uses existing {{INHERIT:claude}} mechanism at
scripts/resolvers/model-overlay.ts:28-43
scripts/models.ts:
- Add opus-4-7 to ALL_MODEL_NAMES
- resolveModel: claude-opus-4-7-* variants route to opus-4-7,
all other claude-* variants continue to route to claude
scripts/resolvers/utility.ts:
- Update coAuthor trailer fallback: Opus 4.6 -> Opus 4.7
(fallback was missed in the initial migration commit)
scripts/resolvers/preamble/generate-routing-injection.ts:
- Align policy with new SKILL.md.tmpl: soft "when in doubt, invoke"
instead of hard "ALWAYS invoke... Do NOT answer directly"
- Replace stale /checkpoint reference with /context-save +
/context-restore (skills were renamed in v1.0.1.0)
- Expand route coverage to match full skill inventory:
/plan-devex-review, /qa-only, /devex-review, /land-and-deploy,
/setup-deploy, /canary, /open-gstack-browser,
/setup-browser-cookies, /benchmark, /learn, /plan-tune, /health
scripts/resolvers/preamble/generate-voice-directive.ts:
- Voice example closing: "Want me to ship it?" -> "Want me to fix it?"
- Preserves directness while routing through review gates
SKILL.md.tmpl:
- Add routing triggers for skills that were missing from the list:
/plan-devex-review, /qa-only, /devex-review, /land-and-deploy,
/setup-deploy, /canary, /open-gstack-browser,
/setup-browser-cookies, /benchmark, /learn, /plan-tune, /health
- Within Opus 4.7 overlay, added scope boundary to
"Literal interpretation" nudge ("fix tests that this branch
introduced or is responsible for")
- Added pacing exception to "Batch your questions" nudge so skills
that require one-question-at-a-time pacing still win
Follow-up commit will regenerate SKILL.md files + update goldens.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(opus-4.7): regenerate SKILL.md files + update golden fixtures
Mechanical consequence of the preceding source changes (overlay split,
routing alignment, voice example, routing expansion). No behavior change
beyond what that commit introduced.
- 36 SKILL.md files regenerated via bun run gen:skill-docs
- 3 golden fixtures updated (claude, codex, factory ship skill)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(routing): assert slash-prefixed skills + new policy + current names
Align gen-skill-docs.test.ts routing assertions with the remediated
routing-injection output:
- Expect '/office-hours' slash-prefixed form (matches SKILL.md.tmpl style)
- Add test asserting /context-save + /context-restore references
(guards against stale '/checkpoint' name regression)
- Add test asserting "When in doubt, invoke the skill" soft policy
(guards against "Do NOT answer directly" hard policy regression)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(binary-guard): replace xargs-per-file loops with fs.statSync + mode filter
The "no compiled binaries in git" describe block had two flaky tests:
- "git tracks no files larger than 2MB" timed out at 5s regularly because
it spawned one `sh -c` per tracked file via `xargs -I{}` (~571 shells
on every run, ~11s locally).
- "git tracks no Mach-O or ELF binaries" ran `file --mime-type` over every
tracked file (~3-10s, flaky near the timeout).
Both were pre-existing — not caused by any recent change — but showed up
as red in every local `bun test` run and masked legit failures in the
same suite.
Rewrites:
- 2MB test: `fs.statSync(f).size` in a filter. Millisecond-fast.
- Mach-O test: pre-filter to mode 100755 files via `git ls-files -s`,
then batch-invoke `file --mime-type` once across all executables.
With zero executables tracked, the `file` invocation is skipped.
Test suite: 320 pass, 0 fail, 907ms (was ~12.7s with 2 fails).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(team-mode): give setup -q / setup --local tests a 3-minute budget
./setup runs a full install, Bun binary build, and skill regeneration.
On a cold cache it takes 60-90s, comfortably above bun test's 5s default.
Both "setup -q produces no stdout" and "setup --local prints deprecation
warning" have been flaky-to-failing for a while with [5001.78ms] timeouts.
The test logic was fine, the budget wasn't. Bumped both to 180s via the
third-arg timeout.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(opus-4.7): E2E eval for fanout rate + routing precision
Closes the measurement gap flagged by the ship-quality review: "zero
tests exercise Opus 4.7 behavior; every skill-e2e hardcodes 4.6."
Two cases, both pinned to claude-opus-4-7:
1. Fanout rate (A/B)
- Arm A: regen SKILL.md with --model opus-4-7 (overlay ON, includes
"Fan out explicitly" nudge).
- Arm B: regen SKILL.md with --model claude (overlay OFF, only
model-agnostic nudges).
- Prompt: "Read alpha.txt, beta.txt, gamma.txt. These are independent."
- Measure: parallel tool calls in first assistant turn.
- Assert: arm A >= arm B.
2. Routing precision (6-case mini-benchmark)
- 3 positive prompts that should route (wtf bug, send it, does it work)
- 3 negative prompts that match keywords but should NOT route
(syntax question, algorithm question, slack message)
- Assert: TP rate >= 66%, FP rate <= 33%.
Cost estimate: ~$3-5 per full run. Classified as periodic tier per
CLAUDE.md convention (Opus model, non-deterministic). Runs only with
EVALS=1 env var, touchfile-gated so unrelated diffs don't trigger it.
Test plan artifact at
~/.gstack/projects/garrytan-gstack/garrytan-feat-opus-4.7-migration-eng-review-test-plan-20260421-230611.md
tracks the full specification.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* refactor(opus-4.7): rewrite fanout nudge to show parallel tool_use pattern
The original fanout nudge told 4.7 to "spawn subagents in the same turn"
and "run independent checks concurrently" in prose. An E2E eval on
claude-opus-4-7 reading 3 independent files showed zero effect: both
overlay-ON and overlay-OFF arms emitted serial Reads across 3-4 turns.
Rewrite follows the same "show not tell" principle the PR introduced for
voice examples. The nudge now includes a concrete wrong/right contrast
showing the exact tool_use structure:
Wrong (3 turns):
Turn 1: Read(foo.ts), then wait
Turn 2: Read(bar.ts), then wait
Turn 3: Read(baz.ts)
Right (1 turn, 3 parallel tool_use blocks in one assistant message):
Turn 1: [Read(foo.ts), Read(bar.ts), Read(baz.ts)]
Applies to Read, Bash, Grep, Glob, WebFetch, Agent, and any tool where
sub-calls don't depend on each other's output.
Effect on test/skill-e2e-opus-47.test.ts fanout eval: unchanged (both
arms still 0 parallel in first turn via `claude -p`). May land better in
Claude Code's interactive harness, where the system prompt + tool
handlers differ. Tracked as P0 TODO for follow-up verification in the
correct harness.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(opus-4.7): tighten ambiguous /qa routing prompt
"does this feature work on mobile? can you check the deploy?" was too
vague — a reasonable agent asks "which feature?" via AskUserQuestion
instead of routing to /qa. That's not a routing miss, it's an under-
specified prompt.
Replaced with "I just pushed the login flow changes. Test the deployed
site and find any bugs." — concrete subject + clear QA verb.
Result: pos-does-it-work went from MISS to OK, routing TP rate 2/3 -> 3/3.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(opus-4.7): rewrite scratch-root helper + add afterAll cleanup
First run of the Opus 4.7 eval exposed two test-setup gaps that made
results misleading:
- Only the root gstack SKILL.md was installed. Claude Code does
auto-discovery per-directory under .claude/skills/{name}/SKILL.md, so
without individual skill dirs the Skill tool had nothing to route to.
Positive routing cases all failed.
- `claude -p` does not load SKILL.md content as system context the way
the Claude Code harness does. The overlay nudges in SKILL.md were
invisible to the model, so the fanout A/B could not actually differ.
New `mkEvalRoot(suffix, includeOverlay)` helper, modelled on the pattern
in skill-routing-e2e.test.ts:
- Installs per-skill SKILL.md under .claude/skills/ for ~14 key skills
so the Skill tool has discoverable targets.
- Writes an explicit routing block into project CLAUDE.md.
- When includeOverlay is true, inlines the content of
model-overlays/opus-4-7.md into CLAUDE.md too. This is what makes the
fanout A/B observable in `claude -p`: arm ON gets the overlay in
context, arm OFF does not.
Plus an afterAll that re-runs gen-skill-docs at the default model so
the working tree is not left with opus-4-7-generated SKILL.md files
after the eval finishes (would break golden-file tests in the next
`bun test` run otherwise).
With this setup in place: routing went from 3/3 FAIL to 3/3 PASS
(correct skill or clarification in every positive case, zero false
positives on negatives). Fanout A/B is now a fair comparison; still
shows 0 parallel in both arms under `claude -p` (tracked as a P0 TODO
for re-measurement inside Claude Code's harness, where fanout may land
differently).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs(todos): verify Opus 4.7 fanout nudge in Claude Code harness (P0)
v1.6.1.0 shipped a rewritten "Fan out explicitly" nudge with a concrete
tool_use example. Under `claude -p` on claude-opus-4-7, the A/B eval
showed zero parallel tool calls in the first turn for both arms
(overlay ON and OFF). Routing verified 3/3 in the same harness, so the
gap is specific to fanout and likely to `claude -p`'s system prompt +
tool wiring.
This TODO closes the measurement loop the ship-quality review flagged:
re-run the fanout A/B inside Claude Code's real harness (or a faithful
replica) before landing another Opus migration claim.
P0 because it is a ship-quality commitment from the v1.6.1.0 release
notes, not a nice-to-have.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(release): v1.6.1.0 — Opus 4.7 migration, reviewed
Bump VERSION + package.json from 1.6.0.0 to 1.6.1.0. New CHANGELOG
entry describing the ship-quality remediation of PR #1117:
- Overlay split (model-agnostic claude.md + opus-4-7.md with INHERIT)
- Routing-injection aligned with SKILL.md.tmpl ("when in doubt" policy,
current skill names, full skill inventory)
- utility.ts trailer fallback updated
- Voice example closes through review gate instead of ship-bypass
- Literal-interpretation nudge bounded to branch scope
- Batch-questions nudge has explicit pacing exception
- First Opus 4.7 eval: routing verified 3/3, fanout A/B unverified
under `claude -p` (tracked as P0 TODO for next rev)
- Pre-existing test failures fixed: fs.statSync binary guard, 180s
setup timeout, golden-file updates
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(opus-4.7): key touchfile entries by testName, not describe text
TOUCHFILES completeness scan in test/touchfiles.test.ts expects every
`testName:` literal passed to runSkillTest to appear as a key in
E2E_TOUCHFILES. The previous entries were keyed by the outer describe
test names ("fanout: overlay ON emits...") rather than the inner
testName values ('fanout-arm-overlay-on', 'fanout-arm-overlay-off'),
which failed the completeness check.
Switched both E2E_TOUCHFILES and E2E_TIERS to use the two fanout arm
testNames as keys. The routing sub-tests use a template literal
(`routing-${c.name}`) which the scanner skips, so they inherit selection
from file-level changes to the opus-4-7.md / routing-injection.ts paths
already covered by the fanout entries.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: gstack <ship@gstack.dev>
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
54d4cde773 |
security: tunnel dual-listener + SSRF + envelope + path wave (v1.6.0.0) (#1137)
* refactor(security): loosen /connect rate limit from 3/min to 300/min
Setup keys are 24 random bytes (unbruteforceable), so a tight rate limit
does not meaningfully prevent key guessing. It exists only to cap
bandwidth, CPU, and log-flood damage from someone who discovered the
ngrok URL. A legitimate pair-agent session hits /connect once; 300/min
is 60x that pattern and never hit accidentally.
3/min caused pairing to fail on any retry flow (network blip, second
paired client) with no upside. Per-IP tracking was considered and
rejected — adds a bounded Map + LRU for defense already adequate at the
global layer.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(security): add tunnel-denial-log module for attack visibility
Append-only log of tunnel-surface auth denials to
~/.gstack/security/attempts.jsonl. Gives operators visibility into who
is probing tunneled daemons so the next security wave can be driven by
real attack data instead of speculation.
Design notes:
- Async via fs.promises.appendFile. Never appendFileSync — blocking the
event loop on every denial during a flood is what an attacker wants
(prior learning: sync-audit-log-io, 10/10 confidence).
- In-process rate cap at 60 writes/minute globally. Excess denials are
counted in memory but not written to disk — prevents disk DoS.
- Writes to the same ~/.gstack/security/attempts.jsonl used by the
prompt-injection attempt log. File rotation is handled by the existing
security pipeline (10MB, 5 generations).
No consumers in this commit; wired up in the dual-listener refactor that
follows.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(security): dual-listener tunnel architecture
The /health endpoint leaked AUTH_TOKEN to any caller that hit the ngrok
URL (spoofing chrome-extension:// origin, or catching headed mode).
Surfaced by @garagon in PR #1026; the original fix was header-inference
on the single port. Codex's outside-voice review during /plan-ceo-review
called that approach brittle (ngrok header behavior could change, local
proxies would false-positive), and pushed for the structural fix.
This is that fix. Stop making /health a root-token bootstrap endpoint on
any surface the tunnel can reach. The server now binds two HTTP
listeners when a tunnel is active. The local listener (extension, CLI,
sidebar) stays on 127.0.0.1 and is never exposed to ngrok. ngrok
forwards only to the tunnel listener, which serves only /connect
(unauth, rate-limited) and /command with a locked allowlist of
browser-driving commands. Security property comes from physical port
separation, not from header inference — a tunnel caller cannot reach
/health or /cookie-picker or /inspector because they live on a
different TCP socket.
What this commit adds to browse/src/server.ts:
* Surface type ('local' | 'tunnel') and TUNNEL_PATHS +
TUNNEL_COMMANDS allowlists near the top of the file.
* makeFetchHandler(surface) factory replacing the single fetch arrow;
closure-captures the surface so the filter that runs before route
dispatch knows which socket accepted the request.
* Tunnel filter at dispatch entry: 404s anything not on TUNNEL_PATHS,
403s root-token bearers with a clear pairing hint, 401s non-/connect
requests that lack a scoped token. Every denial is logged via
logTunnelDenial (from tunnel-denial-log).
* GET /connect alive probe (unauth on both surfaces) so /pair and
/tunnel/start can detect dead ngrok tunnels without reaching
/health — /health is no longer tunnel-reachable.
* Lazy tunnel listener lifecycle. /tunnel/start binds a dedicated
Bun.serve on an ephemeral port, points ngrok.forward at THAT port
(not the local port), hard-fails on bind error (no local fallback),
tears down cleanly on ngrok failure. BROWSE_TUNNEL=1 startup uses
the same pattern.
* closeTunnel() helper — single teardown path for both the ngrok
listener and the tunnel Bun.serve listener.
* resolveNgrokAuthtoken() helper — shared authtoken lookup across
/tunnel/start and BROWSE_TUNNEL=1 startup (was duplicated).
* TUNNEL_COMMANDS check in /command dispatch: on the tunnel surface,
commands outside the allowlist return 403 with a list of allowed
commands as a hint.
* Probe paths in /pair and /tunnel/start migrated from /health to
GET /connect — the only unauth path reachable on the tunnel surface
under the new architecture.
Test updates in browse/test/server-auth.test.ts:
* /pair liveness-verify test: assert via closeTunnel() helper instead
of the inline `tunnelActive = false; tunnelUrl = null` lines that
the helper subsumes.
* /tunnel/start cached-tunnel test: same closeTunnel() adaptation.
Credit
Derived from PR #1026 by @garagon — thanks for flagging the critical
bug that drove the architectural rewrite. The per-request
isTunneledRequest approach from #1026 is superseded by physical port
separation here; the underlying report remains the root cause for the
entire v1.6.0.0 wave.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(security): add source-level guards for dual-listener architecture
23 source-level assertions that keep future contributors from silently
widening the tunnel surface during a routine refactor. Covers:
* Surface type + tunnelServer state variable shape
* TUNNEL_PATHS is a closed set of /connect, /command, /sidebar-chat
(and NOT /health, /welcome, /cookie-picker, /inspector/*, /pair,
/token, /refs, /activity/stream, /tunnel/{start,stop})
* TUNNEL_COMMANDS includes browser-driving ops only (and NOT
launch-browser, tunnel-start, token-mint, cookie-import, etc.)
* makeFetchHandler(surface) factory exists and is wired to both
listeners with the correct surface parameter
* Tunnel filter runs BEFORE any route dispatch, with 404/403/401
responses and logged denials for each reason
* GET /connect returns {alive: true} unauth
* /command dispatch enforces TUNNEL_COMMANDS on tunnel surface
* closeTunnel() helper tears down ngrok + Bun.serve listener
* /tunnel/start binds on ephemeral port, points ngrok at TUNNEL_PORT
(not local port), hard-fails on bind error (no fallback), probes
cached tunnel via GET /connect (not /health), tears down on
ngrok.forward failure
* BROWSE_TUNNEL=1 startup uses the dual-listener pattern
* logTunnelDenial wired for all three denial reasons
* /connect rate limit is 300/min, not 3/min
All 23 tests pass. Behavioral integration tests (spawn subprocess, real
network) live in the E2E suite that lands later in this wave.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* security: gate download + scrape through validateNavigationUrl (SSRF)
The `goto` command was correctly wired through validateNavigationUrl,
but `download` and `scrape` called page.request.fetch(url, ...) directly.
A caller with the default write scope could hit the /command endpoint
and ask the daemon to fetch http://169.254.169.254/latest/meta-data/
(AWS IMDSv1) or the GCP/Azure/internal equivalents. The response body
comes back as base64 or lands on disk where GET /file serves it.
Fix: call validateNavigationUrl(url) immediately before each
page.request.fetch() call site in download and in the scrape loop.
Same blocklist that already protects `goto`: file://, javascript:,
data:, chrome://, cloud metadata (IPv4 all encodings, IPv6 ULA,
metadata.*.internal).
Tests: extend browse/test/url-validation.test.ts with a source-level
guard that walks every `await page.request.fetch(` call site and
asserts a validateNavigationUrl call precedes it within the same
branch. Regression trips before code review if a future refactor
drops the gate.
* security: route splitForScoped through envelope sentinel escape
The scoped-token snapshot path in snapshot.ts built its untrusted
block by pushing the raw accessibility-tree lines between the literal
`═══ BEGIN UNTRUSTED WEB CONTENT ═══` / `═══ END UNTRUSTED WEB CONTENT ═══`
sentinels. The full-page wrap path in content-security.ts already
applied a zero-width-space escape on those exact strings to prevent
sentinel injection, but the scoped path skipped it.
Net effect: a page whose rendered text contains the literal sentinel
can close the envelope early from inside untrusted content and forge
a fake "trusted" block for the LLM. That includes fabricating
interactive `@eN` references the agent will act on.
Fix:
* Extract the zero-width-space escape into a named, exported helper
`escapeEnvelopeSentinels(content)` in content-security.ts.
* Have `wrapUntrustedPageContent` call it (behavior unchanged on
that path — same bytes out).
* Import the helper in snapshot.ts and map it over `untrustedLines`
in the `splitForScoped` branch before pushing the BEGIN sentinel.
Tests: add a describe block in content-security.test.ts that covers
* `escapeEnvelopeSentinels` defuses BEGIN and END markers;
* `escapeEnvelopeSentinels` leaves normal text untouched;
* `wrapUntrustedPageContent` still emits exactly one real envelope
pair when hostile content contains forged sentinels;
* snapshot.ts imports the helper;
* the scoped-snapshot branch calls `escapeEnvelopeSentinels` before
pushing the BEGIN sentinel (source-level regression — if a future
refactor reorders this, the test trips).
* security: extend hidden-element detection to all DOM-reading channels
The Confusion Protocol envelope wrap (`wrapUntrustedPageContent`)
covers every scoped PAGE_CONTENT_COMMAND, but the hidden-element
ARIA-injection detection layer only ran for `text`. Other DOM-reading
channels (html, links, forms, accessibility, attrs, data, media,
ux-audit) returned their output through the envelope with no hidden-
content filter, so a page serving a display:none div that instructs
the agent to disregard prior system messages, or an aria-label that
claims to put the LLM in admin mode, leaked the injection payload on
any non-text channel. The envelope alone does not mitigate this, and
the page itself never rendered the hostile content to the human
operator.
Fix:
* New export `DOM_CONTENT_COMMANDS` in commands.ts — the subset of
PAGE_CONTENT_COMMANDS that derives its output from the live DOM.
Console and dialog stay out; they read separate runtime state.
* server.ts runs `markHiddenElements` + `cleanupHiddenMarkers` for
every scoped command in this set. `text` keeps its existing
`getCleanTextWithStripping` path (hidden elements physically
stripped before the read). All other channels keep their output
format but emit flagged elements as CONTENT WARNINGS on the
envelope, so the LLM sees what it would otherwise have consumed
silently.
* Hidden-element descriptions merge into `combinedWarnings`
alongside content-filter warnings before the wrap call.
Tests: new describe block in content-security.test.ts covering
* `DOM_CONTENT_COMMANDS` export shape and channel membership;
* dispatch gates on `DOM_CONTENT_COMMANDS.has(command)`, not the
literal `text` string;
* hiddenContentWarnings plumbs into `combinedWarnings` and reaches
wrapUntrustedPageContent;
* DOM_CONTENT_COMMANDS is a strict subset of PAGE_CONTENT_COMMANDS.
Existing datamarking, envelope wrap, centralized-wrapping, and chain
security suites stay green (52 pass, 0 fail).
* security: validate --from-file payload paths for parity with direct paths
The direct `load-html <file>` path runs every caller-supplied file path
through validateReadPath() so reads stay confined to SAFE_DIRECTORIES
(cwd, TEMP_DIR). The `load-html --from-file <payload.json>` shortcut
and its sibling `pdf --from-file <payload.json>` skipped that check and
went straight to fs.readFileSync(). An MCP caller that picks the
payload path (or any caller whose payload argument is reachable from
attacker-influenced text) could use --from-file as a read-anywhere
escape hatch for the safe-dirs policy.
Fix: call validateReadPath(path.resolve(payloadPath)) before readFileSync
at both sites. Error surface mirrors the direct-path branch so ops and
agent errors stay consistent.
Test coverage in browse/test/from-file-path-validation.test.ts:
- source-level: validateReadPath precedes readFileSync in the load-html
--from-file branch (write-commands.ts) and the pdf --from-file parser
(meta-commands.ts)
- error-message parity: both sites reference SAFE_DIRECTORIES
Related security audit pattern: R3 F002 (validateNavigationUrl gap on
download/scrape) and R3 F008 (markHiddenElements gap on 10 DOM commands)
were the same shape — a defense that existed on the primary code path
but not its shortcut sibling. This PR closes the same class of gap on
the --from-file shortcuts.
* fix(design): escape url.origin when injecting into served HTML
serve.ts injected url.origin into a single-quoted JS string in
the response body. A local request with a crafted Host header
(e.g. Host: "evil'-alert(1)-'x") would break out of the string
and execute JS in the 127.0.0.1:<port> origin opened by the
design board. Low severity — bound to localhost, requires a
local attacker — but no reason not to escape.
Fix: JSON.stringify(url.origin) produces a properly quoted,
escaped JS string literal in one call.
Also includes Prettier reformatting (single→double quotes,
trailing commas, line wrapping) applied by the repo's
PostToolUse formatter hook. Security change is the one line
in the HTML injection; everything else is whitespace/style.
* fix(scripts): drop shell:true from slop-diff npx invocations
spawnSync('npx', [...], { shell: true }) invokes /bin/sh -c
with the args concatenated, subjecting them to shell parsing
(word splitting, glob expansion, metacharacter interpretation).
No user input reaches these calls today, so not exploitable —
but the posture is wrong: npx + shell args should be direct.
Fix: scope shell:true to process.platform === 'win32' where
npx is actually a .cmd requiring the shell. POSIX runs the
npx binary directly with array-form args.
Also includes Prettier reformatting (single→double quotes,
trailing commas, line wrapping) applied by the repo's
PostToolUse formatter hook. Security-relevant change is just
the two shell:true -> shell: process.platform === 'win32'
lines; everything else is whitespace/style.
* security(E3): gate GSTACK_SLUG on /welcome path traversal
The /welcome handler interpolates GSTACK_SLUG directly into the filesystem
path used to locate the project-local welcome page. Without validation, a
slug like "../../etc/passwd" would resolve to
~/.gstack/projects/../../etc/passwd/designs/welcome-page-20260331/finalized.html
— classic path traversal.
Not exploitable today: GSTACK_SLUG is set by the gstack CLI at daemon launch,
and an attacker would already need local env-var access to poison it. But
the gate is one regex (^[a-z0-9_-]+$), and a defense-in-depth pass costs us
nothing when the cost of being wrong is arbitrary file read via /welcome.
Fall back to the safe 'unknown' literal when the slug fails validation —
same fallback the code already uses when GSTACK_SLUG is unset. No behavior
change for legitimate slugs (they all match the regex).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* security(N1): replace ?token= SSE auth with HttpOnly session cookie
Activity stream and inspector events SSE endpoints accepted the root
AUTH_TOKEN via `?token=` query param (EventSource can't send Authorization
headers). URLs leak to browser history, referer headers, server logs,
crash reports, and refactoring accidents. Codex flagged this during the
/plan-ceo-review outside voice pass.
New auth model: the extension calls POST /sse-session with a Bearer token
and receives a view-only session cookie (HttpOnly, SameSite=Strict, 30-min
TTL). EventSource is opened with `withCredentials: true` so the browser
sends the cookie back on the SSE connection. The ?token= query param is
GONE — no more URL-borne secrets.
Scope isolation (prior learning cookie-picker-auth-isolation, 10/10
confidence): the SSE session cookie grants access to /activity/stream and
/inspector/events ONLY. The token is never valid against /command, /token,
or any mutating endpoint. A leaked cookie can watch activity; it cannot
execute browser commands.
Components
* browse/src/sse-session-cookie.ts — registry: mint/validate/extract/
build-cookie. 256-bit tokens, 30-min TTL, lazy expiry pruning,
no imports from token-registry (scope isolation enforced by module
boundary).
* browse/src/server.ts — POST /sse-session mint endpoint (requires
Bearer). /activity/stream and /inspector/events now accept Bearer
OR the session cookie, and reject ?token= query param.
* extension/sidepanel.js — ensureSseSessionCookie() bootstrap call,
EventSource opened with withCredentials:true on both SSE endpoints.
Tested via the source guards; behavioral test is the E2E pairing
flow that lands later in the wave.
* browse/test/sse-session-cookie.test.ts — 20 unit tests covering
mint entropy, TTL enforcement, cookie flag invariants, cookie
parsing from multi-cookie headers, and scope-isolation contract
guard (module must not import token-registry).
* browse/test/server-auth.test.ts — existing /activity/stream auth
test updated to assert the new cookie-based gate and the absence
of the ?token= query param.
Cookie flag choices:
* HttpOnly: token not readable from page JS (mitigates XSS
exfiltration).
* SameSite=Strict: cookie not sent on cross-site requests (mitigates
CSRF). Fine for SSE because the extension connects to 127.0.0.1
directly.
* Path=/: cookie scoped to the whole origin.
* Max-Age=1800: 30 minutes, matches TTL. Extension re-mints on
reconnect when daemon restarts.
* Secure NOT set: daemon binds to 127.0.0.1 over plain HTTP. Adding
Secure would block the browser from ever sending the cookie back.
Add Secure when gstack ships over HTTPS.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* security(N2): document Windows v20 ABE elevation path on CDP port
The existing comment around the cookie-import-browser --remote-debugging-port
launch claimed "threat model: no worse than baseline." That's wrong on
Windows with App-Bound Encryption v20. A same-user local process that
opens the cookie SQLite DB directly CANNOT decrypt v20 values (DPAPI
context is bound to the browser process). The CDP port lets them bypass
that: connect to the debug port, call Network.getAllCookies inside Chrome,
walk away with decrypted v20 cookies.
The correct fix is to switch from TCP --remote-debugging-port to
--remote-debugging-pipe so the CDP transport is a stdio pipe, not a
socket. That requires restructuring the CDP WebSocket client in this
module and Playwright doesn't expose the pipe transport out of the box.
Non-trivial, deferred from the v1.6.0.0 wave.
This commit updates the comment to correctly describe the threat and
points at the tracking issue. No code change to the launch itself.
Follow-up: #1136.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs(E2): document dual-listener tunnel architecture in ARCHITECTURE.md
Adds an explicit per-endpoint disposition table to the Security model
section, covering the v1.6.0.0 dual-listener refactor. Every HTTP
endpoint now has a documented local-vs-tunnel answer. Future audits
(and future contributors wondering "is it safe to add X to the tunnel
surface?") can read this instead of reverse-engineering server.ts.
Also documents:
* Why physical port separation beats per-request header inference
(ngrok behavior drift, local proxies can forge headers, etc.)
* Tunnel surface denial logging → ~/.gstack/security/attempts.jsonl
* SSE session cookie model (gstack_sse, 30-min TTL, stream-scope only,
module-boundary-enforced scope isolation)
* N2 non-goal for Windows v20 ABE via CDP port (tracking #1136)
No code changes.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test(E1): end-to-end pair-agent flow against a spawned daemon
Spawns the browse daemon as a subprocess with BROWSE_HEADLESS_SKIP=1 so
the HTTP layer runs without a real browser. Exercises:
* GET /health — token delivery for chrome-extension origin, withheld
otherwise (the F1 + PR #1026 invariant)
* GET /connect — alive probe returns {alive:true} unauth
* POST /pair — root Bearer required (403 without), returns setup_key
* POST /connect — setup_key exchange mints a distinct scoped token
* POST /command — 401 without auth
* POST /sse-session — Bearer required, Set-Cookie has HttpOnly +
SameSite=Strict (the N1 invariant)
* GET /activity/stream — 401 without auth
* GET /activity/stream?token= — 401 (the old ?token= query param is
REJECTED, which is the whole point of N1)
* GET /welcome — serves HTML, does not leak /etc/passwd content under
the default 'unknown' slug (E3 regex gate)
12 behavioral tests, ~220ms end-to-end, no network dependencies, no
ngrok, no real browser. This is the receipt for the wave's central
'pair-agent still works + the security boundary holds' claim.
Tunnel-port binding (/tunnel/start) is deliberately NOT exercised here
— it requires an ngrok authtoken and live network. The dual-listener
route allowlist is covered by source-level guards in
dual-listener.test.ts; behavioral tunnel testing belongs in a separate
paid-evals harness.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* release(v1.6.0.0): bump VERSION + CHANGELOG for security wave
Architectural bump, not patch: dual-listener HTTP refactor changes the
daemon's tunnel-exposure model. See CHANGELOG for the full release
summary (~950 words) covering the five root causes this wave closes:
1. /health token leak over ngrok (F1 + E3 + test infra)
2. /cookie-picker + /inspector exposed over the tunnel (F1)
3. ?token=<ROOT> in SSE URLs leaking to logs/referer/history (N1)
4. /welcome GSTACK_SLUG path traversal (E3)
5. Windows v20 ABE elevation via CDP port (N2 — documented non-goal,
tracked as #1136)
Plus the base PRs: SSRF gate (#1029), envelope sentinel escape (#1031),
DOM-channel hidden-element coverage (#1032), --from-file path validation
(#1103), and 2 commits from #1073 (@theqazi).
VERSION + package.json bumped to 1.6.0.0. CHANGELOG entry covers
credits (@garagon, @Hybirdss, @HMAKT99, @theqazi), review lineage (CEO
→ Codex outside voice → Eng), and the non-goal tracking issue.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix: pre-landing review findings (4 auto-fixes)
Addresses 4 findings from the Claude adversarial subagent on the
v1.6.0.0 security wave diff. No user-visible behavior change; all
are defense-in-depth hardening of newly-introduced code.
1. GET /connect rate-limited (was POST-only) [HIGH conf 8/10]
Attacker discovering the ngrok URL could probe unlimited GETs for
daemon enumeration. Now shares the global /connect counter.
2. ngrok listener leak on tunnel startup failure [MEDIUM conf 8/10]
If ngrok.forward() resolved but tunnelListener.url() or the
state-file write threw, the Bun listener was torn down but the
ngrok session was leaked. Fixed in BOTH /tunnel/start and
BROWSE_TUNNEL=1 startup paths.
3. GSTACK_SKILL_ROOT path-traversal gate [MEDIUM conf 8/10]
Symmetric with E3's GSTACK_SLUG regex gate — reject values
containing '..' before interpolating into the welcome-page path.
4. SSE session registry pruning [LOW conf 7/10]
pruneExpired() only checked 10 entries per mint call. Now runs
on every validate too, checks 20 entries, with a hard 10k cap as
backstop. Prevents registry growth under sustained extension
reconnect pressure.
Tests remain green (56/56 in sse-session-cookie + dual-listener +
pair-agent-e2e suites).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs: update project documentation for v1.6.0.0
Reflect the dual-listener tunnel architecture, SSE session cookies,
SSRF guards, and Windows v20 ABE non-goal across the three docs
users actually read for remote-agent and browser auth context:
- docs/REMOTE_BROWSER_ACCESS.md: rewrote Architecture diagram for
dual listeners, fixed /connect rate limit (3/min → 300/min),
removed stale "/health requires no auth" (now 404 on tunnel),
added SSE cookie auth, expanded Security Model with tunnel
allowlist, SSRF guards, /welcome path traversal defense, and
the Windows v20 ABE tracking note.
- BROWSER.md: added dual-listener paragraph to Authentication and
linked to ARCHITECTURE.md endpoint table. Replaced the stale
?token= SSE auth note with the HttpOnly gstack_sse cookie flow.
- CLAUDE.md: added Transport-layer security section above the
sidebar prompt-injection stack so contributors editing server.ts,
sse-session-cookie.ts, or tunnel-denial-log.ts see the load-bearing
module boundaries before touching them.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(make-pdf): write --from-file payload to /tmp, not os.tmpdir()
make-pdf's browseClient wrote its --from-file payload to os.tmpdir(),
which is /var/folders/... on macOS. v1.6.0.0's PR #1103 cherry-pick
tightened browse load-html --from-file to validate against the
safe-dirs allowlist ([TEMP_DIR, cwd] where TEMP_DIR is '/tmp' on
macOS/Linux, os.tmpdir() on Windows). This closed a CLI/API parity
gap but broke make-pdf on macOS because /var/folders/... is outside
the allowlist.
Fix: mirror browse's TEMP_DIR convention — use '/tmp' on non-Windows,
os.tmpdir() on Windows. The make-pdf-gate CI failure on macOS-latest
(run 72440797490) is caused by exactly this: the payload file was
rejected by validateReadPath.
Verified locally: the combined-gate e2e test now passes after
rebuilding make-pdf/dist/pdf.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(sidebar): killAgent resets per-tab state; align tests with current agent event format
Two pre-existing bugs surfaced while running the full e2e suite on the
sec-wave branch. Both pre-date v1.6.0.0 (same failures on main at
|
||
|
Garry Tan
|
e23ff280a1 |
fix(v1.4.1.0): /make-pdf — page numbers, entity escape, Linux fonts (#1098)
* fix(make-pdf): single-source page numbers via CSS, honor --no-page-numbers end-to-end
Two page-number sources were stacking in every PDF: Chromium's native footer
and our @page @bottom-center CSS. The CLI flag --page-numbers/--no-page-numbers
also never reached the CSS layer, because RenderOptions didn't carry it.
Passing --footer-template likewise dropped the "custom footer replaces stock
footer" semantic.
- orchestrator.ts: browseClient.pdf() gets pageNumbers:false unconditionally.
CSS is the single source of truth. Chromium native numbering always off.
- render.ts: RenderOptions gains pageNumbers + footerTemplate. render() computes
showPageNumbers = pageNumbers !== false && !footerTemplate and passes to
printCss(), preserving the prior footerTemplate-suppresses-stock semantic.
- print-css.ts: PrintCssOptions.pageNumbers wraps @bottom-center in a conditional
matching the existing showConfidential pattern.
- types.ts: PreviewOptions.pageNumbers so preview path compiles and matches CLI.
- render.test.ts: 7 regression tests covering printCss({pageNumbers}) in
isolation AND the full render() data flow incl. footerTemplate path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(make-pdf): decode HTML entities in titles and TOC to prevent double-escape
A markdown title like "# Herbert & Garry" rendered as "Herbert &amp; Garry"
in <title>, cover block, and TOC entries. marked emits "&" (correct HTML),
but extractFirstHeading and extractHeadings only stripTags — leaving the entity
intact. That string then flows through escapeHtml, producing the double-encode.
- render.ts: new decodeTextEntities helper, distinct from decodeTypographicEntities
(which runs on in-pipeline HTML and intentionally preserves &). Covers
named entities (lt/gt/quot/apos/39/x27/amp) AND numeric (decimal + hex) so
inputs like "©" or "—" don't create the same partial-fix bug.
Amp-last ordering prevents double-decode on "&lt;" et al.
- Apply in both extractFirstHeading and extractHeadings. extractHeadings feeds
buildTocBlock → escapeHtml, so the TOC site had the same bug.
- render.test.ts: 8 tests covering the contract — parameterized across &, <, >,
©, — chars; single-escape in <title>/cover; TOC double-escape check; numeric
entity decode; smartypants-interacts-with-quotes contract (no raw equality).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(make-pdf): Liberation Sans font fallback for Linux rendering
On Linux (Docker, CI, servers), neither Helvetica nor Arial exist. Our CSS
stacks were falling through to DejaVu Sans — wider letterforms that look like
Verdana, not the intended Helvetica/Faber look. Liberation Sans is the standard
metric-compatible Arial clone (SIL OFL 1.1, apt package fonts-liberation).
- print-css.ts: all four font stacks (body + @top-center + @bottom-center +
@bottom-right CONFIDENTIAL) gain "Liberation Sans" between Helvetica and
Arial. File-header docblock updated to reflect the new stack.
- .github/docker/Dockerfile.ci: explicit apt-get install fonts-liberation +
fontconfig with retry, fc-cache -f, and a verify step that fails the build
loud if the font disappears. Playwright's install-deps happens to pull this
in today but the dep is implicit and could silently regress.
- SKILL.md.tmpl: one-sentence note pointing Linux users at fonts-liberation.
- SKILL.md: regenerated via bun run gen:skill-docs --host all (only make-pdf's
generated file changed — verified clean diff scope).
- render.test.ts: 2 assertions — Liberation Sans in body stack AND in at least
one @page margin-box rule (proves all four intended stacks got touched, not
just one).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: bump version and changelog (v1.4.1.0)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: anonymize test fixtures, drop VC-partner framing
- CHANGELOG + render.test.ts fixtures use "Faber & Faber" instead of a
personal name. Same regression coverage (ampersand in <title>, cover,
TOC, body), neutral subject.
- make-pdf/SKILL.md.tmpl description drops the "send to a VC partner, a
book agent, a judge, or Rick Rubin's team" line. "Not a draft artifact
— a finished artifact" stands on its own without the audience posturing.
- SKILL.md regenerated.
No functional changes. All 58 make-pdf tests still pass.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
97584f9a59 |
feat(security): ML prompt injection defense for sidebar (v1.4.0.0) (#1089)
* chore(deps): add @huggingface/transformers for prompt injection classifier Dependency needed for the ML prompt injection defense layer coming in the follow-up commits. @huggingface/transformers will host the TestSavantAI BERT-small classifier that scans tool outputs for indirect prompt injection. Note: this dep only runs in non-compiled bun contexts (sidebar-agent.ts). The compiled browse binary cannot load it because transformers.js v4 requires onnxruntime-node (native module, fails to dlopen from bun compile's temp extract dir). See docs/designs/ML_PROMPT_INJECTION_KILLER.md for the full architectural decision. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): add security.ts foundation for prompt injection defense Establishes the module structure for the L5 canary and L6 verdict aggregation layers. Pure-string operations only — safe to import from the compiled browse binary. Includes: * THRESHOLDS constants (BLOCK 0.85 / WARN 0.60 / LOG_ONLY 0.40), calibrated against BrowseSafe-Bench smoke + developer content benign corpus. * combineVerdict() implementing the ensemble rule: BLOCK only when the ML content classifier AND the transcript classifier both score >= WARN. Single-layer high confidence degrades to WARN to prevent any one classifier's false-positives from killing sessions (Stack Overflow instruction-writing-style FPs at 0.99 on TestSavantAI alone). * generateCanary / injectCanary / checkCanaryInStructure — session-scoped secret token, recursively scans tool arguments, URLs, file writes, and nested objects per the plan's all-channel coverage decision. * logAttempt with 10MB rotation (keeps 5 generations). Salted SHA-256 hash, per-device salt at ~/.gstack/security/device-salt (0600). * Cross-process session state at ~/.gstack/security/session-state.json (atomic temp+rename). Required because server.ts (compiled) and sidebar-agent.ts (non-compiled) are separate processes. * getStatus() for shield icon rendering via /health. ML classifier code will live in a separate module (security-classifier.ts) loaded only by sidebar-agent.ts — compiled browse binary cannot load the native ONNX runtime. Plan: ~/.gstack/projects/garrytan-gstack/ceo-plans/2026-04-19-prompt-injection-guard.md Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): wire canary injection into sidebar spawnClaude Every sidebar message now gets a fresh CANARY-XXXXXXXXXXXX token embedded in the system prompt with an instruction for Claude to never output it on any channel. The token flows through the queue entry so sidebar-agent.ts can check every outbound operation for leaks. If Claude echoes the canary into any outbound channel (text stream, tool arguments, URLs, file write paths), the sidebar-agent terminates the session and the user sees the approved canary leak banner. This operation is pure string manipulation — safe in the compiled browse binary. The actual output-stream check (which also has to be safe in compiled contexts) lives in sidebar-agent.ts (next commit). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(security): make sidebar-agent destructure check regex-tolerant The test asserted the exact string `const { prompt, args, stateFile, cwd, tabId } = queueEntry` which breaks whenever security or other extensions add fields (canary, pageUrl, etc.). Switch to a regex that requires the core fields in order but tolerates additional fields in between. Preserves the test's intent (args come from the queue entry, not rebuilt) while allowing the destructure to grow. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): canary leak check across all outbound channels The sidebar-agent now scans every Claude stream event for the session's canary token before relaying any data to the sidepanel. Channels covered (per CEO review cross-model tension #2): * Assistant text blocks * Assistant text_delta streaming * tool_use arguments (recursively, via checkCanaryInStructure — catches URLs, commands, file paths nested at any depth) * tool_use content_block_start * tool_input_delta partial JSON * Final result payload If the canary leaks on any channel, onCanaryLeaked() fires once per session: 1. logAttempt() writes the event to ~/.gstack/security/attempts.jsonl with the canary's salted hash (never the payload content). 2. sends a `security_event` to the sidepanel so it can render the approved canary-leak banner (variant A mockup — ceo-plan 2026-04-19). 3. sends an `agent_error` for backward-compat with existing error surfaces. 4. SIGTERM's the claude subprocess (SIGKILL after 2s if still alive). The leaked content itself is never relayed to the sidepanel — the event is dropped at the boundary. Canary detection is pure-string substring match, so this all runs safely in the sidebar-agent (non-compiled bun) context. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): add security-classifier.ts with TestSavantAI + Haiku This module holds the ML classifier code that the compiled browse binary cannot link (onnxruntime-node native dylib doesn't load from Bun compile's temp extract dir — see CEO plan §"Pre-Impl Gate 1 Outcome"). It's imported ONLY by sidebar-agent.ts, which runs as a non-compiled bun script. Two layers: L4 testsavant_content — TestSavantAI BERT-small ONNX classifier. First call triggers a one-time 112MB model download to ~/.gstack/models/testsavant-small/ (files staged into the onnx/ layout transformers.js v4 expects). Classifies page snapshots and tool outputs for indirect prompt injection + jailbreak attempts. On benign-corpus dry-run: Wikipedia/HN/Reddit/tech-blog all score SAFE 0.98+, attack text scores INJECTION 0.99+, Stack Overflow instruction-writing now scores SAFE 0.98 on the shorter form (was 0.99 INJECTION on the longer form — instruction-density threshold). Ensemble combiner downgrades single-layer high to WARN to cover this case. L4b transcript_classifier — Claude Haiku reasoning-blind pre-tool-call scan. Sees only {user_message, last 3 tool_calls}, never Claude's chain-of-thought or tool results (those are how self-persuasion attacks leak). 2000ms hard timeout. Fail-open on any subprocess failure so sidebar stays functional. Gated by shouldRunTranscriptCheck() — only runs when another layer already fired at >= LOG_ONLY, saving ~70% of Haiku spend. Both layers degrade gracefully: load/spawn failures set status to 'degraded' and return confidence=0. Shield icon reflects this via getClassifierStatus() which security.ts's getStatus() composes. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): wire TestSavantAI + ensemble into sidebar-agent pre-spawn scan The sidebar-agent now runs a ML security check on the user message BEFORE spawning claude. If the content classifier and (gated) transcript classifier ensemble returns BLOCK, the session is refused with a security_event + agent_error — the sidepanel renders the approved banner. Two pieces: 1. On agent startup, loadTestsavant() warms the classifier in the background. First run triggers a 112MB model download from HuggingFace (~30s on average broadband). Non-blocking — sidebar stays functional during cold-start, shield just reports 'off' until warmed. 2. preSpawnSecurityCheck() runs the ensemble against the user message: - L4 (testsavant_content) always runs - L4b (transcript_classifier via Haiku) runs only if L4 flagged at >= LOG_ONLY — plan §E1 gating optimization, saves ~70% of Haiku spend combineVerdict() applies the BLOCK-requires-both-layers rule, which downgrades any single-layer high confidence to WARN. Stack Overflow-style instruction-heavy writing false-positives on TestSavantAI alone are caught by this degrade — Haiku corrects them when called. Fail-open everywhere: any subprocess/load/inference error returns confidence=0 so the sidebar keeps working on architectural controls alone. Shield icon reflects degraded state via getClassifierStatus(). BLOCK path emits both: - security_event {verdict, reason, layer, confidence, domain} (for the approved canary-leak banner UX mockup — variant A) - agent_error "Session blocked — prompt injection detected..." (backward-compat with existing error surface) Regression test suite still passes (12/12 sidebar-security tests). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(security): add security.ts unit tests (25 tests, 62 assertions) Covers the pure-string operations that must behave deterministically in both compiled and source-mode bun contexts: * THRESHOLDS ordering invariant (BLOCK > WARN > LOG_ONLY > 0) * combineVerdict ensemble rule — THE critical path: - Empty signals → safe - Canary leak always blocks (regardless of ML signals) - Both ML layers >= WARN → BLOCK (ensemble_agreement) - Single layer >= BLOCK → WARN (single_layer_high) — the Stack Overflow FP mitigation that prevents one classifier killing sessions alone - Max-across-duplicates when multiple signals reference the same layer * Canary generation + injection + recursive checking: - Unique CANARY-XXXXXXXXXXXX tokens (>= 48 bits entropy) - Recursive structure scan for tool_use inputs, nested URLs, commands - Null / primitive handling doesn't throw * Payload hashing (salted sha256) — deterministic per-device, differs across payloads, 64-char hex shape * logAttempt writes to ~/.gstack/security/attempts.jsonl * writeSessionState + readSessionState round-trip (cross-process) * getStatus returns valid SecurityStatus shape * extractDomain returns hostname only, empty string on bad input All 25 tests pass in 18ms — no ML, no network, no subprocess spawning. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): expose security status on /health for shield icon The /health endpoint now returns a `security` field with the classifier status, suitable for driving the sidepanel shield icon: { status: 'protected' | 'degraded' | 'inactive', layers: { testsavant, transcript, canary }, lastUpdated: ISO8601 } Backend plumbing: * server.ts imports getStatus from security.ts (pure-string, safe in compiled binary) and includes it in the /health response. * sidebar-agent.ts writes ~/.gstack/security/session-state.json when the classifier warmup completes (success OR failure). This is the cross- process handoff — server.ts reads the state file via getStatus() to surface the result to the sidepanel. The sidepanel rendering (SVG shield icon + color states + tooltip) is a follow-up commit in the extension/ code. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(security): document the sidebar security stack in CLAUDE.md Adds a security section to the Browser interaction block. Covers: * Layered defense table showing which modules live where (content-security.ts in both contexts vs security-classifier.ts only in sidebar-agent) and why the split exists (onnxruntime-node incompatibility with compiled Bun) * Threshold constants (0.85 / 0.60 / 0.40) and the ensemble rule that prevents single-classifier false-positives (the Stack Overflow FP story) * Env knobs — GSTACK_SECURITY_OFF kill switch, cache paths, salt file, attack log rotation, session state file This is the "before you modify the security stack, read this" doc. It lives next to the existing Sidebar architecture note that points at SIDEBAR_MESSAGE_FLOW.md. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(todos): mark ML classifier v1 in-progress + file v2 follow-ups Reframes the P0 item to reflect v1 scope (branch 2 architecture, TestSavantAI pivot, what shipped) and splits v2 work into discrete TODOs: * Shield icon + canary leak banner UI (P0, blocks v1 user-facing completion) * Attack telemetry via gstack-telemetry-log (P1) * Full BrowseSafe-Bench at gate tier (P2) * Cross-user aggregate attack dashboard (P2) * DeBERTa-v3 as third signal in ensemble (P2) * Read/Glob/Grep ingress coverage (P2, flagged by Codex review) * Adversarial + integration + smoke-bench test suites (P1) * Bun-native 5ms inference (P3 research) Each TODO carries What / Why / Context / Effort / Priority / Depends-on so it's actionable by someone picking it up cold. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(telemetry): add attack_attempt event type to gstack-telemetry-log Extends the existing telemetry pipe with 5 new flags needed for prompt injection attack reporting: --url-domain hostname only (never path, never query) --payload-hash salted sha256 hex (opaque — no payload content ever) --confidence 0-1 (awk-validated + clamped; malformed → null) --layer testsavant_content | transcript_classifier | aria_regex | canary --verdict block | warn | log_only Backward compatibility: * Existing skill_run events still work — all new fields default to null * Event schema is a superset of the old one; downstream edge function can filter by event_type No new auth, no new SDK, no new Supabase migration. The same tier gating (community → upload, anonymous → local only, off → no-op) and the same sync daemon carry the attack events. This is the "E6 RESOLVED" path from the CEO plan — riding the existing pipe instead of spinning up parallel infra. Verified end-to-end: * attack_attempt event with all fields emits correctly to skill-usage.jsonl * skill_run event with no security flags still works (backward compat) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(security): wire logAttempt to gstack-telemetry-log (fire-and-forget) Every local attempt.jsonl write now also triggers a subprocess call to gstack-telemetry-log with the attack_attempt event type. The binary handles tier gating internally (community → Supabase upload, anonymous → local JSONL only, off → no-op), so security.ts doesn't need to re-check. Binary resolution follows the skill preamble pattern — never relies on PATH, which breaks in compiled-binary contexts: 1. ~/.claude/skills/gstack/bin/gstack-telemetry-log (global install) 2. .claude/skills/gstack/bin/gstack-telemetry-log (symlinked dev) 3. bin/gstack-telemetry-log (in-repo dev) Fire-and-forget: * spawn with stdio: 'ignore', detached: true, unref() * .on('error') swallows failures * Missing binary is non-fatal — local attempts.jsonl still gives audit trail Never throws. Never blocks. Existing 37 security tests pass unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(ui): add security banner markup + styles (approved variant A) HTML + CSS for the canary leak / ML block banner. Structure matches the approved mockup from /plan-design-review 2026-04-19 (variant A — centered alert-heavy): * Red alert-circle SVG icon (no stock shield, intentional — matches the "serious but not scary" tone the review chose) * "Session terminated" Satoshi Bold 18px red headline * "— prompt injection detected from {domain}" DM Sans zinc subtitle * Expandable "What happened" chevron button (aria-expanded/aria-controls) * Layer list rendered in JetBrains Mono with amber tabular-nums scores * Close X in top-right, 28px hit area, focus-visible amber outline Enter animation: slide-down 8px + fade, 250ms, cubic-bezier(0.16,1,0.3,1) — matches DESIGN.md motion spec. Respects `role="alert"` + `aria-live="assertive"` so screen readers announce on appearance. Escape-to-dismiss hook is in the JS follow-up commit. Design tokens all via CSS variables (--error, --amber-400, --amber-500, --zinc-*, --font-display, --font-mono, --radius-*) — already established in the stylesheet. No new color constants introduced. JS wiring lands in the next commit so this diff stays focused on presentation layer only. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(ui): wire security banner to security_event + interactivity Adds showSecurityBanner() and hideSecurityBanner() plus the addChatEntry routing for entry.type === 'security_event'. When the sidebar-agent emits a security_event (canary leak or ML BLOCK), the banner renders with: * Title ("Session terminated") * Subtitle with {domain} if present, otherwise generic * Expandable layer list — each row: SECURITY_LAYER_LABELS[layer] + confidence.toFixed(2) in mono. Readable + auditable — user can see which layer fired at what score Interactivity, wired once on DOMContentLoaded: * Close X → hideSecurityBanner() * Expand/collapse "What happened" → toggles details + aria-expanded + chevron rotation (200ms css transition already in place) * Escape key dismisses while banner is visible (a11y) No shield icon yet — that's a separate commit that will consume the `security` field now returned by /health. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(ui): add security shield icon in sidepanel header (3 states) Small "SEC" badge in the top-right of the sidepanel that reflects the security module's current state. Three states drive color: protected green — all layers ok (TestSavantAI + transcript + canary) degraded amber — one+ ML layer offline but canary + arch controls active inactive red — security module crashed, arch controls only Consumes /health.security (surfaced in commit |
||
|
Garry Tan
|
d0782c4c4d |
feat(v1.4.0.0): /make-pdf — markdown to publication-quality PDFs (#1086)
* feat(browse): full $B pdf flag contract + tab-scoped load-html/js/pdf
Grow $B pdf from a 2-line wrapper (hard-coded A4) into a real PDF engine
frontend so make-pdf can shell out to it without duplicating Playwright:
- pdf: --format, --width/--height, --margins, --margin-*, --header-template,
--footer-template, --page-numbers, --tagged, --outline, --print-background,
--prefer-css-page-size, --toc. Mutex rules enforced. --from-file <json>
dodges Windows argv limits (8191 char CreateProcess cap).
- load-html: add --from-file <json> mode for large inline HTML. Size + magic
byte checks still apply to the inline content, not the payload file path.
- newtab: add --json returning {"tabId":N,"url":...} for programmatic use.
- cli: extract --tab-id flag and route as body.tabId to the HTTP layer so
parallel callers can target specific tabs without racing on the active
tab (makes make-pdf's per-render tab isolation possible).
- --toc: non-fatal 3s wait for window.__pagedjsAfterFired. Paged.js ships
later; v1 renders TOC statically via the markdown renderer.
Codex round 2 flagged these P0 issues during plan review. All resolved.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(resolvers): add MAKE_PDF_SETUP + makePdfDir host paths
Skill templates can now embed {{MAKE_PDF_SETUP}} to resolve $P to the
make-pdf binary via the same discovery order as $B / $D: env override
(MAKE_PDF_BIN), local skill root, global install, or PATH.
Mirrors the pattern established by generateBrowseSetup() and
generateDesignSetup() in scripts/resolvers/design.ts.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(make-pdf): new /make-pdf skill + orchestrator binary
Turn markdown into publication-quality PDFs. $P generate input.md out.pdf
produces a PDF with 1in margins, intelligent page breaks, page numbers,
running header, CONFIDENTIAL footer, and curly quotes/em dashes — all on
Helvetica so copy-paste extraction works ("S ai li ng" bug avoided).
Architecture (per Codex round 2):
markdown → render.ts (marked + sanitize + smartypants) → orchestrator
→ $B newtab --json → $B load-html --tab-id → $B js (poll Paged.js)
→ $B pdf --tab-id → $B closetab
browseClient.ts shells out to the compiled browse CLI rather than
duplicating Playwright. --tab-id isolation per render means parallel
$P generate calls don't race on the active tab. try/finally tab cleanup
survives Paged.js timeouts, browser crashes, and output-path failures.
Features in v1:
--cover left-aligned cover page (eyebrow + title + hairline rule)
--toc clickable static TOC (Paged.js page numbers deferred)
--watermark <text> diagonal DRAFT/CONFIDENTIAL layer
--no-chapter-breaks opt out of H1-starts-new-page
--page-numbers "N of M" footer (default on)
--tagged --outline accessible PDF + bookmark outline (default on)
--allow-network opt in to external image loading (default off for privacy)
--quiet --verbose stderr control
Design decisions locked from the /plan-design-review pass:
- Helvetica everywhere (Chromium emits single-word Tj operators for
system fonts; bundled webfonts emit per-glyph and break extraction).
- Left-aligned body, flush-left paragraphs, no text-indent, 12pt gap.
- Cover shares 1in margins with body pages; no flexbox-center, no
inset padding.
- The reference HTMLs at .context/designs/*.html are the implementation
source of truth for print-css.ts.
Tests (56 unit + 1 E2E combined-features gate):
- smartypants: code/URL-safe, verified against 10 fixtures
- sanitizer: strips <script>/<iframe>/on*/javascript: URLs
- render: HTML assembly, CJK fallback, cover/TOC/chapter wrap
- print-css: all @page rules, margin variants, watermark
- pdftotext: normalize()+copyPasteGate() cross-OS tolerance
- browseClient: binary resolution + typed error propagation
- combined-features gate (P0): 2-chapter fixture with smartypants +
hyphens + ligatures + bold/italic + inline code + lists + blockquote
passes through PDF → pdftotext → expected.txt diff
Deferred to Phase 4 (future PR): Paged.js vendored for accurate TOC page
numbers, highlight.js for syntax highlighting, drop caps, pull quotes,
two-column, CMYK, watermark visual-diff acceptance.
Plan: .context/ceo-plans/2026-04-19-perfect-pdf-generator.md
References: .context/designs/make-pdf-*.html
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore(build): wire make-pdf into build/test/setup/bin + add marked dep
- package.json: compile make-pdf/dist/pdf as part of bun run build; add
"make-pdf" to bin entry; include make-pdf/test/ in the free test pass;
add marked@18.0.2 as a dep (markdown parser, ~40KB).
- setup: add make-pdf/dist/pdf to the Apple Silicon codesign loop.
- .gitignore: add make-pdf/dist/ (matches browse/dist/ and design/dist/).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* ci(make-pdf): matrix copy-paste gate on Ubuntu + macOS
Runs the combined-features P0 gate on pull requests that touch make-pdf/
or browse's PDF surface. Installs poppler (macOS) / poppler-utils (Ubuntu)
per OS. Windows deferred to tolerant mode (Xpdf / Poppler-Windows
extraction variance not yet calibrated against the normalized comparator —
Codex round 2 #18).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs(skills): regenerate SKILL.md for make-pdf addition + browse pdf flags
bun run gen:skill-docs picks up:
- the new /make-pdf skill (make-pdf/SKILL.md)
- updated browse command descriptions for 'pdf', 'load-html', 'newtab'
reflecting the new flag contract and --from-file mode
Source of truth stays the .tmpl files + COMMAND_DESCRIPTIONS;
these are regenerated artifacts.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(tests): repair stale test expectations + emit _EXPLAIN_LEVEL / _QUESTION_TUNING from preamble
Three pre-existing test failures on main were blocking /ship:
- test/skill-validation.test.ts "Step 3.4 test coverage audit" expected the
literal strings "CODE PATH COVERAGE" and "USER FLOW COVERAGE" which were
removed when the Step 7 coverage diagram was compressed. Updated assertions
to check the stable `Code paths:` / `User flows:` labels that still ship.
- test/skill-validation.test.ts "ship step numbering" allowed-substeps list
didn't include 15.0 (WIP squash) and 15.1 (bisectable commits) which were
added for continuous checkpoint mode. Extended the allowlist.
- test/writing-style-resolver.test.ts and test/plan-tune.test.ts expected
`_EXPLAIN_LEVEL` and `_QUESTION_TUNING` bash variables in the preamble but
generate-preamble-bash.ts had been refactored and those lines were dropped.
Without them, downstream skills can't read `explain_level` or
`question_tuning` config at runtime — terse mode and /plan-tune features
were silently broken.
Added the two bash echo blocks back to generatePreambleBash and refreshed
the golden-file fixtures to match. All three preamble-related golden
baselines (claude/codex/factory) are synchronized with the new output.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: bump version and changelog (v1.4.0.0)
New /make-pdf skill + $P binary.
Turn any markdown file into a publication-quality PDF. Default output is
a 1in-margin Helvetica letter with page numbers in the footer. `--cover`
adds a left-aligned cover page, `--toc` generates a clickable table of
contents, `--watermark DRAFT` overlays a diagonal watermark. Copy-paste
extraction from the PDF produces clean words, not "S a i l i n g"
spaced out letter by letter. CI gate (macOS + Ubuntu) runs a combined-
features fixture through pdftotext on every PR.
make-pdf shells out to browse rather than duplicating Playwright.
$B pdf grew into a real PDF engine with full flag contract (--format,
--margins, --header-template, --footer-template, --page-numbers,
--tagged, --outline, --toc, --tab-id, --from-file). $B load-html and
$B js gained --tab-id. $B newtab --json returns structured output.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* docs(changelog): rewrite v1.4.0.0 headline — positive voice, no VC framing
The original headline led with "a PDF you wouldn't be embarrassed to send
to a VC": double-negative voice and audience-too-narrow. /make-pdf works
for essays, letters, memos, reports, proposals, and briefs. Framing the
whole release around founders-to-investors misses the wider audience.
New headline: "Turn any markdown file into a PDF that looks finished."
New tagline: "This one reads like a real essay or a real letter."
Positive voice. Broader aperture. Same energy.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
||
|
Garry Tan
|
22a4451e0e |
feat(v1.3.0.0): open agents learnings + cross-model benchmark skill (#1040)
* chore: regenerate stale ship golden fixtures
Golden fixtures were missing the VENDORED_GSTACK preamble section that
landed on main. Regression tests failed on all three hosts (claude, codex,
factory). Regenerated from current preamble output.
No code changes, unblocks test suite.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: anti-slop design constraints + delete duplicate constants
Tightens design-consultation and design-shotgun to push back on the
convergence traps every AI design tool falls into.
Changes:
- scripts/resolvers/constants.ts: add "system-ui as primary font" to
AI_SLOP_BLACKLIST. Document Space Grotesk as the new "safe alternative
to Inter" convergence trap alongside the existing overused fonts.
- scripts/gen-skill-docs.ts: delete duplicate AI slop constants block
(dead code — scripts/resolvers/constants.ts is the live source).
Prevents drift between the two definitions.
- design-consultation/SKILL.md.tmpl: add Space Grotesk + system-ui to
overused/slop lists. Add "anti-convergence directive" — vary across
generations in the same project. Add Phase 1 "memorable-thing forcing
question" (what's the one thing someone will remember?). Add Phase 5
"would a human designer be embarrassed by this?" self-gate before
presenting variants.
- design-shotgun/SKILL.md.tmpl: anti-convergence directive — each
variant must use a different font, palette, and layout. If two
variants look like siblings, one of them failed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: context health soft directive in preamble (T2+)
Adds a "periodically self-summarize" nudge to long-running skills.
Soft directive only — no thresholds, no enforcement, no auto-commit.
Goal: self-awareness during /qa, /investigate, /cso etc. If you notice
yourself going in circles, STOP and reassess instead of thrashing.
Codex review caught that fake precision thresholds (15/30/45 tool calls)
were unimplementable — SKILL.md is a static prompt, not runtime code.
This ships the soft version only.
Changes:
- scripts/resolvers/preamble.ts: add generateContextHealth(), wire into
T2+ tier. Format: [PROGRESS] ... summary line. Explicit rule that
progress reporting must never mutate git state.
- All T2+ skill SKILL.md files regenerated to include the new section.
- Golden ship fixtures updated (T4 skill, picks up the change).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: model overlays with explicit --model flag (no auto-detect)
Adds a per-model behavioral patch layer orthogonal to the host axis.
Different LLMs have different tendencies (GPT won't stop, Gemini
over-explains, o-series wants structured output). Overlays nudge each
model toward better defaults for gstack workflows.
Codex review caught three landmines the prior reviews missed:
1. Host != model — Claude Code can run any Claude model, Codex runs
GPT/o-series, Cursor fronts multiple providers. Auto-detecting from
host would lie. Dropped auto-detect. --model is explicit (default
claude). Missing overlay file → empty string (graceful).
2. Import cycle — putting Model in resolvers/types.ts would cycle
through hosts/index. Created neutral scripts/models.ts instead.
3. "Final say" is dangerous — overlay at the end of preamble could
override STOP points, AskUserQuestion gates, /ship review gates.
Placed overlay after spawned-session-check but before voice + tier
sections. Wrapper heading adds explicit subordination language on
every overlay: "subordinate to skill workflow, STOP points,
AskUserQuestion gates, plan-mode safety, and /ship review gates."
Changes:
- scripts/models.ts: new neutral module. ALL_MODEL_NAMES, Model type,
resolveModel() for family heuristics (gpt-5.4-mini → gpt-5.4, o3 →
o-series, claude-opus-4-7 → claude), validateModel() helper.
- scripts/resolvers/types.ts: import Model, add ctx.model field.
- scripts/resolvers/model-overlay.ts: new resolver. Reads
model-overlays/{model}.md. Supports {{INHERIT:base}} directive at
top of file for concat (gpt-5.4 inherits gpt). Cycle guard.
- scripts/resolvers/index.ts: register MODEL_OVERLAY resolver.
- scripts/resolvers/preamble.ts: wire generateModelOverlay into
composition before voice. Print MODEL_OVERLAY: {model} in preamble
bash so users can see which overlay is active. Filter empty sections.
- scripts/gen-skill-docs.ts: parse --model CLI flag. Default claude.
Unknown model → throw with list of valid options.
- model-overlays/{claude,gpt,gpt-5.4,gemini,o-series}.md: behavioral
patches per model family. gpt-5.4.md uses {{INHERIT:gpt}} to extend
gpt.md without duplication.
- test/gen-skill-docs.test.ts: fix qa-only guardrail regex scope.
Was matching Edit/Glob/Grep anywhere after `allowed-tools:` in the
whole file. Now scoped to frontmatter only. Body prose (Claude
overlay references Edit as a tool) correctly no longer breaks it.
Verification:
- bun run gen:skill-docs --host all --dry-run → all fresh
- bun run gen:skill-docs --model gpt-5.4 → concat works, gpt.md +
gpt-5.4.md content appears in order
- bun run gen:skill-docs --model unknown → errors with valid list
- All generated skills contain MODEL_OVERLAY: claude in preamble
- Golden ship fixtures regenerated
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: continuous checkpoint mode with non-destructive WIP squash
Adds opt-in auto-commit during long sessions so work survives Claude
Code crashes, Conductor workspace handoffs, and context switches.
Local-only by default — pushing requires explicit opt-in.
Codex review caught multiple landmines that would have shipped:
1. checkpoint_push=true default would push WIP commits to shared
branches, trigger CI/deploys, expose secrets. Now default false.
2. Plan's original /ship squash (git reset --soft to merge base) was
destructive — uncommitted ALL branch commits, not just WIP, and
caused non-fast-forward pushes. Redesigned: rebase --autosquash
scoped to WIP commits only, with explicit fallback for WIP-only
branches and STOP-and-ask for conflicts.
3. gstack-config get returned empty for missing keys with exit 0,
ignoring the annotated defaults in the header comments. Fixed:
get now falls back to a lookup_default() table that is the
canonical source for defaults.
4. Telemetry default mismatched: header said 'anonymous' but runtime
treated empty as 'off'. Aligned: default is 'off' everywhere.
5. /checkpoint resume only read markdown checkpoint files, not the
WIP commit [gstack-context] bodies the plan referenced. Wired up
parsing of [gstack-context] blocks from WIP commits as a second
recovery trail alongside the markdown checkpoints.
Changes:
- bin/gstack-config: add checkpoint_mode (default explicit) and
checkpoint_push (default false) to CONFIG_HEADER. Add lookup_default()
as canonical default source. get() falls back to defaults when key
absent. list now shows value + source (set/default). New 'defaults'
subcommand to inspect the table.
- scripts/resolvers/preamble.ts: preamble bash reads _CHECKPOINT_MODE
and _CHECKPOINT_PUSH, prints CHECKPOINT_MODE: and CHECKPOINT_PUSH: so
the mode is visible. New generateContinuousCheckpoint() section in
T2+ tier describes WIP commit format with [gstack-context] body and
the rules (never git add -A, never commit broken tests, push only
if opted in). Example deliberately shows a clean-state context so
it doesn't contradict the rules.
- ship/SKILL.md.tmpl: new Step 5.75 WIP Commit Squash. Detects WIP
count, exports [gstack-context] blocks before squash (as backup),
uses rebase --autosquash for mixed branches and soft-reset only when
VERIFIED WIP-only. Explicit anti-footgun rules against blind soft-
reset. Aborts with BLOCKED status on conflict instead of destroying
non-WIP commits.
- checkpoint/SKILL.md.tmpl: new Step 1.5 to parse [gstack-context]
blocks from WIP commits via git log --grep="^WIP:". Merges with
markdown checkpoint for fuller session recovery.
- Golden ship fixtures regenerated (ship is T4, preamble change shows up).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: feature discovery flow gated by per-feature markers
Extends generateUpgradeCheck() to surface new features once per user
after a just-upgraded session. No more silent features.
Codex review caught: spawned sessions (OpenClaw, etc.) must skip the
discovery prompt entirely — they can't interactively answer. Feature
discovery now checks SPAWNED_SESSION first and is silent in those.
Discovery is per-feature, not per-upgrade. Each feature has its own
marker file at ~/.claude/skills/gstack/.feature-prompted-{name}. Once
the user has been shown a feature (accepted, shown docs, or skipped),
the marker is touched and the prompt never fires again for that
feature. Future features get their own markers.
V1 features surfaced:
- continuous-checkpoint: offer to enable checkpoint_mode=continuous
- model-overlay: inform-only note about --model flag and MODEL_OVERLAY
line in preamble output
Max one prompt per session to avoid nagging. Fires only on JUST_UPGRADED
(not every session), plus spawned-session skip.
Changes:
- scripts/resolvers/preamble.ts: extend generateUpgradeCheck() with
feature discovery rules, per-marker-file semantics, spawned-session
exclusion, and max-one-per-session cap.
- All skill SKILL.md files regenerated to include the new section.
- Golden ship fixtures regenerated.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: design taste engine with persistent schema
Adds a cross-session taste profile that learns from design-shotgun
approval/rejection decisions. Biases future design-consultation and
design-shotgun proposals toward the user's demonstrated preferences.
Codex review caught that the plan had "taste engine" as a vague goal
without schema, decay, migration, or placeholder insertion points. This
commit ships the full spec.
Schema v1 at ~/.gstack/projects/$SLUG/taste-profile.json:
- version, updated_at
- dimensions: fonts, colors, layouts, aesthetics — each with approved[]
and rejected[] preference lists
- sessions: last 50 (FIFO truncation), each with ts/action/variant/reason
- Preference: { value, confidence, approved_count, rejected_count, last_seen }
- Confidence: Laplace-smoothed approved/(total+1)
- Decay: 5% per week of inactivity, computed at read time (not write)
Changes:
- bin/gstack-taste-update: new CLI. Subcommands approved/rejected/show/
migrate. Parses reason string for dimension signals (e.g.,
"fonts: Geist; colors: slate; aesthetics: minimal"). Emits taste-drift
NOTE when a new signal contradicts a strong opposing signal. Legacy
approved.json aggregates migrate to v1 on next write.
- scripts/resolvers/design.ts: new generateTasteProfile() resolver.
Produces the prose that skills see: how to read the profile, how to
factor into proposals, conflict handling, schema migration.
- scripts/resolvers/index.ts: register TASTE_PROFILE and a BIN_DIR
resolver (returns ctx.paths.binDir, used by templates that shell out
to gstack-* binaries).
- design-consultation/SKILL.md.tmpl: insert {{TASTE_PROFILE}} placeholder
in Phase 1 right after the memorable-thing forcing question so the
Phase 3 proposal can factor in learned preferences.
- design-shotgun/SKILL.md.tmpl: taste memory section now reads
taste-profile.json via {{TASTE_PROFILE}}, falls back to per-session
approved.json (legacy). Approval flow documented to call
gstack-taste-update after user picks/rejects a variant.
Known gap: v1 extracts dimension signals from a reason string passed
by the caller ("fonts: X; colors: Y"). Future v2 can read EXIF or an
accompanying manifest written by design-shotgun alongside each variant
for automatic dimension extraction without needing the reason argument.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: multi-provider model benchmark (boil the ocean)
Adds the full spec Codex asked for: real provider adapters with auth
detection, normalized RunResult, pricing tables, tool compatibility
maps, parallel execution with error isolation, and table/JSON/markdown
output. Judge stays on Anthropic SDK as the single stable source of
quality scoring, gated behind --judge.
Codex flagged the original plan as massively under-scoped — the
existing runner is Claude-only and the judge is Anthropic-only. You
can't benchmark GPT or Gemini without real provider infrastructure.
This commit ships it.
New architecture:
test/helpers/providers/types.ts ProviderAdapter interface
test/helpers/providers/claude.ts wraps `claude -p --output-format json`
test/helpers/providers/gpt.ts wraps `codex exec --json`
test/helpers/providers/gemini.ts wraps `gemini -p --output-format stream-json --yolo`
test/helpers/pricing.ts per-model USD cost tables (quarterly)
test/helpers/tool-map.ts which tools each CLI exposes
test/helpers/benchmark-runner.ts orchestrator (Promise.allSettled)
test/helpers/benchmark-judge.ts Anthropic SDK quality scorer
bin/gstack-model-benchmark CLI entry
test/benchmark-runner.test.ts 9 unit tests (cost math, formatters, tool-map)
Per-provider error isolation:
- auth → record reason, don't abort batch
- timeout → record reason, don't abort batch
- rate_limit → record reason, don't abort batch
- binary_missing → record in available() check, skip if --skip-unavailable
Pricing correction: cached input tokens are disjoint from uncached
input tokens (Anthropic/OpenAI report them separately). Original
math subtracted them, producing negative costs. Now adds cached at
the 10% discount alongside the full uncached input cost.
CLI:
gstack-model-benchmark --prompt "..." --models claude,gpt,gemini
gstack-model-benchmark ./prompt.txt --output json --judge
gstack-model-benchmark ./prompt.txt --models claude --timeout-ms 60000
Output formats: table (default), json, markdown. Each shows model,
latency, in→out tokens, cost, quality (when --judge used), tool calls,
and any errors.
Known limitations for v1:
- Claude adapter approximates toolCalls as num_turns (stream-json
would give exact counts; v2 can upgrade).
- Live E2E tests (test/providers.e2e.test.ts) not included — they
require CI secrets for all three providers. Unit tests cover the
shape and math.
- Provider CLIs sometimes return non-JSON error text to stdout; the
parsers fall back to treating raw output as plain text in that case.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat: standalone methodology skill publishing via gstack-publish
Ships the marketplace-distribution half of Item 5 (reframed): publish
the existing standalone OpenClaw methodology skills to multiple
marketplaces with one command.
Codex review caught that the original plan assumed raw generated
multi-host skills could be published directly. They can't — those
depend on gstack binaries, generated host paths, tool names, and
telemetry. The correct artifact class is hand-crafted standalone
skills in openclaw/skills/gstack-openclaw-* (already exist and work
without gstack runtime). This commit adds the wrapper that publishes
them to ClawHub + SkillsMP + Vercel Skills.sh with per-marketplace
error isolation and dry-run validation.
Changes:
- skills.json: root manifest with 4 skills (office-hours, ceo-review,
investigate, retro) each pointing at its openclaw/skills source.
Each skill declares per-marketplace targets with a slug, a publish
flag, and a compatible-hosts list. Marketplace configs include CLI
name, login command, publish command template (with placeholder
substitution), docs URL, and auth_check command.
- bin/gstack-publish: new CLI. Subcommands:
gstack-publish Publish all skills
gstack-publish <slug> Publish one skill
gstack-publish --dry-run Validate + auth-check without publishing
gstack-publish --list List skills + marketplace targets
Features:
* Manifest validation (missing source files, missing slugs, empty
marketplace list all reported).
* Per-marketplace auth check before any publish attempt.
* Per-skill / per-marketplace error isolation: one failure doesn't
abort the batch.
* Idempotent — re-running with the same version is safe; markets
that reject duplicate versions report it as a failure for that
single target without affecting others.
* --dry-run walks the full pipeline but skips execSync; useful in
CI to validate manifest before bumping version.
Tested locally: clawhub auth detected, skillsmp/vercel CLIs not
installed (marked NOT READY and skipped cleanly in dry-run).
Follow-up work (tracked in TODOS.md later):
- Version-bump helper that reads openclaw/skills/*/SKILL.md frontmatter
and updates skills.json in lockstep.
- CI workflow that runs gstack-publish --dry-run on every PR and
gstack-publish on tags.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* refactor: split preamble.ts into submodules (byte-identical output)
Splits scripts/resolvers/preamble.ts (841 lines, 18 generator functions +
composition root) into one file per generator under
scripts/resolvers/preamble/. Root preamble.ts becomes a thin composition
layer (~80 lines of imports + generatePreamble).
Before:
scripts/resolvers/preamble.ts 841 lines
After:
scripts/resolvers/preamble.ts 83 lines
scripts/resolvers/preamble/generate-preamble-bash.ts 97 lines
scripts/resolvers/preamble/generate-upgrade-check.ts 48 lines
scripts/resolvers/preamble/generate-lake-intro.ts 16 lines
scripts/resolvers/preamble/generate-telemetry-prompt.ts 37 lines
scripts/resolvers/preamble/generate-proactive-prompt.ts 25 lines
scripts/resolvers/preamble/generate-routing-injection.ts 49 lines
scripts/resolvers/preamble/generate-vendoring-deprecation.ts 36 lines
scripts/resolvers/preamble/generate-spawned-session-check.ts 11 lines
scripts/resolvers/preamble/generate-ask-user-format.ts 16 lines
scripts/resolvers/preamble/generate-completeness-section.ts 19 lines
scripts/resolvers/preamble/generate-repo-mode-section.ts 12 lines
scripts/resolvers/preamble/generate-test-failure-triage.ts 108 lines
scripts/resolvers/preamble/generate-search-before-building.ts 14 lines
scripts/resolvers/preamble/generate-completion-status.ts 161 lines
scripts/resolvers/preamble/generate-voice-directive.ts 60 lines
scripts/resolvers/preamble/generate-context-recovery.ts 51 lines
scripts/resolvers/preamble/generate-continuous-checkpoint.ts 48 lines
scripts/resolvers/preamble/generate-context-health.ts 31 lines
Byte-identity verification (the real gate per Codex correction):
- Before refactor: snapshotted 135 generated SKILL.md files via
`find -name SKILL.md -type f | grep -v /gstack/` across all hosts.
- After refactor: regenerated with `bun run gen:skill-docs --host all`
and re-snapshotted.
- `diff -r baseline after` returned zero differences and exit 0.
The `--host all --dry-run` gate passes too. No template or host behavior
changes — purely a code-organization refactor.
Test fix: audit-compliance.test.ts's telemetry check previously grepped
preamble.ts directly for `_TEL != "off"`. After the refactor that logic
lives in preamble/generate-preamble-bash.ts. Test now concatenates all
preamble submodule sources before asserting — tracks the semantic contract,
not the file layout. Doing the minimum rewrite preserves the test's intent
(conditional telemetry) without coupling it to file boundaries.
Why now: we were in-session with full context. Codex had downgraded this
from mandatory to optional, but the preamble had grown to 841 lines and
was getting harder to navigate. User asked "why not?" given the context
was hot. Shipping it as a clean bisectable commit while all the prior
preamble.ts changes are fresh reduces rebase pain later.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: bump version and changelog (v0.19.0.0)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: trim verbose preamble + coverage audit prose
Compress without removing behavior or voice. Three targeted cuts:
1. scripts/resolvers/testing.ts coverage diagram example: 40 lines → 14
lines. Two-column ASCII layout instead of stacked sections.
Preserves all required regression-guard phrases (processPayment,
refundPayment, billing.test.ts, checkout.e2e.ts, COVERAGE, QUALITY,
GAPS, Code paths, User flows, ASCII coverage diagram).
2. scripts/resolvers/preamble/generate-completion-status.ts Plan Status
Footer: was 35 lines with embedded markdown table example, now 7
lines that describe the table inline. The footer fires only at
ExitPlanMode time — Claude can construct the placeholder table from
the inline description without copying a literal example.
3. Same file's Plan Mode Safe Operations + Skill Invocation During Plan
Mode sections compressed from ~25 lines combined to ~12. Preserves
all required test phrases (precedence over generic plan mode behavior,
Do not continue the workflow, cancel the skill or leave plan mode,
PLAN MODE EXCEPTION).
NOT touched:
- Voice directive (Garry's voice — protected per CLAUDE.md)
- Office-hours Phase 6 Handoff (Garry's voice + YC pitch)
- Test bootstrap, review army, plan completion (carefully tuned behavior)
Token savings (per skill, system-wide):
ship/SKILL.md 35474 → 34992 tokens (-482)
plan-ceo-review 29436 → 28940 (-496)
office-hours 26700 → 26204 (-496)
Still over the 25K ceiling. Bigger reduction requires restructure
(move large resolvers to externally-referenced docs, split /ship into
ship-quick + ship-full, or refactor the coverage audit + review army
into shorter prose). That's a follow-up — added to TODOS.
Tests: 420/420 pass on gen-skill-docs.test.ts + host-config.test.ts.
Goldens regenerated for claude/codex/factory ship.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(ci): install Node.js from official tarball instead of NodeSource apt setup
The CI Dockerfile's Node install was failing on ubicloud runners. NodeSource's
setup_22.x script runs two internal apt operations that both depend on
archive.ubuntu.com + security.ubuntu.com being reachable:
1. apt-get update (to refresh package lists)
2. apt-get install gnupg (as a prerequisite for its gpg keyring)
Ubicloud's CI runners frequently can't reach those mirrors — last build hit
~2min of connection timeouts to every security.ubuntu.com IP (185.125.190.82,
91.189.91.83, 91.189.92.24, etc.) plus archive.ubuntu.com mirrors. Compounding
this: on Ubuntu 24.04 (noble) "gnupg" was renamed to "gpg" and "gpgconf".
NodeSource's setup script still looks for "gnupg", so even when apt works,
it fails with "Package 'gnupg' has no installation candidate." The subsequent
apt-get install nodejs then fails because the NodeSource repo was never added.
Fix: drop NodeSource entirely. Download Node.js v22.20.0 from nodejs.org as a
tarball, extract to /usr/local. One host, no apt, no script, no keyring.
Before:
RUN curl -fsSL https://deb.nodesource.com/setup_22.x | bash - \
&& apt-get install -y --no-install-recommends nodejs ...
After:
ENV NODE_VERSION=22.20.0
RUN curl -fsSL "https://nodejs.org/dist/v${NODE_VERSION}/node-v${NODE_VERSION}-linux-x64.tar.xz" -o /tmp/node.tar.xz \
&& tar -xJ -C /usr/local --strip-components=1 --no-same-owner -f /tmp/node.tar.xz \
&& rm -f /tmp/node.tar.xz \
&& node --version && npm --version
Same installed path (/usr/local/bin/node and npm). Pinned version for
reproducibility. Version is bump-visible in the Dockerfile now.
Does not address the separate apt flakiness that affects the GitHub CLI
install (line 17) or `npx playwright install-deps chromium` (line 33) —
those use apt too. If those fail on a future build we can address then.
Failing job: build-image (71777913820)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* chore: raise skill token ceiling warning from 25K to 40K
The 25K ceiling predated flagship models with 200K-1M windows and assumed
every skill prompt dominates context cost. Modern reality: prompt caching
amortizes the skill load across invocations, and three carefully-tuned
skills (ship, plan-ceo-review, office-hours) legitimately pack 25-35K
tokens of behavior that can't be cut without degrading quality or removing
protected content (Garry's voice, YC pitch, specialist review instructions).
We made the safe prose cuts earlier (coverage diagram, plan status footer,
plan mode operations). The remaining gap is structural — real compression
would require splitting /ship into ship-quick vs ship-full, externalizing
large resolvers to reference docs, or removing detailed skill behavior.
Each is 1-2 days of work. The cost of the warning firing is zero (it's
a warning, not an error). The cost of hitting it is ~15¢ per invocation
at worst, amortized further by prompt caching.
Raising to 40K catches what it's supposed to catch — a runaway 10K+ token
growth in a single release — without crying wolf on legitimately big
skills. Reference doc in CLAUDE.md updated to reflect the new philosophy:
when you hit 40K, ask WHAT grew, don't blindly compress tuned prose.
scripts/gen-skill-docs.ts: TOKEN_CEILING_BYTES 100_000 → 160_000.
CLAUDE.md: document the "watch for feature bloat, not force compression"
intent of the ceiling.
Verification: `bun run gen:skill-docs --host all` shows zero TOKEN
CEILING warnings under the new 40K threshold.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(ci): install xz-utils so Node tarball extraction works
The direct-tarball Node install (switched from NodeSource apt in the last
CI fix) failed with "xz: Cannot exec: No such file or directory" because
Ubuntu 24.04 base doesn't include xz-utils. Node ships .tar.xz by default,
and `tar -xJ` shells out to xz, which was missing.
Add xz-utils to the base apt install alongside git/curl/unzip/etc.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* fix(benchmark): pass --skip-git-repo-check to codex adapter
The gpt provider adapter spawns `codex exec -C <workdir>` with arbitrary
working directories (benchmark temp dirs, non-git paths). Without
`--skip-git-repo-check`, codex refuses to run and returns "Not inside a
trusted directory" — surfaced as a generic error.code='unknown' that
looks like an API failure.
Benchmarks don't care about codex's git-repo trust model; we just want
the prompt executed. Surfaced by the new provider live E2E test on a
temp workdir.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(benchmark): add --dry-run flag to gstack-model-benchmark
Matches gstack-publish --dry-run semantics. Validates the provider list,
resolves per-adapter auth, echoes the resolved flag values, and exits
without invoking any provider CLI. Zero-cost pre-flight for CI pipelines
and for catching auth drift before starting a paid benchmark run.
Output shape:
== gstack-model-benchmark --dry-run ==
prompt: <truncated>
providers: claude, gpt, gemini
workdir: /tmp/...
timeout_ms: 300000
output: table
judge: off
Adapter availability:
claude: OK
gpt: NOT READY — <reason>
gemini: NOT READY — <reason>
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* test: lite E2E coverage for benchmark, taste engine, publish
Fills real coverage gaps in v0.19.0.0 primitives. 44 new deterministic
tests (gate tier, ~3s) + 8 live-API tests (periodic tier).
New gate-tier test files (free, <3s total):
- test/taste-engine.test.ts — 24 tests against gstack-taste-update:
schema shape, Laplace-smoothed confidence, 5%/week decay clamped at 0,
multi-dimension extraction, case-insensitive matching, session cap,
legacy profile migration with session truncation, taste-drift conflict
warning, malformed-JSON recovery, missing-variant exit code.
- test/publish-dry-run.test.ts — 13 tests against gstack-publish --dry-run:
manifest parsing, missing/malformed JSON, per-skill validation errors
(missing source file / slug / version / marketplaces), slug filter,
unknown-skill exit, per-marketplace auth isolation (fake marketplaces
with always-pass / always-fail / missing-binary CLIs), and a sanity
check against the real repo manifest.
- test/benchmark-cli.test.ts — 11 tests against gstack-model-benchmark
--dry-run: provider default, unknown-provider WARN, empty list
fallback, flag passthrough (timeout/workdir/judge/output), long-prompt
truncation, prompt resolution (inline vs file vs positional), missing
prompt exit.
New periodic-tier test file (paid, gated EVALS=1):
- test/skill-e2e-benchmark-providers.test.ts — 8 tests hitting real
claude, codex, gemini CLIs with a trivial prompt (~$0.001/provider).
Verifies output parsing, token accounting, cost estimation, timeout
error.code semantics, Promise.allSettled parallel isolation.
Per-provider availability gate — unauthed providers skip cleanly.
This suite already caught one real bug (codex adapter missing
--skip-git-repo-check, fixed in
|
||
|
Garry Tan
|
12260262ea |
fix(checkpoint): rename /checkpoint → /context-save + /context-restore (v1.0.1.0) (#1064)
* rename /checkpoint → /context-save + /context-restore (split) Claude Code ships /checkpoint as a native alias for /rewind (Esc+Esc), which was shadowing the gstack skill. Training-data bleed meant agents saw /checkpoint and sometimes described it as a built-in instead of invoking the Skill tool, so nothing got saved. Fix: rename the skill and split save from restore so each skill has one job. Restore now loads the most recent saved context across ALL branches by default (the previous flow was ambiguous between mode="restore" and mode="list" and agents applied list-flow filtering to restore). New commands: - /context-save → save current state - /context-save list → list saved contexts (current branch default) - /context-restore → load newest saved context across all branches - /context-restore X → load specific saved context by title fragment Storage directory unchanged at ~/.gstack/projects/$SLUG/checkpoints/ so existing saved files remain loadable. Canonical ordering is now the filename YYYYMMDD-HHMMSS prefix, not filesystem mtime — filenames are stable across copies/rsync, mtime is not. Empty-set handling in both restore and list flows uses find+sort instead of ls -1t, which on macOS falls back to listing cwd when the input is empty. Sources for the collision: - https://code.claude.com/docs/en/checkpointing - https://claudelog.com/mechanics/rewind/ * preamble: split 'checkpoint' routing rule into context-save + context-restore scripts/resolvers/preamble.ts:238 is the source of truth for the routing rules that gstack writes into users' CLAUDE.md on first skill run, AND gets baked into every generated SKILL.md. A single 'invoke checkpoint' line points at a skill that no longer exists. Replace with two lines: - Save progress, save state, save my work → invoke context-save - Resume, where was I, pick up where I left off → invoke context-restore Tier comment at :750 also updated. All SKILL.md files regenerated via bun run gen:skill-docs. * tests: split checkpoint-save-resume into context-save + context-restore E2Es Renames the combined E2E test to match the new skill split: - checkpoint-save-resume → context-save-writes-file Extracts the Save flow from context-save/SKILL.md, asserts a file gets written with valid YAML frontmatter. - New: context-restore-loads-latest Seeds two saved-context files with different YYYYMMDD-HHMMSS prefixes AND scrambled filesystem mtimes (so mtime DISAGREES with filename order). Hand-feeds the restore flow and asserts the newer- by-filename file is loaded. Locks in the "newest by filename prefix, not mtime" guarantee. touchfiles.ts: old 'checkpoint-save-resume' key removed from both E2E_TOUCHFILES and E2E_TIERS maps; new keys added to both. Leaving a key in one map but not the other silently breaks test selection. Golden baselines (claude/codex/factory ship skill) regenerated to match the new preamble routing rules from the previous commit. * migration: v0.18.5.0 removes stale /checkpoint install with ownership guard gstack-upgrade/migrations/v0.18.5.0.sh removes the stale on-disk /checkpoint install so Claude Code's native /rewind alias is no longer shadowed. Ownership guard inspects the directory itself (not just SKILL.md) and handles 3 install shapes: 1. ~/.claude/skills/checkpoint is a directory symlink whose canonical path resolves inside ~/.claude/skills/gstack/ → remove. 2. ~/.claude/skills/checkpoint is a directory containing exactly one file SKILL.md that's a symlink into gstack → remove (gstack's prefix-install shape). 3. Anything else (user's own regular file/dir, or a symlink pointing elsewhere) → leave alone, print a one-line notice. Also removes ~/.claude/skills/gstack/checkpoint/ unconditionally (gstack owns that dir). Portable realpath: `realpath` with python3 fallback for macOS BSD which lacks readlink -f. Idempotent: missing paths are no-ops. test/migration-checkpoint-ownership.test.ts ships 7 scenarios covering all 3 install shapes + idempotency + no-op-when-gstack-not-installed + SKILL.md-symlink-outside-gstack. Critical safety net for a migration that mutates user state. Free tier, ~85ms. * docs: bump VERSION to 0.18.5.0, CHANGELOG + TODOS entry User-facing changelog leads with the problem: /checkpoint silently stopped saving because Claude Code shipped a native /checkpoint alias for /rewind. The fix is a clean rename to /context-save + /context-restore, with the second bug (restore was filtering by current branch and hiding most recent saves) called out separately under Fixed. TODOS entry for the deferred lane feature points at the existing lane data model in plan-eng-review/SKILL.md.tmpl:240-249 so a future session can pick it up without re-discovering the source. * chore: bump package.json to 0.18.5.0 (match VERSION) * fix(test): skill-e2e-autoplan-dual-voice was shipped broken The test shipped on main in v0.18.4.0 used wrong option names and wrong result fields throughout. It could not have passed in any environment: Broken API calls: - `workdir` → should be `workingDirectory` The fixture setup (git init, copy autoplan + plan-*-review dirs, write TEST_PLAN.md) was completely ignored. claude -p spawned with undefined cwd instead of the tmp workdir. - `timeoutMs: 300_000` → should be `timeout: 300_000` Fell back to default 120s. Explains the observed ~170s failure (test harness overhead + retry startup). - `name: 'autoplan-dual-voice'` → should be `testName: 'autoplan-dual-voice'` No per-test run directory was created. - `evalCollector` → not a recognized `runSkillTest` option at all. Broken result access: - `result.stdout + result.stderr` → SkillTestResult has neither field. `out` was literally "undefinedundefined" every time. - Every regex match fired false. All 3 assertions (claudeVoiceFired, codex-or-unavailable, reachedPhase1) failed on every attempt. - `logCost(result)` → signature is `logCost(label, result)`. - `recordE2E('autoplan-dual-voice', result)` → signature is `recordE2E(evalCollector, name, suite, result, extra)`. Fixes: - Renamed all 4 broken options in the runSkillTest call. - Changed assertion source to `result.output` plus JSON-serialized `result.transcript` (broader net for voice fingerprints in tool inputs/outputs). - Widened regex alternatives: codex voice now matches "CODEX SAYS" and "codex-plan-review"; Claude voice now matches subagent_type; unavailable matches CODEX_NOT_AVAILABLE. - Added Agent + Skill + Edit + Grep + Glob to allowedTools. Without Agent, /autoplan can't spawn subagents and never reaches Phase 1. - Raised maxTurns 15 → 30 (autoplan is a long multi-phase skill). - Fixed logCost + recordE2E signatures, passing `passed:` flag into recordE2E per the neighboring context-save pattern. * security: harden migration + context-save after adversarial review Adversarial review (Claude + Codex, both high confidence) identified 6 critical production-harm findings in the /ship pre-landing pass. All folded in. Migration v1.0.1.0.sh hardening: - Add explicit `[ -z "${HOME:-}" ]` guard. HOME="" survives set -u and expands paths to /.claude/skills/... which could hit absolute paths under root/containers/sudo-without-H. - Add python3 fallback inside resolve_real() (was missing; broken symlinks silently defeated ownership check). - Ownership-guard Shape 2 (~/.claude/skills/gstack/checkpoint/). Was unconditional rm -rf. Now: if symlink, check target resolves inside gstack; if regular dir, check realpath resolves inside gstack. A user's hand-edited customization or a symlink pointing outside gstack is preserved with a notice. - Use `rm --` and `rm -r --` consistently to resist hostile basenames. - Use `find -type f -not -name .DS_Store -not -name ._*` instead of `ls -A | grep`. macOS sidecars no longer mask a legit prefix-mode install. Strip sidecars explicitly before removing the dir. context-save/SKILL.md.tmpl: - Sanitize title in bash, not LLM prose. Allowlist [a-z0-9.-], cap 60 chars, default to "untitled". Closes a prompt-injection surface where `/context-save $(rm -rf ~)` could propagate into subsequent commands. - Collision-safe filename. If ${TIMESTAMP}-${SLUG}.md already exists (same-second double-save with same title), append a 4-char random suffix. The skill contract says "saved files are append-only" — this enforces it. Silent overwrite was a data-loss bug. context-restore/SKILL.md.tmpl: - Cap `find ... | sort -r` at 20 entries via `| head -20`. A user with 10k+ saved files no longer blows the context window just to pick one. /context-save list still handles the full-history listing path. test/skill-e2e-autoplan-dual-voice.test.ts: - Filter transcript to tool_use / tool_result / assistant entries before matching, so prompt-text mentions of "plan-ceo-review" don't force the reachedPhase1 assertion to pass. Phase-1 assertion now requires completion markers ("Phase 1 complete", "Phase 2 started"), not mere name occurrence. - claudeVoiceFired now requires JSON evidence of an Agent tool_use (name:"Agent" or subagent_type field), not the literal string "Agent(" which could appear anywhere. - codexVoiceFired now requires a Bash tool_use with a `codex exec/review` command string, not prompt-text mentions. All SKILL.md files regenerated. Golden fixtures updated. bun test: 0 failures across 80+ targeted tests and the full suite. Review source: /ship Step 11 adversarial pass (claude subagent + codex exec). Same findings independently surfaced by both reviewers — this is cross-model high confidence. * test: tier-2 hardening tests for context-save + context-restore 21 unit-level tests covering the security + correctness hardening that landed in commit |
||
|
Garry Tan
|
8ee16b867b |
feat: mode-posture energy fix for /plan-ceo-review and /office-hours (v1.1.2.0) (#1065)
* feat: restore mode-posture energy to expansion + forcing + builder output
Rewrites Writing Style rule 2-4 examples in scripts/resolvers/preamble.ts
to cover three framing families (pain reduction, upside/delight, forcing
pressure) instead of diagnostic-pain only. Adds inline exemplars to
plan-ceo-review (0D-prelude shared between SCOPE + SELECTIVE EXPANSION)
and office-hours (Q3 forcing exemplar with career/day/weekend domain
gating, builder operating principles wild exemplar).
V1 shipped rule 2-4 examples that all pointed to diagnostic-pain framing
("3-second spinner", "double-click button"). Models follow concrete
examples over abstract taxonomies, so any skill with a non-diagnostic
mode posture (expansion, forcing, delight) got flattened at runtime
even when the template itself said "dream big" or "direct to the point
of discomfort." This change targets the actual lever: swap the single
diagnostic example for three paired framings, one per posture family.
Preserves V1 clarity gains — rules 2, 3, 4 principles unchanged, only
examples expanded. Terse mode (EXPLAIN_LEVEL: terse) still skips the
block entirely.
* chore: regenerate SKILL.md after preamble + template changes
Mechanical cascade from `bun run gen:skill-docs --host all` after the
Writing Style rule 2-4 example rewrite and the plan-ceo-review /
office-hours template exemplar additions. No hand edits — every change
flows from the prior commit's templates.
* test: add gate-tier mode-posture regression tests
Three gate-tier E2E tests detect when preamble / template changes
flatten the distinctive posture of /plan-ceo-review SCOPE EXPANSION or
/office-hours (startup Q3, builder mode). The V1 regression that this
PR fixes shipped without anyone catching it at ship time — this is the
ongoing signal so the same thing doesn't happen again.
Pieces:
- `judgePosture(mode, text)` in `test/helpers/llm-judge.ts`. Sonnet
judge with mode-specific dual-axis rubric (expansion: surface_framing
+ decision_preservation; forcing: stacking_preserved +
domain_matched_consequence; builder: unexpected_combinations +
excitement_over_optimization). Pass threshold 4/5 on both axes.
- Three fixtures in `test/fixtures/mode-posture/` — deterministic input
for expansion proposal generation, Q3 forcing question, and builder
adjacent-unlock riffing.
- `plan-ceo-review-expansion-energy` case appended to
`test/skill-e2e-plan.test.ts`. Generator: Opus (skill default). Judge:
Sonnet.
- New `test/skill-e2e-office-hours.test.ts` with
`office-hours-forcing-energy` + `office-hours-builder-wildness`
cases. Generator: Sonnet. Judge: Sonnet.
- Touchfile registration in `test/helpers/touchfiles.ts` — all three as
`gate` tier in `E2E_TIERS`, triggered by changes to
`scripts/resolvers/preamble.ts`, the relevant skill template, the
judge helper, or any mode-posture fixture.
Cost: ~$0.50-$1.50 per triggered PR. Sonnet judge is cheap; Opus
generator for the plan-ceo-review case dominates.
Known V1.1 tradeoff: judges test prose markers more than deep behavior.
V1.2 candidate is a cross-provider (Codex) adversarial judge on the
same output to decouple house-style bias.
* test: update golden ship baselines + touchfile count for mode-posture entries
Mechanical test updates after the mode-posture work:
- Golden ship SKILL.md baselines (claude + codex + factory hosts) regenerate with
the rewritten Writing Style rule 2-4 examples from preamble.ts.
- Touchfile selection test expects 6 matches for a plan-ceo-review/ change (was 5)
because E2E_TOUCHFILES now includes plan-ceo-review-expansion-energy.
* chore: bump version and changelog (v1.1.2.0)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|