Garry TanandClaude Fable 5
410b4928e7
* ci: bump CI image Bun 1.3.10 -> 1.3.13
Matches the local toolchain and brings native `bun test --shard=M/N` /
--parallel to CI (needed by the free-test lane and shard runner work).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci: stop version bumps rebuilding the eval Docker image (cache key trio)
Three coupled fixes, atomic because any subset is worse than none:
1. Image tag keys on hashFiles(Dockerfile.ci, bun.lock) — package.json is
out: its version field changed on 60/60 recent commits, forcing a ~2min
image rebuild per PR for a dependency set only bun.lock determines.
2. ci-image.yml now pushes that same content-hash tag (previously only
:latest/:sha, so the weekly prebuild never warmed the tag the eval
matrix actually looks up) and both eval workflows get registry layer
cache (cache-to export gated to same-repo runs; fork tokens cannot
write GHCR).
3. Dockerfile bakes /opt/node_modules_cache/.bun.lock and the runtime
Restore-deps guard diffs bun.lock instead of package.json — otherwise
every version-only bump made all 14 matrix jobs fall back to a live
bun install, which is slower than today's behavior.
Worst-case failure mode is self-healing: a missing tag or cache falls
back to exactly the previous rebuild-and-install path.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci: stop double-running lint + skill-docs on every PR commit
Both fired on unrestricted push AND pull_request, so each PR push ran
them twice (12 duplicate (headSha, workflow) pairs in the last 200 runs).
push is now main-only; pull_request covers PR branches.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci: run actionlint from the prebuilt image (16s -> ~2s)
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci: right-size five single-core jobs to ubicloud-standard-2
actionlint, skill-docs, version-gate, pr-title-sync, and the evals report
job never exceed one core; standard-8 was ~4x the cost for zero wall-clock.
build-image and the eval matrix keep standard-8.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci: fix workflow_dispatch concurrency collisions (head_ref || run_id)
head_ref is empty on workflow_dispatch, so every manual dispatch of these
four workflows shared one empty-suffix group and cancelled each other.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci(windows): cache bun installs; run the curated suite, not a hand list
- actions/cache on ~/.bun/install/cache keyed on bun.lock (install was
35-45s of both 55-64s jobs, all network) and Bun pinned to 1.3.13 to
match the other lanes.
- windows-free-tests now runs `bun run test:windows` (the runner's
--windows-only curation) instead of a hand-listed 13-file subset that
had drifted from the registry it sampled. POSIX-bound tests get
excluded in ONE place (the curation patterns), not two.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: retry 1, not 2, on every paid path
Measured on the llm-judge shard: --retry 2 amplified 25 tests into 46
executions (+84%), with retried runs at 138s vs a 10-12s baseline (429
backoff), and a permanently-failing test paying 3x. One retry still
absorbs one-off flakes; chronic flakes become visible fix-work instead
of silent wall-clock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: split skill-e2e-review into three per-file CI shards
Bun runs describe blocks as concurrency barriers, so the e2e-review CI
job executed its tests serially: 741s of an 860s PR critical path for
tests whose slowest member is 224s. The per-file matrix is the repo's
parallelism unit, so the split moves:
- Retro E2E + retro-base-branch -> test/skill-e2e-retro.test.ts
- review/ship base-branch + Review Dashboard Via Attribution
-> test/skill-e2e-review-attribution.test.ts
- sql-injection / enum-completeness / design-lite stay in
test/skill-e2e-review.test.ts
One 741s job becomes three ~180-250s jobs. Locally the worst paid shard
drops from 1705s (94.7% of the 1800s kill) to under 700s. Test names,
bodies, suite strings, and eval-store collectors are unchanged, so
baselines carry over. Matrix rows added to both eval workflows
(attribution is gate-only, so no periodic row); the report job's
hardcoded runner count is gone (drift-proof).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: gate security-bench on SECURITY_BENCH=1, not model-cache existence
The existsSync gate ran ~12s of ONNX inference (plus a HuggingFace
dataset fetch) on every free-suite run on any dev box that had ever
warmed the classifier, while CI (no cache) silently skipped it. Now
explicit opt-in: SECURITY_BENCH=1 bun test browse/test/security-bench.test.ts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: watchdog E2E in 1.5s instead of 22.7s (tunable poll interval)
server.ts gains BROWSE_WATCHDOG_INTERVAL_MS (floor 50ms, default 15s
unchanged). The #994 stay-alive test runs a 250ms tick and waits for the
stay-alive log line instead of blind-sleeping 2s + 20s past the
production interval.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: dedupe coverage gates; route both walks through skill-census
skill-coverage-floor duplicated two matrix assertions (registry
completeness, gate-tier floor) with a DIFFERENT hand-rolled directory
walk — matrix's skipped nothing, floor's skipped node_modules/docs/test.
Two 'same' gates disagreeing on the census is the bug class
test/helpers/skill-census.ts was written to kill. Registry assertions
now live in matrix only (with floor's better error message), both files
walk via skillCensus().authoredSkills, and floor keeps the per-skill
structural checks it owns.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: EVALS_JOBS for shard processes; explicit within-shard concurrency
EVALS_CONCURRENCY was overloaded: the legacy bun-test path used it as
--max-concurrency (default 15) while the sharded runner read it as the
process count — exporting the legacy value gave 15 concurrent Bun
processes each spawning claude (the 429 storm). Now: EVALS_JOBS = shard
processes (default 4); EVALS_CONCURRENCY = bun --max-concurrency inside
a shard (default 4, explicit in shard args — omitting it made
within-shard parallelism silently differ from the legacy path). Stale
49/59 header math replaced with the live-count rule.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: enforce detach-timeout floor from the live shard census
New free tripwire: eval:bg:gate / eval:bg:periodic --timeout must cover
ceil(shards/jobs) x shard-timeout x 1.05, recomputed from the actual paid
test census every run. Hand-derived numbers go stale every time a paid
file lands — the review split just proved it: periodic's 28800s dropped
BELOW its new 32130s worst case (raised to 32400s here). An undersized
watchdog kills healthy runs and the tail reports never-started.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: preflight ping once in the sharded parent, not per shard
The Anthropic fail-fast ping ran at module load in every paid test file
importing e2e-helpers — ~30 paid claude -p calls (30s timeout each) per
full sharded run for one bit of information. The parent now pings once
before spawning shards and sets EVALS_PREFLIGHT_OK=1; the module-load
path honors the flag. Extracted to test/helpers/anthropic-preflight.ts
(injectable spawn seam) with regression pins in both directions: the
flag must skip, its absence must ping exactly once, dead API must throw.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: split touchfiles into pure data + selection logic + facade
touchfiles.ts listed ITSELF in GLOBAL_TOUCHFILES, so adding one test's
dep entry forced the full ~$38 / 30-45min suite — measured on 21.9% of
recent commits (42/192). The self-reference existed because data and
logic shared a file: any edit COULD be a selection-logic change.
Now: touchfiles-data.ts (the four maps, literals only, zero imports —
the future map-diff target), test-selection.ts (matchGlob/detectBase
Branch/getChangedFiles/selectTests), and touchfiles.ts as a re-export
facade so all ~12 import sites are untouched. GLOBAL_TOUCHFILES drops
the self-ref, adds test-selection.ts (logic stays maximally
conservative), and TEMPORARILY adds touchfiles-data.ts until the
map-diff change lands. New free test pins the literal-only property
(comment-aware state-machine scan with a self-test) and facade export
parity (===), so neither can silently rot.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: free runner — strict output, parallel execution, stable shard indices
Three coupled changes to scripts/test-free-shards.ts:
1. STRICT OUTPUT: runFreeShard streams through the paid runner's
BunTestOutputClassifier — exit 0 without bun's 'Ran N tests across M
files' summary, with (fail) lines, or with a wrong file count is a
FAILURE (anti-truncation backstop at the runner layer), plus an
external wall-clock timeout that SIGKILLs the process group
(timed-out distinct from failed; exit 124 vs 1). Also fixes a latent
shard-bleed: file selectors now use exactTestFileSelectors (relative
paths were substring filters that matched sibling roots).
2. PARALLEL: full-suite mode is one 'bun test --parallel' invocation
(Bun 1.3.13). Measured semantics recorded in the header: per-file
worker isolation, standard summary, and mid-suite process.exit
surfaces as a crashed-worker FAIL with exit 1 — strictly safer than
serial, where the same exit truncates silently. No static weight
lists; --shards M --shard i keeps deterministic hash partitioning for
CI matrices (native --shard rejected: round-robin renumbers when
files land). Spawned shards get throwaway GSTACK_HOME/TMPDIR so
parallel shards can't contend on real state. Per-shard epilogue
prints files/seconds/status every run.
3. Stable indices: assignFilesToShards no longer drops empty shards, so
a shard's index depends only on the file hash and requested count —
an empty CI matrix slot is a fast no-op success, not a renumbering.
package.json 'test' now delegates to the runner (TEST_ROOTS becomes the
single source of truth for roots; slop:diff tail preserved; the runner
inherits the 30s per-test timeout the old glob passed inline).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci: Linux free-test lane — ~400 files get CI coverage for the first time
New required, secretless free-tests job: the canonical runner's single
'bun test --parallel' invocation with strict-output classification on
ubicloud-standard-8. The free suite previously ran on NO Linux CI — only
a curated Windows subset ran anywhere — so every 'tests pass' claim
about main rested on contributors running them locally.
Secretless by design (no API keys; fork PRs finally get real test
signal) and pinned by test/free-tests-workflow-wiring.test.ts: canonical
runner invoked, zero secrets.* references, pull_request never
pull_request_target, and matrix-count/--shards agreement if anyone
switches to the sharded fallback.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: map-diff selection — a touchfiles-data edit runs only what changed
Editing the eval dep-list data no longer forces the full ~$38 /
30-45min suite (measured on 21.9% of recent commits). When
touchfiles-data.ts is in the diff, selection now evaluates the BASE
version (git show -> mkdtemp -> spawnSync bun child printing the four
maps as JSON — sync because e2e-helpers selects at module scope) and
JSON-diffs per key: added entries, edited dep lists, and tier flips are
selected; keys removed from all maps are reported, never silently
dropped; a GLOBAL_TOUCHFILES edit still runs everything.
FAIL-CLOSED with named causes: missing-base-ref, git-show-failed,
import-failed, shape-mismatch each degrade to run-all and print
'selection: global — touchfiles-data changed (<cause>)' (D9 — silently
expensive beats silently wrong, but never silently). eval:select prints
'selected N of M, reason: ...' + removed tests; --base scopes the
map-diff too.
The temporary conservative GLOBAL entry for touchfiles-data.ts is gone —
its changes route through the map-diff. 23 new free tests: pure-core
fixtures, selectTests wiring incl. a poison-injection guard, and a temp
git repo exercising every fail-closed cause end-to-end.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: selection sees uncommitted work; git errors fail closed
getChangedFiles is now the deduped union of committed (base...HEAD),
staged+unstaged (git diff HEAD), and untracked (git status --porcelain
--untracked-files=all) — an agent that edits files and runs evals
BEFORE committing no longer gets the full $38 suite every time because
the committed diff looked empty. Clean tree still returns [] (run-all
by design for main-branch/periodic runs).
Git failures now THROW with the failing command, stderr, and 'set
EVALS_ALL=1 to deliberately run the full suite' — the old return []
silently became run-all, which is silently expensive. 11 new free tests
cover every source, dedupe, quoted paths, and both failure shapes via
an injectable spawn seam.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: revert GSTACK_HOME injection in the free runner — shared mutable state
The first full run under the strict runner surfaced 12 failures with one
root cause: injecting a single throwaway GSTACK_HOME per invocation made
6,900 tests share a MUTABLE scratch home. gstack-config tests wrote keys
into it; relink and update-check tests then read them (e.g. relink saw
skill_prefix left behind by a config test and produced prefixed names).
All 12 pass when run directly.
TMPDIR isolation stays (mkdtemp inside it is still per-call unique).
Tests needing GSTACK_HOME isolation mkdtemp their own per test — the
repo convention — and hermetic-env covers E2E children. The env-dump pin
now asserts GSTACK_HOME passes through UNTOUCHED so the injection can't
come back.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: rebase parity baseline to v1.64.0.0; fix capture-vs-check drift
The parity ratchet had quietly failed for 7 skills — v1.58-v1.64 growth
landed past the v1.57.7.0 anchors and nothing caught it because this
test had no CI lane (verified pre-existing: SKILL.md content is
byte-identical to origin/main). Same rebase protocol as
v1.53->v1.57.7.0; old baseline retained for the audit trail.
Root-caused a second latent bug while rebasing: captureBaseline recorded
SKELETON-ONLY bytes while the checker compares UNION bytes (skeleton +
carved sections/*.md), so a fresh capture read carved skills at ~2x
ratio (ship: 82KB captured vs 183KB checked). captureBaseline now takes
sectionedSkills and records unions for carved skills — capture and check
measure the same thing, so the NEXT rebase can't hit this. Four
CARVE_GUARDS skeleton caps re-ratcheted to current +headroom
(plan-ceo 92K, plan-eng 70K, office-hours 100K, design-consultation
70K), annotated inline.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix: package.json version matches VERSION (1.64.0.0)
v1.64.0.0 shipped with VERSION bumped but package.json left at 1.63.0.0
— the 'package.json version matches VERSION file' test fails on
origin/main today. Nothing caught it because that test had no CI lane
until this branch's free-tests job.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: fix variants-retry-after HTTP-date flake (TODOS P2)
toUTCString() truncates to whole seconds, so a +3000ms Retry-After date
could mean an effective wait of ~2001ms — flaking against the 2500ms
assertion floor ~1-2 in 9 runs under suite load. +4000ms puts the
truncation floor at 3001ms with the assertion floor safely below it.
Pulled forward from U4 because the free-tests lane is now a required
check and this flake would randomly block PRs.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: skill-fixture helper — extract SKILL.md sections, don't copy files
extractSkillSections (fence-aware H2 scanner, loud-throw on missing
sections with available-heading list), extractSkillBody (drops the
shared generated preamble), extractSkillHead (frontmatter + first 30
lines, for routing fixtures). Pinned section lists per consumer, and
free-tier real-skill pins so a gen-skill-docs heading rename fails the
FREE suite instead of a paid run. skill-fixture.ts joins
GLOBAL_TOUCHFILES (fail-safe polarity: over-select).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): review E2E fixtures extract sections — 1871 -> 207 lines
CLAUDE.md's extract-don't-copy rule, applied: the three review fixtures
carry only the sections the sql-injection/enum/design-lite prompts and
judges exercise (89% cut). Full-file copies made claude -p read 1871
lines per test — the direct cause of the 1705s worst shard (94.7% of
the 1800s kill).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): retro E2E fixtures extract sections — 1821 -> 757 lines
Keeps every section the retro flow exercises incl. base-branch detect;
drops preamble, Global Retrospective Mode, Compare Mode (58% cut).
retro-base-branch was the single slowest CI test at 224s.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): review-army fixture extracts sections — 1871 -> 650 lines
CS1's set plus Step 1.5 (PLAN COMPLETION AUDIT machinery) and Step 4.5
(army dispatch, quality_score, findings schema) that the 7 army tests
assert on. Pin test guards the three load-bearing strings.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): skillify fixtures via extractSkillBody — 63-83% smaller
Tests follow all 11 skillify steps, so the whole body stays; only the
shared generated preamble drops (skillify 1239->453, scrape 958->167).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): context-skills fixtures via extractSkillBody — 74-82% smaller
context-save 1037->267 lines, context-restore 952->168; the 8 tests
exercise full save/restore/list flows so the body stays, preamble drops.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): opus-47 discovery fixtures via extractSkillHead — ~95% smaller
Routing/fanout tests only read frontmatter + opening lines of the 14
installed skills (review 1871->54, office-hours 1706->80).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): codex runner gains sections option — review variant 88% smaller
runCodexSkill/installSkillToTempHome accept sections?: string[] routed
through extractSkillSections; codex-review-findings wired (1465->181
lines). codex-discover-skill deliberately keeps the FULL copy — its
stderr assertions validate that the real generated artifact loads.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(evals): routing fixture installs skill HEADS, not ~18 full SKILL.md
Routing reads frontmatter only; extractSkillHead per skill (root
611->48, ship 1435->54 lines). This was the single worst fixture bloat
site: one fixture dir holding ~18 full skills.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: parent-side shard skipping — a one-test diff runs 3 of 44 shards
The sharded runner spawned every shard regardless of diff; only the
child self-skipped, so a typical single-skill change still paid 44 Bun
boots + container-equivalent setup for shards with zero selected tests.
The parent now computes selection once (mirroring e2e-helpers exactly:
EVALS_ALL -> run-all, empty union -> run-all, git errors propagate the
fail-closed throw) and drops shards where no selected test name maps in.
Mapping = quoted E2E map keys in the file's source UNION keys whose dep
list registers the file (constructed-name families need the second
direction). FAIL-OPEN everywhere it matters: run-all, non-skill-e2e
files, unreadable source, zero mapped names all keep the shard — the
child filter stays authoritative, so a parent bug can only run extra.
New taxonomy status skipped-by-diff (never conflated with
never-started); selection banner prints once; --list is selection-aware.
C6 lands in the same commit: a HARD tier-alignment test — every paid
skill-e2e file must be parent-mappable or provably fail-open-safe.
Note: this change-set's 14 dep-list registrations in touchfiles-data.ts
rode along in f945c841 (concurrent-agent staging); they belong to this
change logically.
Demo: selection of one test -> 'running 3 of 44 shards, 41
skipped-by-diff'. 13 new $0 tests via injected seams.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: fix context-save-list test that was 0-for-26 ($5.28, zero passes)
Disposition for the eval store's only permanently-red test. Root cause:
the hide-other-branches assertions scanned the FULL output surface
(incl. bash tool_results), so any agent that ran ls on the checkpoints
dir — the natural first step of a list flow — surfaced all three seeded
filenames and failed, even when its user-facing listing filtered
correctly. The test punished the agent for looking at the directory.
The hide-assertions now scan the agent's FINAL TEXT (the listing the
user sees); showsMain keeps the broad surface for its documented reason.
Validated live in the final gate run rather than quarantined: the test
guards real behavior (branch filtering) and the assertion was the bug.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: wire 12 orphaned test files into the free suite (D3a)
ios-qa/daemon/test (10 files), ios-qa/scripts/gen-accessors.test.ts, and
browser-skills/hackernews-frontpage/script.test.ts ran under NO script
or CI — written coverage catching nothing. All 174 tests green on
arrival (4.6s), zero quarantines needed. TODOS P2 closed: main wired
design/test in v1.64, the variants-retry-after flake it named is fixed
on this branch, and this commit lands the remaining orphans.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: supabase-provision runs in-process — 16.5s -> 0.45s
bin/gstack-gbrain-supabase-provision (482-line bash) becomes a 26-line
bun-shebang entry over a new importable lib/gbrain-supabase-provision.ts
with an injected-deps seam (fetch/env/stdout/sleep — D7: args, never
env-mutation-before-import). The 33 spawn-per-test cases run in-process
against the same Bun.serve mocks; exactly one spawn smoke keeps the
shebang/CLI/receipt contract covered.
Byte-compat proven by a 25-case differential harness (old bash bin from
git vs new, same mock): stdout, stderr, exit codes identical across all
subcommands, JSON/plain modes, and error paths. Egress receipts stay
per-attempt, receipt-before-send, fail-closed (scanner updated:
SHELL_SINKS -> MODULE_SINKS). No-op sleep injection makes retry/backoff
paths instant.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: kill the 3,372-line zombie monolith; revive 4 never-run tests
test/skill-e2e.test.ts survived the v1.56 split as a zombie: the paid
glob needs the skill-e2e-* hyphen, so with EVALS=1 NOTHING has executed
it for ~8 releases — and it held the ONLY implementations of four
map-registered tests: review-coverage-audit (gate), plan-eng-coverage-
audit (gate), ship-triage (gate), ship-idempotency (periodic). Three
gate tests silently never ran — the exact 0%-execution class this
branch exists to kill.
Rehomed into test/skill-e2e-coverage-audit.test.ts, -triage.test.ts, and
-ship-idempotency-sdk.test.ts with bodies byte-identical modulo collector
wiring and fixture extraction (drift observed in the skills since v1.56
is DOCUMENTED in each header, not fixed — their first paid run in 8
releases must attribute failures to drift, not to this move). All 24
other monolith names were true duplicates of the split files — dropped
with the monolith. Matrix rows added to both eval workflows; the paid
glob's zombie-exclusion is now a commented regression pin.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: fix two parallelism-exposed flakes (probe re-run, live-tree census)
Both pass solo and on main but flaked under the parallel runner:
1. gstack-brain-context-load probed 'gbrain --version' PER QUERY with a
500ms budget — a cold probe on a saturated box timed out (observed
505ms), branding gbrain 'missing' for one query while siblings
passed. The probe is now memoized (availability can't change
mid-invocation) with a generous one-time 5s budget; query calls keep
the tight timeout.
2. skill-size-budget's catalog estimate read the LIVE tree, so a
concurrent worker's transient skill-shaped scratch dirs exactly
doubled it (8356 vs 4177). The ratchet now counts git-TRACKED skills
only — the catalog that ships, immune to sibling workers.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: demote 4 expensive posture tests to periodic (D2a)
design-consultation-research ($0.91/304s) and -preview ($0.89/481s) —
the two most expensive gate tests — plus office-hours-forcing-energy
(LLM-judge posture score; its sibling was already demoted) and
cso-full-audit (250s/$0.57; the targeted cso tests stay gate). Saves
~$8-12 and 10-15 min per gate run. The plan-*-finding-floor tests stay
gate deliberately: cheap insurance on the most-edited skill surface.
Housing files have no whole-file self-gates, so the runtime E2E_TIERS
filter handles both tiers; tier-alignment tripwire green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: judge default Sonnet -> Haiku 4.5 (D1a)
The 25 doc-quality judges are rubric-scoring calls — a duty Haiku is
already proven at in this repo (pty hung/working classifier,
first-task-scaffold, hermetic-canary). Tests needing a stronger judge
pass a model explicitly. Note: eval-store judge costs were hardcoded
synthetic (0.02), so no baseline distortion. Re-baselined by the
periodic run in this branch's final verification.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: SDK runner default Opus -> Sonnet (D1a)
agent-sdk-runner defaulted to Opus 4.7 while session-runner (the claude
-p path) defaulted to Sonnet — an inconsistency between the two runners,
not a decision anyone made. Unpinned tests were implicitly asserting the
expensive model. The 30+ tests that genuinely need Opus already pin it
via opts.model. Re-baselined by the periodic run in this branch's final
verification; regressors get explicit Opus pins.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* docs: CLAUDE.md tells the truth about the free suite; make-pdf gate is macOS-only
The '<2s' claim was off by two orders of magnitude (measured 454s serial
at v1.63; ~90-100s now under the parallel runner), and the bare
'bun test' guidance walked the whole repo, loading paid eval files and
missing the strict classifier. Commands now say 'bun run test' with real
numbers, document the strict-output invariant, the EVALS_JOBS /
EVALS_CONCURRENCY split, the computed detach-timeout floor, and the
required free-tests lane. make-pdf-gate drops its Linux leg (redundant
with the free lane running make-pdf tests on every PR); macOS rendering
coverage stays.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: catalog ratchet reads committed content; SDK unit pins follow D1a default
Two follow-ups from the verification runs:
1. skill-size-budget's catalog estimate still flaked under --parallel
(8356, then 8041, vs 4177 solo) even after filtering to tracked
skills: sibling workers REGENERATE real SKILL.md files mid-run, so
any live-tree read is a moving target. The ratchet now reads each
tracked skill's frontmatter from git show HEAD: — the catalog that
ships — which no concurrent worker can perturb.
2. agent-sdk-runner unit pins asserted the old Opus default through the
default-flow fixtures; flipped to the Sonnet default (the explicit-
override pass-through pins keep Opus — that path is unchanged).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(browse): SIGKILL abandoned Chromium on close-race timeout (suite wedge)
close()'s launched-mode path raced browser.close() against 5s and on
timeout ABANDONED the child: this.browser nulled, process handle lost,
Chromium alive holding keep-alive connections into test servers whose
stop() then waits forever. Reproduced twice as an intermittent (~50%)
whole-suite wedge — a 44min 0.1%-CPU hang pinned by a leaked LISTEN
socket, and a 400s hang with commands.test.ts teardown in flight.
The child handle is now captured BEFORE the race and SIGKILLed on
race-timeout (launched mode only; headed keeps context.close). Race
timers are unref'd so a successful close stops pinning the caller's
event loop for the window. The four browse test servers force-close
keep-alives (stop(true)) as belt-and-braces.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: delete two dead-architecture security contract tests
browse/test/security-source-contracts.test.ts and sidebar-security.test.ts
read browse/src/sidebar-agent.ts at module scope — a file deleted (on main
too) when the sidebar chat-queue path was ripped in favor of the terminal
PTY. Both files have errored on load ever since: the old truncating suite
never surfaced it, and no CI lane ran them. Their subjects (queue-spawn
canary injection, preSpawnSecurityCheck, queued args, chat system prompt)
no longer exist; server.ts retains processAgentEvent only in a comment.
Live security coverage continues in security.test.ts (canary/verdict),
content-security.test.ts (L1-L3), server-sanitize-surrogates.test.ts,
and the security-bench suite. If the terminal-agent path should inherit
any of the deleted contracts, that is a separately scoped piece of work
against the component that actually exists.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(redact): calibrate placeholder recognition for code and doc shapes
Three pushed-secret false positives blocked this branch's push; each is
now recognized as a placeholder in the url_with_password/basic_auth_url
validators, with real passwords still blocking (all pinned):
- ${camelCase} JS template interpolations (the old check only skipped
uppercase env-style ${DB_PASS}, so the supabase-provision bash->TS
port's `postgresql://${dbUser}:${dbPass}@...` flagged as two
pushed secrets).
- The literal PASSWORD/pass placeholder in URL-format doc comments.
- The provision lib's doc comments now use <PASSWORD>/PASSWORD forms.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: opt-in gate for live-playwright ML tests; ios-qa build hygiene
security-live-playwright's L4 tests dlopen onnxruntime inside a bun
--parallel worker whenever the dev box has a warm model cache — the
source of the intermittent 'panic: Segmentation fault' + crashed-worker
retries (and likely the residual run wedges). Same SECURITY_BENCH=1
opt-in as security-bench.test.ts; the L1-L3 tests in the file still run
everywhere.
Also: gitignore the ios-qa gen-accessors-tool Swift .build/ output (a
side-effect of running its tests that kept polluting git status) and
commit its Package.resolved so tool builds resolve reproducibly.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: free runner output contract — name the failure, quiet the noise
Diagnosing a red run used to mean re-running with output captured to a
file and grepping past ~1000 lines of tab-close spam and ASCII art —
several runs today ended with no way to even NAME the failing test, and
a wall-timeout kill said nothing about which file wedged.
New contract: the full child stream ALWAYS lands in a per-run log file
(path printed up front); the console shows only runner lines, (fail)
results, crash markers, and the terminal summary (--verbose restores
the firehose; the strict classifier consumes the full stream in every
mode). After every run a stable epilogue names the outcome:
[test:free] FAIL — k failing test(s) in j file(s), c crashed
worker(s). Full log: <path>
✗ <file> — <test name>
⚠ crashed+retried: <file>
⏱ in flight at kill: <files> (timeout only — the wedge suspects)
Attribution rides bun --parallel's per-file output grouping
(ANSI-stripped — color codes defeated a plain grep today). 12 new pins:
epilogue formats, crash surfacing, quiet/verbose console policy, log
completeness, in-flight-at-kill on a real hang.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: quarantine 5 pre-existing env failures individually (receipts in-file)
Three snapshot tests (stale-ref error, snapshot -D diff, annotation
cleanup) and two extension-sender-auth behavioral tests fail identically
on origin/main v1.64.1.0, solo, on dev machines — verified per the blame
protocol. Main's CI lane skip-lists both FILES wholesale; quarantining
only the five failing tests keeps the other 60 guarding. Each carries
the un-skip condition.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: stealth-webdriver launch gets parallel-load headroom (120s)
Playwright's default 30s launch timeout dies under the full-suite
--parallel run when ~400 workers contend for Chromium launches — bun
reports the hook death as an '(unnamed)' 30006ms failure (named on
sight by the new runner epilogue). Both launch sites get explicit 120s
timeouts; the runner's external wall-clock still bounds the ceiling.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: free-runner wall timeout 15min -> 6min (faster wedge diagnosis)
The suite completes in ~100-160s; a wedge used to mean 15 minutes of
silence before the kill-and-name epilogue fired. 6min keeps ~3.5x
headroom over the slowest observed clean run while naming wedge
suspects in minutes. --wall-timeout <secs> overrides per run.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: run worker-hostile files in a serial child (first entry: security-live-playwright)
The residual full-suite wedge, named by the new epilogue: Bun 1.3.13
segfaults running browse/test/security-live-playwright.test.ts in a
--parallel worker ('panic: Segmentation fault ... a bug in Bun'), and
the crashed-worker retry then wedges the whole invocation past the wall
clock. The file passes serially.
New WORKER_HOSTILE placement list: full-suite mode excludes listed files
from the parallel invocation and runs them in their own strict-classified
serial child afterward — execution placement, not a skip; each entry
carries its reason and removal condition.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: gate compare-board's file-level hooks too — the intermittent staller
Skipped describes do NOT skip file-level hooks: the quarantined
compare-board file still ran its top-level beforeAll (PNG fixtures +
Bun.serve + a BrowserManager launch — exactly the 'needs a
display-shaped env' code) on every run, and under parallel load that
setup wedges. Caught red-handed by the runner's in-flight-at-kill
epilogue: '⏱ in flight at kill: browse/test/compare-board.test.ts'.
This was the suite's intermittent staller. Hooks now honor the same
GSTACK_COMPARE_BOARD_TESTS gate; the gated file drops from 3.7s of live
setup to 0.4s of pure skips.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: full suite runs as N shard processes; scrub spec-sync child env
Two fixes from the wedge-hunt endgame:
1. Full-suite mode switches from one 'bun test --parallel' invocation to
N concurrent shard PROCESSES, serial within each (the paid runner's
proven model; N = min(6, cpus-2)). The single-invocation strategy hit
three distinct Bun 1.3.13 worker pathologies in one day — a segfault
whose crashed-worker retry wedged the run, a quarantined file's
still-running file-level hooks stalling a worker, and spawn-heavy
files hanging under load — and each one stalled the WHOLE invocation.
Process shards isolate any wedge to its own shard. First full run
under this model: no wedge, six epilogues, one real failure named.
WORKER_HOSTILE stays as the paper trail; --parallel remains available
per-shard for a future Bun.
2. That one real failure: spec-template-sync regenerates SKILL.md via a
child that inherited the shard process's env — an earlier test's
GSTACK_*/GBRAIN_* mutations changed generator output (failed in-suite,
passed solo on an identical tree). The child now gets a scrubbed env:
generator output must be a function of the templates, not of whichever
test ran before.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: tree-mutating tests run after the parallel shards; scrub relink env
The flake family's root cause, finally: five test files REGENERATE
shared repo artifacts in place (catalog-mode-full rewrites every
SKILL.md in full-catalog mode; spec-sync and idempotency regenerate all
skills; gen-skill-docs and skill-validation rewrite .agents/). Any
concurrent shard reading those files sees a moving target — this one
family produced the exactly-doubled catalog estimate, the golden-file
drift, and the spec-sync mismatch chased earlier today. Full-suite mode
now runs TREE_MUTATING files in one serial shard AFTER the parallel
shards complete; CI's matrix is unaffected (per-runner checkouts).
Also: relink's run() helper spread process.env into its children, so a
sibling file's leaked GSTACK_HOME made the 'fresh install' test see a
neighbor's skill_prefix. GSTACK_HOME is now dropped unless the test
passes it explicitly.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: gbrain-detection-override joins TREE_MUTATING (mutator #6)
It regenerates SKILL.md in place with --respect-detection (the gbrain
variant adds ~1-3KB per carved skeleton) and git-restores afterward —
its own header documents the approach. During that window the parity
suite in a concurrent shard read inflated skeletons and failed 4 caps.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test: tree-ratchet readers join the serial phase (quiet tree by construction)
Two consecutive runs failed the parity caps with byte-identical inflated
skeletons (+~2KB gbrain-variant blocks) while the tree was clean before
and after — some concurrent regen window keeps escaping the mutator
census. Rather than hunt every present and future mutator, the tests
that MEASURE the shared tree (parity caps, size budgets, carve guards)
now run in the serial phase after the parallel shards: a quiet tree by
construction, immune to any regen we haven't found.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* evals: judge default back to Sonnet — Haiku regressed the rubric family (A/B receipts)
The partial-diff rehearsal was the Haiku judge default's first live run
and it failed all three selected doc-rubric judges. Controlled A/B on
the identical health-rubric prompt: Haiku 2/2/2 vs Sonnet 4/3/4, both
with coherent reasoning — Haiku is simply a harsher grader on
long-document rubrics, and every >=4 threshold in skill-llm-eval was
calibrated against months of Sonnet baselines. Per D1a's
pin-on-regressors protocol the default reverts; a new
GSTACK_EVAL_MODEL_JUDGE override makes future recalibration a one-var
experiment. Haiku keeps the classifier-grade duties (pty hung/working,
warmup, distill via lib/eval-model.ts) and D1a's capture->Sonnet stands.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(test-runner): per-origin classifier buffers — interleaved pipes can't shear lines
stdout and stderr are independent pipes; a chunk from one can arrive
between two halves of a line from the other. The single shared
pending-buffer glued those fragments into garbled lines: a sheared
(fail) line went uncounted (defeating the exit-0-with-failures
backstop) and a sheared terminal summary read as truncation.
Counters stay shared; line assembly is now per-stream, and both
runners tag the stream origin. Also drops the dead ChildProcess
type import left by the killProcessGroup move.
* fix(test-runner): real carve-guard keys in TREE_MUTATING; census pins; size-scaled wall deadlines
TREE_MUTATING listed 'test/carve-guard-checks.test.ts' — a file that
has never existed (the real ratchet readers are
carve-guard-completeness and carve-section-ordering), so the intended
serialization was silently absent. New census pin tests fail on any
key that doesn't name a real free test file, and on a TEST_ROOTS
entry that stops contributing files. Full-suite wall deadlines now
scale with shard size (max(6min, files x 5s)) so a jobs=1 machine or
the ~130-file Windows shards can't false-timeout a healthy run;
explicit --wall-timeout disables scaling. Stale --parallel wording in
the dry-run message, jsdoc, and the TREE_MUTATING ordering comment
corrected to the shipped process-shard model.
* fix(evals): selection under-selection fixes — duplicate keys, self-paths, quotePath
Three under-selection holes: (1) duplicate E2E_TOUCHFILES keys
(ship-plan-completion/-verification) — JS keeps the LAST duplicate, so
the earlier dep lists were dead; pair deleted and a duplicate-key scan
added to the literal-only tripwire. (2) The five rehomed e2e files
didn't list themselves in their own dep lists, so editing the test
never selected it. (3) git C-escapes non-ASCII paths without
core.quotePath=false, so an accented filename matched no glob and
deselected its tests. Also updates the stale --retry cost comment.
* fix(evals): destructive-actions guard actually inspects Bash commands
The rehomed guard filtered on typeof input === 'string', but
session-runner records tool inputs as objects ({command} for Bash) —
the filter matched nothing and the assertion could never fail, even
against a real 'git push'. Now extracts the command from the object
shape, same as the usedGitDiff check above it.
* fix(redact): interpolation allowance can't swallow a real $word password
The placeholder calibration used optional braces on both sides, which
also suppressed bare $lowercase — a real password starting with '$'
would have passed the HIGH gate. Interpolation now means ${identifier}
(braced, any case) or bare $UPPER_SNAKE only; both connection-string
patterns share one validator so they can't drift. Pins added for the
bare-$word block, $UPPER allowance, and mismatched-brace block.
* fix(gbrain): wait --timeout validates up front instead of polling forever on NaN
Number('abc') is NaN, NaN comparisons are always false, and the
poll loop never hit its deadline — an infinite 5s loop where the bash
predecessor errored immediately. die(2) at parse time, with a test.
* ci: least-privilege tokens on the two lanes that execute PR-controlled code
free-tests runs PR code (install lifecycle scripts + the suite) with
whatever the repo-default GITHUB_TOKEN grant is, persisted into
.git/config by checkout. Now: permissions contents:read,
persist-credentials false, pinned by the wiring test. actionlint gets
the same treatment plus a digest pin on the third-party Docker Hub
image (a tag is repointable with no GitHub-side audit trail, and the
image sees the mounted checkout). restore-keys added to both caches so
a lockfile bump warms from the previous cache; stale --parallel header
wording corrected.
* test(browse): unit coverage for the close() SIGKILL fallback
The wedge fix (capture the Chromium child before the close race,
SIGKILL on timeout) shipped without a test of the branch it added —
the coverage audit flagged it as the diff's one regression-gap. The
5s race window becomes an injectable closeRaceMs field, and four unit
tests pin: SIGKILL on hang, no SIGKILL on clean close, no SIGKILL on
an already-exited child, SIGKILL on a rejecting close.
* docs: CLAUDE.md describes the shipped shard-process model, not the abandoned --parallel probe
* fix(test-runner): cancellation terminates the run; win32 kills the whole tree
Installing SIGINT/SIGTERM forwarders suppresses Node's default
terminate-on-signal, so a cancelled run killed the current child and
kept LAUNCHING shards — observed as paid runs continuing to burn API
spend after Ctrl-C (codex adversarial, repro'd ALIVE_AFTER_SIGTERM).
The first signal now also schedules the parent's own exit after the
children's SIGKILL grace, and both shard pools consult
isTerminationRequested() before taking new work. On win32,
killProcessGroup uses taskkill /T /F — detached:true creates no
killable group there, and a bare child.kill orphaned every grandchild
(ports, locks, and the inherited pipes that kept close from firing).
Also: the tree-mutating serial shard prints dirty generated artifacts
when it dies mid-regeneration, and --shard CI-matrix mode gets the
same size-scaled wall deadline as full-suite mode.
* fix(evals): preflight fails fast on spawn error, timeout, and exit 127
The ping only grepped stdout for two connection strings — a missing
claude binary, a 30s timeout kill, or command-not-found all returned
'ok', and the fleet then burned ~30 shard timeouts discovering the
outage one child at a time. Cross-model finding (testing specialist +
codex adversarial). Other non-zero exits stay deliberately fail-open:
a flaky preflight must not block a runnable suite; pinned both ways.
* fix(redact): lowercase 'password'/'pass' at the URL-password position blocks
The case-insensitive placeholder words waved postgres://admin:password@host
through the HIGH gate as a doc placeholder (codex adversarial,
verified zero findings pre-fix). URL-password position is now stricter
than generic placeholder detection: ALL-CAPS doc convention
(USER:PASSWORD), ${identifier} interpolations, bare $UPPER_SNAKE, and
structural shapes (<your-password>) suppress; lowercase dictionary
words block. Pinned in both directions.
* fix(gbrain): DSNs percent-encode the password; body reads retry; stdout drains
Three codex-adversarial findings in the provision port: (1) raw DB_PASS
interpolation — a reserved character (/ # ? % @) restructured the URI,
provisioning succeeded, and every consumer then failed to parse the DSN
(unusable billable orphan); now encodeURIComponent, round-trip pinned.
(2) await res.text() sat outside the transport try — a server that sent
headers then reset the stream was an uncaught exit 1 instead of a
retry-then-exit-8. (3) The bin entrypoint called process.exit() after
unawaited stdout writes, truncating piped JSON; exitCode lets writes
drain.
* fix(evals): selection-path helpers join GLOBAL_TOUCHFILES; base-branch keys self-register
The three-file split moved test-selection.ts into the globals but
dropped the facade — an edit to test/helpers/touchfiles.ts (executable
selection-path code imported by every consumer) selected ZERO paid
tests, the exact invisible-non-execution class this branch exists to
kill (claude adversarial, finding 1). e2e-helpers.ts (the harness every
paid test imports) and paid-test-set.ts (paid-vs-free classification)
had the same gap. The review/ship base-branch keys also register
test/skill-e2e-review-attribution.test.ts so editing those tests
selects them.
* ci(free-tests): PR-number concurrency, failure-log artifact, main-push runs
Three red-team/adversarial findings on the new required lane:
(1) concurrency keyed on bare head_ref — two forks with the same
branch name shared one group, so a push to fork B cancelled fork A's
in-flight REQUIRED check (merge-pipeline DoS with no code fault); key
on the PR number. (2) The runner's full logs die with the runner in
os.tmpdir() — a red check named WHICH test failed but never why;
upload the shard logs as an artifact on failure. (3) PR-only trigger
meant two individually-green PRs could merge into a red main with
nothing running the suite there; add push: branches: [main].
* ci: PR-number concurrency keying on the eval and Windows lanes too
Same fork-branch-name collision as free-tests.yml: bare head_ref
carries no owner prefix, so same-name branches from different forks
shared a cancel-in-progress group.
* test(evals): retro E2E passes require the report on disk
Both retro tests passed with zero work product: error_max_turns
counted as success and the content assertion was guarded by
fs.existsSync — a run that burned 30 turns and wrote nothing recorded
green (red team). The report is now load-bearing for pass/fail.
* chore: bump version and changelog (v1.66.0.0)
Test/evals/CI speedup pass: release summary + itemized changes in
CHANGELOG.md; TODOS.md marks the free-suite exit-code P1 complete and
files the review-army follow-ups.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(redact): fully-braced ${...} interpolations are code, whatever they contain
The identifier-only braced form flagged the DSN builder's own
${encodeURIComponent(dbPass)} call site as a pushed secret — a scan
that cries wolf on the fix for the previous finding. Any ${...}
spanning the whole password segment is template code; bare $word
stays uppercase-only so $hunter2 still blocks. The mismatched-brace
negative fixture assembles at runtime so this file's own pushed bytes
carry no blockable URL shape.
* test(gbrain): assemble the pooler expected-URL from parts (scan-clean pushed bytes)
* docs: sync docs for v1.66.0.0 (test/evals/CI speedup)
CONTRIBUTING.md, AGENTS.md, and ARCHITECTURE.md still taught bare
`bun test` for the suite; the shipped runner deprecates it (walks the
whole repo, loads paid eval files, misses the strict classifier). All
suite-level references now say `bun run test`, the Tier 1 section
describes the strict shard runner (~90-100s, --verbose, --wall-timeout),
the sharded paid-runner paragraph documents diff-based shard skipping
and the EVALS_JOBS / EVALS_CONCURRENCY split, the Tier 3 row points at
the actual judge-only invocation, and GSTACK_EVAL_MODEL_JUDGE is
documented at the judge it overrides.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* ci(free-tests): restore the PR-number concurrency + failure-log artifact; truth-fix stale comments
The workspace-revert incident that hit CHANGELOG/TODOS mid-ship also
caught free-tests.yml between edits: commit 8d6c2ff8's message claims
PR-number concurrency + artifact upload + main-push runs, but only the
push trigger survived to the commit (caught by the /document-release
doc-vs-code audit). Both re-applied. Also: eval-model.ts header said
capture defaults to Opus (it's Sonnet per D1a), paid-shards' header
pinned a stale 44/63 shard census, and two CHANGELOG phrases
over-claimed ('six' -> 'up to six' shard processes; retry-1 scoped to
retry-bearing paid paths).
* ci: setup-buildx before every cache-exporting image build
First live run of the cache trio failed at flag-parse time: the
default buildx `docker` driver hard-errors on cache-to registry
export ('Cache export is not supported for the docker driver'), which
failed build-image on PR #2593 and skipped the entire gate eval
matrix behind it. docker/setup-buildx-action creates the
docker-container builder that supports registry cache export; all
three build sites (evals, evals-periodic, ci-image) get it.
* fix(browse): Xvfb identity is argv[0]'s basename, not a cmdline substring
First Linux CI run: isOurXvfb identified the TEST RUNNER as our Xvfb —
the suite's own argv contains 'xvfb.test.ts', the substring match over
the whole cmdline passed, and the start-time check matched because the
pid was real. Any process whose ARGUMENTS mention xvfb (a runner, an
editor) was killable — the sibling-kill class the identity check
exists to prevent. Identity now rests on argv[0]'s basename ('Xvfb'),
with a sh-$0 regression pin. isDisplayFree falls back to the X
socket/lock files when xdpyinfo isn't installed (x11-utils is absent
on some images that ship Xvfb).
* fix(test-runner): strip GHA ::group:: wrappers before file attribution
On GitHub Actions bun wraps each file's log section in ::group::. The
un-stripped header failed FILE_HEADER_RE, failures attributed to the
PREVIOUS file, and the terminal recap's re-printed (fail) lines landed
under a phantom second file — the first Linux run reported 5 real
failures as 10 across 2 files (one of them innocent). Strip the prefix
before matching; the existing file+test dedupe then absorbs the recap.
* test: first-Linux-run environment fixes — bun-only PATH shim, claude gate, darwin-scoped pdf gates
Three environmental assumptions the Linux lane exposed:
(1) gbrain-detect's deterministic SAFE_PATH lacked the bun runtime, so
every env-shebang spawn exited 127 on CI; a scratch dir holding ONLY a
bun symlink joins the PATH (appending bun's real dir would leak its
siblings — dev boxes keep gbrain there too).
(2) host-config's 'detect finds claude' assumed a claude binary; the
secretless lane deliberately has none — gated on Bun.which.
(3) The four make-pdf render gates hard-required prerequisites on ANY
CI, but the make-pdf gate workflow is macOS-only by decision and the
Linux lane doesn't build dist/pdf — hard-require scoped to darwin.
* ci(free-tests): run the suite under xvfb-run
Headed-browser tests (handoff, extension sidepanel DOM) need a real
DISPLAY; the first Linux run died on Playwright's 'headed browser
without an XServer' banner. xvfb-run -a provides the display; x11-utils
ships xdpyinfo for display probing.
* fix(test-runner): bun's headerless failure recap can't invent a phantom failing file
Round-3 CI showed the remaining half of the recap bug: bun prints
'N tests failed:' then re-prints every (fail) line with NO file
headers, so they attributed to the stale currentFile — an innocent
file (test/uninstall.test.ts) was charged with another file's 5
failures. The recap marker now ends attribution (currentFile=null,
chunk closed) and recap re-prints of already-recorded test names
dedupe; a recap-only failure the main run never attributed still
records, unattributed, as belt and braces.
* test(browse): sidepanel DOM suite launches with --no-sandbox on CI + console capture
The suite's raw chromium.launch had no --no-sandbox — every browse
test that goes through gstack's launcher (which always passes it)
survived the Linux lane, while this file's sandboxed renderer died on
first navigation: waitForFunction hung to the 15s test timeout, then
every newContext failed with Target.createBrowserContext. Also wires
pageerror/console-error capture at all six pages so a page-side
failure reads as itself in CI logs instead of a bare timeout.
* test(browse): delete the sidepanel security-DOM suite — it tests UI removed in v1.14
Another member of the never-ran class: the file skipped everywhere
(Playwright chromium absent locally, no Linux CI until this branch),
so it rotted invisibly through THREE contract changes — the v1.63
/extension-token bootstrap, the endpoint growth (/memory,
/pty-session, /sse-session), and finally the v1.14 sidebar-REPL
rewrite that removed the security shield/banner UI it asserts on
(#security-shield survives in sidepanel.html as a dead hidden stub
with no JS driver; sidepanel.js:87 and :1317 document the removal).
The Linux lane executed it for the first time and it can never pass:
the behavior is gone. The L1-L3 security filters it name-checked stay
covered by the ~83 unit/behavioral security tests. The free-tests
lane also vendors xterm assets (bun run vendor:xterm) so the
sidepanel terminal scripts load for any future DOM coverage.
* ci(evals): per-row retry override — two receipted rows keep the third attempt
Three PR rounds of receipts: pty-plan-smoke failed attempt 2 in two
consecutive rounds with ROTATING members (plan-design-review, then
plan-eng-review) and e2e-workflow's document-release timed out on
attempt 2 in round 4 — while both families pass on branches still
running three attempts, and every other row stayed green at --retry 1
across all rounds. Matrix rows gain an optional retries field
(default 1); only these two rows set 2, keeping the measured
retry-amplification win everywhere else.
* test(windows): curate the seven POSIX-bound files the expanded lane surfaced; fix flag-utils path embedding
First full run of the expanded Windows lane (13 -> ~258 files, PR #2593
run 31918591602) failed in exactly 8 files. One was a real test bug,
fixed: design-flag-utils embedded a raw Windows ROOT into a bun -e
string where backslashes act as escapes (D:\a\gstack imported as
D:agstack) — forward slashes work on every platform. The other seven
are POSIX-bound in ways the content patterns cannot see (sed/ln/bash
ARE their subject, a shebang shim arrives via variable, wall-clock
retry bounds on the slowest runner) — each gets a receipted
KNOWN_WINDOWS_INCOMPATIBLE entry, and the census pin now covers that
list so a renamed file fails the suite instead of silently keeping a
stale exclusion.
* test(windows): curate skill-census + browser-manager-unit; surface unhandled errors in the epilogue
Round-2 Windows census (zero failing TESTS — the first curation wave
held): shard 1 failed on an unhandled module-load throw in
skill-census (the skills-tree symlink layout needs Developer Mode CI
runners lack) and shard 2 wedged to its wall deadline inside
browser-manager-unit — both get receipted exclusions; macOS + Linux
lanes keep covering the files. The unhandled-error class also exposed
an epilogue gap: it fails the shard via the strict classifier but
produces no (fail) lines, so the epilogue read 'FAIL — 0 failing
test(s)' with no culprit. The reporter now attributes each
'# Unhandled error between tests' marker to its chunk and the FAIL
line carries the count.
* docs: file the two Windows-lane follow-ups (browser-manager wedge, skill-census symlinks)
* test(windows): round-3 curation — seven files the round-2 wedge had been truncating
The browser-manager-unit wedge was cutting shard 2 short, so each
Windows round revealed the next segment of never-run files. With the
wedge excluded, shard 2 completes (50s) and shows its real failures:
seven more POSIX-environment files (PID/cmdline identity probing,
bash scripts as the subject under test, env-scrubbed bun spawns).
Shards 1 and 3 (including all tree-mutators) now PASS on
windows-latest — this should be the fixed point: ~234 files of real
Windows coverage vs the 13 hand-picked before.
* test(windows): round-4 curation (spawnSkill env, symlink fixtures) + shard-log artifact
Shard 2 ran all 132 files with zero (fail) lines yet bun exited 1 —
unhandled errors in a shape neither counter names, and the Windows
lane had no log artifact to attribute them. Statically attributed and
excluded: browser-skill-commands (spawnSkill spawns bun with a
constructed env; resolution fails under Windows spawn) and
security-audit-r2 (evil-link symlink fixtures need Developer Mode).
The lane now uploads its shard logs on failure like free-tests.yml,
with os.tmpdir() pointed at runner.temp so the glob can find them.
* evals: Opus pin on the spec AUQ-matrix entry — D1a regressor, receipts in-file
The periodic re-baseline for the capture default (Opus -> Sonnet)
found exactly one regressor across the seven-entry AUQ behavioral
matrix: spec failed twice under Sonnet ('never reached a question in
budget', 242s) while its six siblings passed; the controlled Opus
re-run passed cleanly (7/7 format, substance 5, 160s), and a second
run through the new per-entry model plumbing confirms. MatrixSkill
gains an optional model field wired into captureFirstAuq; only spec
sets it. TODOS gains the re-baseline receipts for the never-baselined
periodic tail (three setup-gbrain files + ship-idempotency, all
local-only).
* test(evals): scope-gate assertion carries its evidence tail; file the detector-flake TODO
The plan-design-review member fails ONLY scopeGateQuestionObserved
intermittently on unchanged code (PR #2593: red rounds 3/11 + rerun,
green rounds 5/6 — every attempt terminal, no plan-mode leak), and a
bare Expected-true/Received-false is undiagnosable from CI logs. The
check now throws with the last-2KB visible evidence, so the next
failure distinguishes a detector-sensitivity miss from a real silent
bypass. TODO filed with the full receipt trail.
* test(evals): review-dashboard-via budget 300s -> 360s — third ratchet of the same contention story
PR #2472 documented the 180s deterministic 0-turn startup timeouts and
ratcheted to 300s; PR #2593 hit 302s timeouts on attempt 2 in two
consecutive runs while five sibling rounds passed — marginal at 300s
under 40-way in-shard concurrency. Same headroom its contention-class
sibling (retro-base-branch) carries; outer bun timeout rises to 480s.
* test(evals): document-release budget 180s -> 300s — same contention ratchet, receipts in-file
Timed out at exactly 180s on its final attempt twice on PR #2593
(rounds 4 and 13) while passing four other rounds — a 30-turn
multi-step doc workflow is marginal at 180s under 40-way in-shard CI
concurrency. Same story and same fix as review-dashboard-via and
retro-base-branch; outer bun timeout rises to 360s.
* docs: file the systemic in-shard-concurrency follow-up behind the timeout-flake family
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
gstack
"I don't think I've typed like a line of code probably since December, basically, which is an extremely large change." — Andrej Karpathy, No Priors podcast, March 2026
When I heard Karpathy say this, I wanted to find out how. How does one person ship like a team of twenty? Peter Steinberger built OpenClaw — 247K GitHub stars — essentially solo with AI agents. The revolution is here. A single builder with the right tooling can move faster than a traditional team.
I'm Garry Tan, President & CEO of Y Combinator. I've worked with thousands of startups — Coinbase, Instacart, Rippling — when they were one or two people in a garage. Before YC, I was one of the first eng/PM/designers at Palantir, cofounded Posterous (sold to Twitter), and built Bookface, YC's internal social network.
gstack is my answer. I've been building products for twenty years, and right now I'm shipping more products than I ever have. In the last 60 days: 3 production services, 40+ shipped features, part-time, while running YC full-time. On logical code change — not raw LOC, which AI inflates — my 2026 run rate is ~810× my 2013 pace (11,417 vs 14 logical lines/day). Year-to-date (through April 18), 2026 has already produced 240× the entire 2013 year. Measured across 40 public + private garrytan/* repos including Bookface, after excluding one demo repo. AI wrote most of it. The point isn't who typed it, it's what shipped.
The LOC critics aren't wrong that raw line counts inflate with AI. They are wrong that normalized-for-inflation, I'm less productive. I'm more productive, by a lot. Full methodology, caveats, and reproduction script: On the LOC Controversy.
2026 — 1,237 contributions and counting:
2013 — when I built Bookface at YC (772 contributions):
Same person. Different era. The difference is the tooling.
gstack is how I do it. It turns Claude Code into a virtual engineering team — a CEO who rethinks the product, an eng manager who locks architecture, a designer who catches AI slop, a reviewer who finds production bugs, a QA lead who opens a real browser, a security officer who runs OWASP + STRIDE audits, and a release engineer who ships the PR. Twenty-three specialists and eight power tools, all slash commands, all Markdown, all free, MIT license.
This is my open source software factory. I use it every day. I'm sharing it because these tools should be available to everyone.
Fork it. Improve it. Make it yours. And if you want to hate on free open source software — you're welcome to, but I'd rather you just try it first.
Who this is for:
- Founders and CEOs — especially technical ones who still want to ship
- First-time Claude Code users — structured roles instead of a blank prompt
- Tech leads and staff engineers — rigorous review, QA, and release automation on every PR
Quick start
- Install gstack (30 seconds — see below)
- Run
/office-hours— describe what you're building - Run
/plan-ceo-reviewon any feature idea - Run
/reviewon any branch with changes - Run
/qaon your staging URL - Stop there. You'll know if this is for you.
Install — 30 seconds
Requirements: Claude Code, Git, Bun v1.0+, Node.js (Windows only)
Step 1: Install on your machine
Open Claude Code and paste this. Claude does the rest.
Install gstack: run
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setupthen add a "gstack" section to CLAUDE.md that says to use the /browse skill from gstack for all web browsing, never use mcp__claude-in-chrome__* tools, and lists the available skills: /office-hours, /plan-ceo-review, /plan-eng-review, /plan-design-review, /design-consultation, /design-shotgun, /design-html, /review, /ship, /land-and-deploy, /canary, /benchmark, /browse, /connect-chrome, /qa, /qa-only, /design-review, /setup-browser-cookies, /setup-deploy, /setup-gbrain, /retro, /investigate, /document-release, /document-generate, /codex, /cso, /autoplan, /plan-devex-review, /devex-review, /careful, /freeze, /guard, /unfreeze, /gstack-upgrade, /learn. Then ask the user if they also want to add gstack to the current project so teammates get it.
Step 2: Team mode — auto-update for shared repos (recommended)
From inside your repo, paste this. Switches you to team mode, bootstraps the repo so teammates get gstack automatically, and commits the change:
(cd ~/.claude/skills/gstack && ./setup --team) && ~/.claude/skills/gstack/bin/gstack-team-init required && git add .claude/ CLAUDE.md && git commit -m "require gstack for AI-assisted work"
No vendored files in your repo, no version drift, no manual upgrades. Every Claude Code session starts with a fast auto-update check (throttled to once/hour, network-failure-safe, completely silent).
Swap required for optional if you'd rather nudge teammates than block them.
OpenClaw
OpenClaw spawns Claude Code sessions via ACP, so every gstack skill just works when Claude Code has gstack installed. Paste this to your OpenClaw agent:
Install gstack: run
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/.claude/skills/gstack && cd ~/.claude/skills/gstack && ./setupto install gstack for Claude Code. Then add a "Coding Tasks" section to AGENTS.md that says: when spawning Claude Code sessions for coding work, tell the session to use gstack skills. Include these examples — security audit: "Load gstack. Run /cso", code review: "Load gstack. Run /review", QA test a URL: "Load gstack. Run /qa https://...", build a feature end-to-end: "Load gstack. Run /autoplan, implement the plan, then run /ship", plan before building: "Load gstack. Run /office-hours then /autoplan. Save the plan, don't implement."
After setup, just talk to your OpenClaw agent naturally:
| You say | What happens |
|---|---|
| "Fix the typo in README" | Simple — Claude Code session, no gstack needed |
| "Run a security audit on this repo" | Spawns Claude Code with Run /cso |
| "Build me a notifications feature" | Spawns Claude Code with /autoplan → implement → /ship |
| "Help me plan the v2 API redesign" | Spawns Claude Code with /office-hours → /autoplan, saves plan |
See docs/OPENCLAW.md for advanced dispatch routing and the gstack-lite/gstack-full prompt templates.
Native OpenClaw Skills (via ClawHub)
Four methodology skills that work directly in your OpenClaw agent, no Claude Code session needed. Install from ClawHub:
clawhub install gstack-openclaw-office-hours gstack-openclaw-ceo-review gstack-openclaw-investigate gstack-openclaw-retro
| Skill | What it does |
|---|---|
gstack-openclaw-office-hours |
Product interrogation with 6 forcing questions |
gstack-openclaw-ceo-review |
Strategic challenge with 4 scope modes |
gstack-openclaw-investigate |
Root cause debugging methodology |
gstack-openclaw-retro |
Weekly engineering retrospective |
These are conversational skills. Your OpenClaw agent runs them directly via chat.
Other AI Agents
gstack works on 10 AI coding agents, not just Claude. Setup auto-detects which agents you have installed:
git clone --single-branch --depth 1 https://github.com/garrytan/gstack.git ~/gstack
cd ~/gstack && ./setup
Or target a specific agent with ./setup --host <name>:
| Agent | Flag | Skills install to |
|---|---|---|
| OpenAI Codex CLI | --host codex |
~/.codex/skills/gstack-*/ |
| OpenCode | --host opencode |
~/.config/opencode/skills/gstack-*/ |
| Cursor | --host cursor |
~/.cursor/skills/gstack-*/ |
| Factory Droid | --host factory |
~/.factory/skills/gstack-*/ |
| Slate | --host slate |
~/.slate/skills/gstack-*/ |
| Kiro | --host kiro |
~/.kiro/skills/gstack-*/ |
| Hermes | --host hermes |
~/.hermes/skills/gstack-*/ |
| GBrain (mod) | --host gbrain |
~/.gbrain/skills/gstack-*/ |
Want to add support for another agent? See docs/ADDING_A_HOST.md. It's one TypeScript config file, zero code changes.
See it work
You: I want to build a daily briefing app for my calendar.
You: /office-hours
Claude: [asks about the pain — specific examples, not hypotheticals]
You: Multiple Google calendars, events with stale info, wrong locations.
Prep takes forever and the results aren't good enough...
Claude: I'm going to push back on the framing. You said "daily briefing
app." But what you actually described is a personal chief of
staff AI.
[extracts 5 capabilities you didn't realize you were describing]
[challenges 4 premises — you agree, disagree, or adjust]
[generates 3 implementation approaches with effort estimates]
RECOMMENDATION: Ship the narrowest wedge tomorrow, learn from
real usage. The full vision is a 3-month project — start with
the daily briefing that actually works.
[writes design doc → feeds into downstream skills automatically]
You: /plan-ceo-review
[reads the design doc, challenges scope, runs 10-section review]
You: /plan-eng-review
[ASCII diagrams for data flow, state machines, error paths]
[test matrix, failure modes, security concerns]
You: Approve plan. Exit plan mode.
[writes 2,400 lines across 11 files. ~8 minutes.]
You: /review
[AUTO-FIXED] 2 issues. [ASK] Race condition → you approve fix.
You: /qa https://staging.myapp.com
[opens real browser, clicks through flows, finds and fixes a bug]
You: /ship
Tests: 42 → 51 (+9 new). PR: github.com/you/app/pull/42
You said "daily briefing app." The agent said "you're building a chief of staff AI" — because it listened to your pain, not your feature request. Eight commands, end to end. That is not a copilot. That is a team.
The sprint
gstack is a process, not a collection of tools. The skills run in the order a sprint runs:
Think → Plan → Build → Review → Test → Ship → Reflect
Each skill feeds into the next. /office-hours writes a design doc that /plan-ceo-review reads. /plan-eng-review writes a test plan that /qa picks up. /review catches bugs that /ship verifies are fixed. Nothing falls through the cracks because every step knows what came before it.
| Skill | Your specialist | What they do |
|---|---|---|
/office-hours |
YC Office Hours | Start here. Six forcing questions that reframe your product before you write code. Pushes back on your framing, challenges premises, generates implementation alternatives. Design doc feeds into every downstream skill. |
/plan-ceo-review |
CEO / Founder | Rethink the problem. Find the 10-star product hiding inside the request. Four modes: Expansion, Selective Expansion, Hold Scope, Reduction. |
/plan-eng-review |
Eng Manager | Lock in architecture, data flow, diagrams, edge cases, and tests. Forces hidden assumptions into the open. |
/plan-design-review |
Senior Designer | Rates each design dimension 0-10, explains what a 10 looks like, then edits the plan to get there. AI Slop detection. Interactive — one AskUserQuestion per design choice. |
/plan-devex-review |
Developer Experience Lead | Interactive DX review: explores developer personas, benchmarks against competitors' TTHW, designs your magical moment, traces friction points step by step. Three modes: DX EXPANSION, DX POLISH, DX TRIAGE. 20-45 forcing questions. |
/design-consultation |
Design Partner | Build a complete design system from scratch. Researches the landscape, proposes creative risks, generates realistic product mockups. |
/review |
Staff Engineer | Find the bugs that pass CI but blow up in production. Auto-fixes the obvious ones. Flags completeness gaps. |
/investigate |
Debugger | Systematic root-cause debugging. Iron Law: no fixes without investigation. Traces data flow, tests hypotheses, stops after 3 failed fixes. |
/design-review |
Designer Who Codes | Same audit as /plan-design-review, then fixes what it finds. Atomic commits, before/after screenshots. |
/devex-review |
DX Tester | Live developer experience audit. Actually tests your onboarding: navigates docs, tries the getting started flow, times TTHW, screenshots errors. Compares against /plan-devex-review scores — the boomerang that shows if your plan matched reality. |
/design-shotgun |
Design Explorer | "Show me options." Generates 4-6 AI mockup variants, opens a comparison board in your browser, collects your feedback, and iterates. Taste memory learns what you like. Repeat until you love something, then hand it to /design-html. |
/design-html |
Design Engineer | Turn a mockup into production HTML that actually works. Pretext computed layout: text reflows, heights adjust, layouts are dynamic. 30KB, zero deps. Detects React/Svelte/Vue. Smart API routing per design type (landing page vs dashboard vs form). The output is shippable, not a demo. |
/qa |
QA Lead | Test your app, find bugs, fix them with atomic commits, re-verify. Auto-generates regression tests for every fix. |
/qa-only |
QA Reporter | Same methodology as /qa but report only. Pure bug report without code changes. |
/pair-agent |
Multi-Agent Coordinator | Share your browser with any AI agent. One command, one paste, connected. Works with OpenClaw, Hermes, Codex, Cursor, or anything that can curl. Each agent gets its own tab. Auto-launches headed mode so you watch everything. Auto-starts ngrok tunnel for remote agents. Scoped tokens, tab isolation, rate limiting, activity attribution. |
/cso |
Chief Security Officer | OWASP Top 10 + STRIDE threat model. Zero-noise: 17 false positive exclusions, 8/10+ confidence gate, independent finding verification. Each finding includes a concrete exploit scenario. |
/ship |
Release Engineer | Sync main, run tests, audit coverage, push, open PR. Bootstraps test frameworks if you don't have one. |
/land-and-deploy |
Release Engineer | Merge the PR, wait for CI and deploy, verify production health. One command from "approved" to "verified in production." |
/canary |
SRE | Post-deploy monitoring loop. Watches for console errors, performance regressions, and page failures. |
/benchmark |
Performance Engineer | Baseline page load times, Core Web Vitals, and resource sizes. Compare before/after on every PR. |
/document-release |
Technical Writer | Update all project docs to match what you just shipped. Catches stale READMEs automatically. Builds a Diataxis coverage map (reference / how-to / tutorial / explanation) so gaps are visible in the PR body. |
/document-generate |
Documentation Author | Generate missing docs from scratch using the Diataxis framework. Researches the codebase first, then writes reference / how-to / tutorial / explanation docs that actually match the code. Invokable standalone or chained from /document-release when the coverage map finds gaps. Learn more: tutorial • how-to • why Diataxis. |
/retro |
Eng Manager | Team-aware weekly retro. Per-person breakdowns, shipping streaks, test health trends, growth opportunities. /retro global runs across all your projects and AI tools (Claude Code, Codex, Gemini). |
/browse |
QA Engineer | Give the agent eyes. Real Chromium browser, real clicks, real screenshots. ~100ms per command. /open-gstack-browser launches GStack Browser with sidebar, anti-bot stealth, and auto model routing. |
/setup-browser-cookies |
Session Manager | Import cookies from your real browser (Chrome, Arc, Brave, Edge) into the headless session. Test authenticated pages. |
/autoplan |
Review Pipeline | One command, fully reviewed plan. Runs CEO → design → eng review automatically with encoded decision principles. Surfaces only taste decisions for your approval. |
/spec |
Spec Author | Turn vague intent into a precise, executable spec in five phases (why, scope, technical with mandatory code-reading, draft, file). Codex quality gate before file (blocks below 7/10), fail-closed secret redaction, dedupe against existing issues, archive to $GSTACK_STATE_ROOT/projects/$SLUG/specs/ for team-corpus recall. --execute spawns claude -p in a fresh worktree; /ship auto-closes the source issue on merge. Plan-mode aware. |
/learn |
Memory | Manage what gstack learned across sessions. Review, search, prune, and export project-specific patterns, pitfalls, and preferences. Learnings compound across sessions so gstack gets smarter on your codebase over time. |
/make-pdf |
Publisher | Markdown in, publication-quality document out. Mermaid and excalidraw fences render as vector diagrams, fully offline. Images scale to the page and never truncate; wide diagrams get their own landscape page. --to html emits one self-contained file, --to docx a Word doc. |
/diagram |
Diagram Maker | English in, editable diagram out. Emits a triplet: mermaid source, .excalidraw you can open and edit on excalidraw.com (hand-drawn style), and rendered SVG/PNG. Zero network. Embed the source in markdown and /make-pdf renders it. |
Which review should I use?
| Building for... | Plan stage (before code) | Live audit (after shipping) |
|---|---|---|
| End users (UI, web app, mobile) | /plan-design-review |
/design-review |
| Developers (API, CLI, SDK, docs) | /plan-devex-review |
/devex-review |
| Architecture (data flow, perf, tests) | /plan-eng-review |
/review |
| All of the above | /autoplan (runs CEO → design → eng → DX, auto-detects which apply) |
— |
Power tools
| Skill | What it does |
|---|---|
/codex |
Second Opinion — independent code review from OpenAI Codex CLI. Three modes: review (pass/fail gate), adversarial challenge, and open consultation. Cross-model analysis when both /review and /codex have run. |
/careful |
Safety Guardrails — warns before destructive commands (rm -rf, DROP TABLE, force-push). Say "be careful" to activate. Override any warning. |
/freeze |
Edit Lock — restrict file edits to one directory. Prevents accidental changes outside scope while debugging. |
/guard |
Full Safety — /careful + /freeze in one command. Maximum safety for prod work. |
/unfreeze |
Unlock — remove the /freeze boundary. |
/open-gstack-browser |
GStack Browser — launch GStack Browser with sidebar, anti-bot stealth, auto model routing (Sonnet for actions, Opus for analysis), one-click cookie import, and Claude Code integration. Clean up pages, take smart screenshots, edit CSS, and pass info back to your terminal. |
/setup-deploy |
Deploy Configurator — one-time setup for /land-and-deploy. Detects your platform, production URL, and deploy commands. |
/setup-gbrain |
GBrain Onboarding — from zero to running gbrain in under 5 minutes. PGLite local, Supabase existing URL, or auto-provision a new Supabase project via Management API. MCP registration for Claude Code + per-repo trust triad (read-write/read-only/deny). Full guide. |
/sync-gbrain |
Keep Brain Current — re-index this repo's code into gbrain via gbrain sources add + gbrain sync --strategy code, refresh the ## GBrain Search Guidance block in CLAUDE.md, and auto-remove guidance when the capability check fails. --incremental (default), --full, --dry-run. Idempotent; safe to re-run. |
/gstack-upgrade |
Self-Updater — upgrade gstack to latest. Detects global vs vendored install, syncs both, shows what changed. |
/ios-qa |
iOS Live-Device QA (v1.43.0.0+) — drive a real iPhone over USB CoreDevice via an embedded StateServer in the app. Read Swift source, codegen typed @Observable accessors, run the agent loop. Optional --tailnet flag exposes the device to OpenClaw or any HTTP-capable agent on your Tailscale tailnet so remote agents can run iOS QA without ever touching the hardware. Capability-tier allowlist (observe/interact/mutate/restore), per-device session lock, audit log. |
/ios-fix, /ios-design-review, /ios-clean, /ios-sync |
iOS bug-fix loop, designer's-eye HIG audit, debug-bridge cleanup, and accessor resync. See docs/skills.md. End-to-end walkthrough: docs/howto-ios-testing-with-gstack.md. |
Standalone binaries
Beyond the slash-command skills, gstack ships standalone CLIs for workflows that don't belong inside a session:
| Command | What it does |
|---|---|
gstack-model-benchmark |
Cross-model benchmark — run the same prompt through Claude, GPT (via Codex CLI), and Gemini; compare latency, tokens, cost, and (optionally) LLM-judge quality score. Auth detected per provider, unavailable providers skip cleanly. Output as table, JSON, or markdown. --dry-run validates flags + auth without spending API calls. |
gstack-taste-update |
Design taste learning — writes approvals and rejections from /design-shotgun into a persistent per-project taste profile. Decays 5%/week. Feeds back into future variant generation so the system learns what you actually pick. |
gstack-egress |
Egress receipt auditor — every gstack-initiated off-machine send writes a tamper-evident, hash-chained receipt to ~/.gstack/security/egress.jsonl before the send. list shows what gstack attempted to send and to which host, grants shows the standing consent settings plus the exact command that revokes each, verify recomputes the hash chain and exits 3 on tamper. |
gstack-context-bill |
Token bill-of-materials — read-only, offline audit of what an installed skills tree costs in tokens: always-on frontmatter every session pays vs per-invocation SKILL.md + forced references. --diff compares two trees, --budget enforces a ceiling, --exact opts into Anthropic count_tokens (sends file text off-machine; writes an egress receipt first, degrades to the offline estimate if the receipt can't be written). |
gstack-code-intelligence |
Code-intelligence provider picker — wraps GBrain, Sourcebot, and Graphify behind one interface: options/status to see what's available, select to pick one, index/search to use it, suggest to check whether the one-time indexing offer should fire here. The offer triggers on large repos (1,000+ tracked files; a decline is persisted). Non-local providers refuse to index or search until you record per-repo consent (consent <repo> yes|no — the query text is repo-derived content), the per-repo trust policy's deny and read-only tiers veto write-class operations regardless of consent, and every off-machine send writes an egress receipt. Fully optional — with nothing selected, gstack falls back to grep. |
gstack-verify-gate |
Verification stop hook (opt-in) — blocks a Claude Code turn from ending until the project's declared verify command passes (after 3 blocked re-entries it yields with a loud still-RED warning instead of looping forever). Declare it on one line in CLAUDE.md: <!-- gstack:verify: bun test -->. Hooks bypass the permission system, so a declared command never runs until you trust it once per repo (gstack-verify-gate --trust); editing the command invalidates trust until re-granted, and every grant is audit-logged. ./setup never registers it for you — opt in with gstack-settings-hook add-event --event Stop --command ~/.claude/skills/gstack/bin/gstack-verify-gate --source verify-gate, remove with gstack-settings-hook remove-source --source verify-gate. |
gstack-ios-qa-daemon |
iOS QA daemon — Mac-side broker between an agent and a connected iPhone over USB CoreDevice. Loopback by default; --tailnet opens a Tailscale-facing listener with identity-gated capability tiers. Single-instance via flock on ~/.gstack/ios-qa-daemon.pid. See docs/howto-ios-testing-with-gstack.md. |
gstack-ios-qa-mint |
iOS allowlist manager — owner-grant CLI for the tailnet allowlist. grant/revoke/list against ~/.gstack/ios-qa-allowlist.json (mode 0600). Remote agents never auto-allowlist; this is the explicit-intent path. |
gstack-ios-qa-regen |
iOS bridge regenerator — deterministically installs the canonical DebugBridge package, generates typed state accessors, and records the installed gstack version. Safe to rerun after source changes or upgrades. |
Continuous checkpoint mode (opt-in, local by default)
Set gstack-config set checkpoint_mode continuous and skills auto-commit your work as you go with a WIP: prefix plus a structured [gstack-context] body (decisions, remaining work, failed approaches). Survives crashes and context switches. /context-restore reads those commits to reconstruct session state. /ship filter-squashes WIP commits before the PR (preserving non-WIP commits) so bisect stays clean. Push is opt-in via checkpoint_push=true — default is local-only so you don't trigger CI on every WIP commit.
Domain skills + raw CDP escape hatch
Two new browser primitives compound the gstack agent over time:
$B domain-skill save— agent saves a per-site note (e.g., "LinkedIn's Apply button lives in an iframe") that fires automatically next time it visits that hostname. Quarantined → active after 3 successful uses → optional cross-project promotion via$B domain-skill promote-to-global. Storage lives alongside/learn's per-project learnings file. Full reference: docs/domain-skills.md.$B cdp <Domain.method>— raw Chrome DevTools Protocol escape hatch for the rare case curated commands miss. Deny-default: methods must be explicitly added tobrowse/src/cdp-allowlist.tswith a one-line justification. Two-tier mutex serializes browser-scoped CDP calls against per-tab work. Output for data-exfil methods is wrapped in the UNTRUSTED envelope.
Want raw CDP with no rails, no allowlist, no daemon — just thin transport from agent to Chrome? browser-use/browser-harness-js is a different philosophy (agent-authored helpers vs gstack's curated commands) and a good fit if you don't want gstack's security stack. The two can coexist: gstack's
$B cdpand harness can both attach to the same Chrome via Playwright'snewCDPSession.
Deep dives with examples and philosophy for every skill →
Karpathy's four failure modes? Already covered.
Andrej Karpathy's AI coding rules (17K stars) nail four failure modes: wrong assumptions, overcomplexity, orthogonal edits, imperative over declarative. gstack's workflow skills enforce all four. /office-hours forces assumptions into the open before code is written. The Confusion Protocol stops Claude from guessing on architectural decisions. /review catches unnecessary complexity and drive-by edits. /ship transforms tasks into verifiable goals with test-first execution. If you already use Karpathy-style CLAUDE.md rules, gstack is the workflow enforcement layer that makes them stick across entire sprints, not just single prompts.
Parallel sprints
gstack works well with one sprint. It gets interesting with ten running at once.
Design is at the heart. /design-consultation builds your design system from scratch, researches what's out there, proposes creative risks, and writes DESIGN.md. But the real magic is the shotgun-to-HTML pipeline.
/design-shotgun is how you explore. You describe what you want. It generates 4-6 AI mockup variants using GPT Image. Then it opens a comparison board in your browser with all variants side by side. You pick favorites, leave feedback ("more whitespace", "bolder headline", "lose the gradient"), and it generates a new round. Repeat until you love something. Taste memory kicks in after a few rounds so it starts biasing toward what you actually like. No more describing your vision in words and hoping the AI gets it. You see options, pick the good ones, and iterate visually.
/design-html makes it real. Take that approved mockup (from /design-shotgun, a CEO plan, a design review, or just a description) and turn it into production-quality HTML/CSS. Not the kind of AI HTML that looks fine at one viewport width and breaks everywhere else. This uses Pretext for computed text layout: text actually reflows on resize, heights adjust to content, layouts are dynamic. 30KB overhead, zero dependencies. It detects your framework (React, Svelte, Vue) and outputs the right format. Smart API routing picks different Pretext patterns depending on whether it's a landing page, dashboard, form, or card layout. The output is something you'd actually ship, not a demo.
/qa was a massive unlock. It let me go from 6 to 12 parallel workers. Claude Code saying "I SEE THE ISSUE" and then actually fixing it, generating a regression test, and verifying the fix — that changed how I work. The agent has eyes now.
Smart review routing. Just like at a well-run startup: CEO doesn't have to look at infra bug fixes, design review isn't needed for backend changes. gstack tracks what reviews are run, figures out what's appropriate, and just does the smart thing. The Review Readiness Dashboard tells you where you stand before you ship.
Test everything. /ship bootstraps test frameworks from scratch if your project doesn't have one. Every /ship run produces a coverage audit. Every /qa bug fix generates a regression test. 100% test coverage is the goal — tests make vibe coding safe instead of yolo coding.
/document-release is the engineer you never had. It reads every doc file in your project, cross-references the diff, and updates everything that drifted. README, ARCHITECTURE, CONTRIBUTING, CLAUDE.md, TODOS — all kept current automatically. And now /ship auto-invokes it — docs stay current without an extra command.
Real browser mode. /open-gstack-browser launches GStack Browser, an AI-controlled Chromium with anti-bot stealth, custom branding, and the sidebar extension baked in. Sites like Google and NYTimes work without captchas. The menu bar says "GStack Browser" instead of "Chrome for Testing." Your regular Chrome stays untouched. All existing browse commands work unchanged. $B disconnect returns to headless. The browser stays alive as long as the window is open... no idle timeout killing it while you're working.
Sidebar agent — your AI browser assistant. Type natural language in the Chrome side panel and a child Claude instance executes it. "Navigate to the settings page and screenshot it." "Fill out this form with test data." "Go through every item in this list and extract the prices." The sidebar auto-routes to the right model: Sonnet for fast actions (click, navigate, screenshot) and Opus for reading and analysis. Each task gets up to 5 minutes. The sidebar agent runs in an isolated session, so it won't interfere with your main Claude Code window. One-click cookie import right from the sidebar footer.
Personal automation. The sidebar agent isn't just for dev workflows. Example: "Browse my kid's school parent portal and add all the other parents' names, phone numbers, and photos to my Google Contacts." Two ways to get authenticated: (1) log in once in the headed browser, your session persists, or (2) click the "cookies" button in the sidebar footer to import cookies from your real Chrome. Once authenticated, Claude navigates the directory, extracts the data, and creates the contacts.
Prompt injection defense. Hostile web pages try to hijack your sidebar agent. gstack ships a layered defense: a 22MB ML classifier bundled with the browser scans every page and tool output locally, a Claude Haiku transcript check votes on the full conversation shape, a random canary token in the system prompt catches session exfil attempts across text, tool args, URLs, and file writes, and a verdict combiner requires two classifiers to agree before blocking (prevents single-model false positives on Stack Overflow-style instruction pages). A shield icon in the sidebar header shows status (green/amber/red). Opt in to a 721MB DeBERTa-v3 ensemble via GSTACK_SECURITY_ENSEMBLE=deberta for 2-of-3 agreement. Emergency kill switch: GSTACK_SECURITY_OFF=1. See ARCHITECTURE.md for the full stack.
Browser handoff when the AI gets stuck. Hit a CAPTCHA, auth wall, or MFA prompt? $B handoff opens a visible Chrome at the exact same page with all your cookies and tabs intact. Solve the problem, tell Claude you're done, $B resume picks up right where it left off. The agent even suggests it automatically after 3 consecutive failures.
/pair-agent is cross-agent coordination. You're in Claude Code. You also have OpenClaw running. Or Hermes. Or Codex. You want them both looking at the same website. Type /pair-agent, pick your agent, and a GStack Browser window opens so you can watch. The skill prints a block of instructions. Paste that block into the other agent's chat. It exchanges a one-time setup key for a session token, creates its own tab, and starts browsing. You see both agents working in the same browser, each in their own tab, neither able to interfere with the other. If ngrok is installed, the tunnel starts automatically so the other agent can be on a completely different machine. Same-machine agents get a zero-friction shortcut that writes credentials directly. This is the first time AI agents from different vendors can coordinate through a shared browser with real security: scoped tokens, tab isolation, rate limiting, domain restrictions, and activity attribution.
Multi-AI second opinion. /codex gets an independent review from OpenAI's Codex CLI — a completely different AI looking at the same diff. Three modes: code review with a pass/fail gate, adversarial challenge that actively tries to break your code, and open consultation with session continuity. When both /review (Claude) and /codex (OpenAI) have reviewed the same branch, you get a cross-model analysis showing which findings overlap and which are unique to each.
Safety guardrails on demand. Say "be careful" and /careful warns before any destructive command — rm -rf, DROP TABLE, force-push, git reset --hard. /freeze locks edits to one directory while debugging so Claude can't accidentally "fix" unrelated code. /guard activates both. /investigate auto-freezes to the module being investigated.
Proactive skill suggestions. gstack notices what stage you're in — brainstorming, reviewing, debugging, testing — and suggests the right skill. Don't like it? Say "stop suggesting" and it remembers across sessions.
10-15 parallel sprints
gstack is powerful with one sprint. It is transformative with ten running at once.
Conductor runs multiple Claude Code sessions in parallel — each in its own isolated workspace. One session running /office-hours on a new idea, another doing /review on a PR, a third implementing a feature, a fourth running /qa on staging, and six more on other branches. All at the same time. I regularly run 10-15 parallel sprints — that's the practical max right now.
The sprint structure is what makes parallelism work. Without a process, ten agents is ten sources of chaos. With a process — think, plan, build, review, test, ship — each agent knows exactly what to do and when to stop. You manage them the way a CEO manages a team: check in on the decisions that matter, let the rest run.
Voice input (AquaVoice, Whisper, etc.)
gstack skills have voice-friendly trigger phrases. Say what you want naturally — "run a security check", "test the website", "do an engineering review" — and the right skill activates. You don't need to remember slash command names or acronyms.
Uninstall
Option 1: Run the uninstall script
If gstack is installed on your machine:
~/.claude/skills/gstack/bin/gstack-uninstall
This handles skills, symlinks, global state (~/.gstack/), project-local state, browse daemons, and temp files. Use --keep-state to preserve config and analytics. Use --force to skip confirmation.
Option 2: Manual removal (no local repo)
If you don't have the repo cloned (e.g. you installed via a Claude Code paste and later deleted the clone):
# 1. Stop browse daemons
pkill -f "gstack.*browse" 2>/dev/null || true
# 2. Remove per-skill directories whose SKILL.md points into gstack/
find ~/.claude/skills -mindepth 1 -maxdepth 1 -type d ! -name gstack 2>/dev/null |
while IFS= read -r dir; do
link="$dir/SKILL.md"
[ -L "$link" ] || continue
target=$(readlink "$link" 2>/dev/null) || continue
case "$target" in
gstack/*|*/gstack/*)
rm -f "$link"
rmdir "$dir" 2>/dev/null || true
;;
esac
done
# 3. Remove gstack
rm -rf ~/.claude/skills/gstack
# 4. Remove global state
rm -rf ~/.gstack
# 5. Remove integrations (skip any you never installed)
rm -rf ~/.codex/skills/gstack* 2>/dev/null
rm -rf ~/.factory/skills/gstack* 2>/dev/null
rm -rf ~/.kiro/skills/gstack* 2>/dev/null
rm -rf ~/.openclaw/skills/gstack* 2>/dev/null
# 6. Remove temp files
rm -f /tmp/gstack-* 2>/dev/null
# 7. Per-project cleanup (run from each project root)
rm -rf .gstack .gstack-worktrees .claude/skills/gstack 2>/dev/null
rm -rf .agents/skills/gstack* .factory/skills/gstack* 2>/dev/null
Clean up CLAUDE.md
The uninstall script does not edit CLAUDE.md. In each project where gstack was added, remove the ## gstack and ## Skill routing sections.
Playwright
~/Library/Caches/ms-playwright/ (macOS) is left in place because other tools may share it. Remove it if nothing else needs it.
Free, MIT licensed, open source. No premium tier, no waitlist.
I open sourced how I build software. You can fork it and make it your own.
We're hiring. Want to ship real products at AI-coding speed and help harden gstack? Come work at YC — ycombinator.com/software Extremely competitive salary and equity. San Francisco, Dogpatch District.
GBrain — persistent knowledge for your coding agent
GBrain is a persistent knowledge base for AI agents — think of it as the memory your agent actually keeps between sessions. GStack gives you a one-command path from zero to "it's running, my agent can call it."
/setup-gbrain
Four paths, pick one:
- Supabase, existing URL — your cloud agent already provisioned a brain; paste the Session Pooler URL, now this laptop uses the same data.
- Supabase, auto-provision — paste a Supabase Personal Access Token; the skill creates a new project, polls to healthy, fetches the pooler URL, hands it to
gbrain init. ~90 seconds end-to-end. - PGLite local — zero accounts, zero network, ~30 seconds. Isolated brain on this Mac only. Great for try-first; migrate to Supabase later with
/setup-gbrain --switch. - Remote gbrain MCP — your brain runs on another machine (Tailscale, ngrok, internal LAN) or a teammate's server; paste an MCP URL and bearer token. Optionally pair with a local PGLite for symbol-aware code search in split-engine mode. Best for cross-machine memory without standing up a local DB.
After init, the skill offers to register gbrain as an MCP server for Claude Code (claude mcp add gbrain -- gbrain serve) so gbrain search, gbrain put, etc. show up as first-class typed tools — not bash shell-outs.
Keeping the brain current. Run /sync-gbrain from any repo to re-index its code into gbrain (incremental by default, --full for a full reindex, --dry-run to preview). The skill registers the cwd as a federated source via gbrain sources add, runs gbrain sync --strategy code, and writes a ## GBrain Search Guidance block to your project's CLAUDE.md so the agent prefers gbrain search/code-def/code-refs over Grep. The block is removed automatically if the capability check fails — no stale guidance pointing at tools that aren't installed.
Per-remote trust policy. Each repo on your machine gets one of three tiers:
read-write— agent can search the brain AND write new pages back from this reporead-only— agent can search but never writes (best for multi-client consultants: search the shared brain, don't contaminate it with Client A's work while in Client B's repo)deny— no gbrain interaction at all
The skill asks once per repo. The decision is sticky across worktrees and branches of the same remote.
GStack memory sync (different feature, same private-repo infra). Optionally pushes your gstack state (learnings, CEO plans, design docs, retros, developer profile) to a private git repo so your memory follows you across machines, with a one-time privacy prompt (everything allowlisted / artifacts only / off) and a defense-in-depth secret scanner that blocks AWS keys, tokens, PEM blocks, and JWTs before they leave your machine.
gstack-artifacts-init
Running gstack in Conductor? Conductor explicitly strips ANTHROPIC_API_KEY and OPENAI_API_KEY from every workspace's process env, so paid evals and gbrain embeddings won't work out of the box. Set GSTACK_ANTHROPIC_API_KEY and GSTACK_OPENAI_API_KEY in Conductor's workspace env config instead — gstack's TS entry points promote them to canonical names at runtime. Full details and the contributor checklist for adding the import to new entry points: Conductor + GSTACK_* env vars.
Full monty — every scenario, every flag, every bin helper, every troubleshooting step: USING_GBRAIN_WITH_GSTACK.md
Other references: docs/gbrain-sync.md (sync-specific guide) • docs/gbrain-sync-errors.md (error index)
Docs
| Doc | What it covers |
|---|---|
| Skill Deep Dives | Philosophy, examples, and workflow for every skill (includes Greptile integration) |
| Diagrams & Document Formats | Mermaid/excalidraw fences in PDFs, image sizing and safety defaults, --to html|docx, /diagram triplets |
| Builder Ethos | Builder philosophy: Boil the Ocean, Search Before Building, three layers of knowledge |
| Using GBrain with GStack | Every path, flag, bin helper, and troubleshooting step for /setup-gbrain |
| GBrain Sync | Cross-machine memory setup, privacy modes, troubleshooting |
| Architecture | Design decisions and system internals |
| Browser Reference | Full command reference for /browse |
| Contributing | Dev setup, testing, contributor mode, and dev mode |
| Changelog | What's new in every version |
Privacy & Telemetry
gstack includes opt-in usage telemetry to help improve the project. Here's exactly what happens:
- Default is off. Nothing is sent anywhere unless you explicitly say yes.
- On first run, gstack asks if you want to share anonymous usage data. You can say no.
- What's sent (if you opt in): skill name, duration, success/fail, gstack version, OS. That's it.
- What's never sent: code, file paths, repo names, branch names, prompts, or any user-generated content.
- Change anytime:
gstack-config set telemetry offdisables everything instantly. - Every off-machine send is receipted. Any gstack-initiated network send — telemetry included — writes a hash-chained, tamper-evident receipt to
~/.gstack/security/egress.jsonlbefore the send; sensitive sinks refuse to send at all if the receipt can't be written. Audit withgstack-egress list, verify the chain withgstack-egress verify(exit 3 on tamper), see the standing consent settings withgstack-egress grants. The ledger records attempted sends so accidents are auditable — it's an audit trail, not a network firewall.
Data is stored in Supabase (open source Firebase alternative). The schema is in supabase/migrations/ — you can verify exactly what's collected. The Supabase publishable key in the repo is a public key (like a Firebase API key) — row-level security policies deny all direct access. Telemetry flows through validated edge functions that enforce schema checks, event type allowlists, and field length limits.
Local analytics are always available. Run gstack-analytics to see your personal usage dashboard from the local JSONL file — no remote data needed.
Troubleshooting
Skill not showing up? cd ~/.claude/skills/gstack && ./setup
/browse fails? cd ~/.claude/skills/gstack && bun install && bun run build
Stale install? Run /gstack-upgrade — or set auto_upgrade: true in ~/.gstack/config.yaml
Want shorter commands? cd ~/.claude/skills/gstack && ./setup --no-prefix — switches from /gstack-qa to /qa. Your choice is remembered for future upgrades.
Want namespaced commands? cd ~/.claude/skills/gstack && ./setup --prefix — switches from /qa to /gstack-qa. Useful if you run other skill packs alongside gstack.
Codex says "Skipped loading skill(s) due to invalid SKILL.md"? Your Codex skill descriptions are stale. Fix: cd ~/.codex/skills/gstack && git pull && ./setup --host codex — or for repo-local installs: cd "$(readlink -f .agents/skills/gstack)" && git pull && ./setup --host codex
Windows users: gstack works on Windows 11 via Git Bash or WSL. Node.js is required in addition to Bun — Bun has a known bug with Playwright's pipe transport on Windows (bun#4253). The browse server automatically falls back to Node.js. Make sure both bun and node are on your PATH.
On Windows without Developer Mode (MSYS2 / Git Bash), setup falls back to file copies instead of symlinks because ln -snf produces frozen copies that don't refresh on git pull. Re-run cd ~/.claude/skills/gstack && ./setup after every git pull so your skill files match the repo. setup prints a one-line note reminding you. Unix and WSL keep symlinks and don't need the re-run.
Claude says it can't see the skills? Make sure your project's CLAUDE.md has a gstack section. Add this:
## gstack
Use /browse from gstack for all web browsing. Never use mcp__claude-in-chrome__* tools.
Available skills: /office-hours, /plan-ceo-review, /plan-eng-review, /plan-design-review,
/design-consultation, /design-shotgun, /design-html, /review, /ship, /land-and-deploy,
/canary, /benchmark, /browse, /open-gstack-browser, /qa, /qa-only, /design-review,
/setup-browser-cookies, /setup-deploy, /setup-gbrain, /sync-gbrain, /retro, /investigate,
/document-release, /document-generate, /codex, /cso, /autoplan, /pair-agent, /careful, /freeze,
/guard, /unfreeze, /gstack-upgrade, /learn.
License
MIT. Free forever. Go build something.