mirror of
https://github.com/garrytan/gstack.git
synced 2026-10-31 22:14:25 +01:00
4d69ff02dfca4e2dc352dde43745dcc53fe0a83c
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Garry Tan
|
55f8151fbe |
v1.91.30.0 fix: honest eval failure reports, Bun floor with safe auto-update, make-pdf and /review on Codex (#3033)
* docs: approved reliability follow-ups plan (Oct 2026)
* D1: score overlay wrong answers as comparison results (contract v4)
A completed answer that fails the fixture's output check is now a valid
measurement with taskCorrect false and answerError set, in either arm.
Only execution, scope and metric failures invalidate a trial. ON
correctness still gates the case, so an OFF wrong answer stays in the
comparison. OVERLAY_CONTRACT moves to v4; v3 verdicts are not
reinterpreted. Replays the 37198445662 sonnet trials (OFF-0 wrong).
* C4: score auq-matrix substance with a 3-sample judge panel
Each skill keeps one native capture in the same execution; the
recommendation-substance grader now runs as a JUDGE_PANEL_SAMPLES panel on
that capture and gates on the mean against the unchanged minimum of 4. A
failed sample fails the panel without resampling. Approved at the gate
2026-10-04 for censuses after this change.
* E1: one Bun version source; setup refuses below the 1.3.3 floor, updater checks before advancing
bin/gstack-bun-version.sh holds the security floor (1.3.3) and the tested
version (1.4.0, the CI pin and engines.bun). setup refuses Bun below the
floor before writing anything, warns between floor and tested, warns and
continues on an unparseable version, and prints a 1.4.0 install hint.
Prereleases sort below their release (semver).
gstack-session-update now fetches, reads the incoming release's floor and
checks the bun setup will run before fast-forwarding to that exact
revision; below it the update is held with reason bun-too-old and a
resume command, leaving the revision and installs untouched.
Fixture copies of setup (team-mode, claude-code migration) now include the
sourced file; the slow-git lock test sleeps on fetch, the updater's
network step.
* E3: record the owned Chromium's PID over CDP so the #2709 reap runs on Playwright 1.62
Playwright 1.62's Browser has no process(), so chromiumProcInfo was always
null and browse stop never reaped a surviving headless Chromium. The owned
headless launch now reads the browser PID from CDP SystemInfo.getProcessInfo
(bounded at 2s; a hung or failed call logs once and records nothing). The
reaper's start-time and cmdline identity checks are unchanged, so a recycled
PID or an externally owned browser is never signalled.
* A1: ship docs gate records post-child hashes with gstack-docs-candidate, saves each record once
Stored transcripts (37198445662 gate-census-2) show the parent's reads of
release files take ~1 s; the parent's own model time dominates (completion
268 s of 392 s, store ~270 s of 387 s), and 45-78 s of it is retyping hash
maps, child output and records into several artifacts after the child
returns. The child contract is unchanged.
* A2: /plan-ceo-review sends the mode handoff line before any other tool call
plan-ceo-mode-routing's HOLD SCOPE session spends its 240 s post-selection
window on model work: 37176837432 skipped the handoff helper and paraphrased
the mode, 37198445662 ran it, then wrote the plan, loaded sections and probed
Codex for 114 s before the posture chat. The handoff line is the posture.
* B3: accept wrapped unchanged rows in the cropped Edit preview
PR run 37176835584 timed out with the floor's owned Edit menu on screen:
unchanged diff rows soft-wrap under their gutter with a blank marker
column, and a trimmed empty unchanged row keeps only its number. The
parser accepted wrapped +/- rows only, so it never verified the owned
edit. Both forms are now parsed and still compared with the exact source
lines; the captured viewport is replayed with negative controls.
* A3: record per-phase headroom diagnosis for docsync, plan-mode-no-op, mode-routing and gate slice 11
* E4: /review Step 1 keeps going when the base fetch fails and reports stale base coverage
Under Codex's read-only .git, 'git fetch && merge-base && diff --stat'
stopped at the fetch, so the empty output read as 'Nothing to review'.
The fetch is now separate, prints BASE_REFRESH fresh or stale with the
local base revision, and the review states stale coverage.
* D2: reword the format-approach question prompt
Both plan-ceo-review-format-approach refusals in census 37195203538
(Fable 5.1 model_refusal_no_fallback, category reasoning_extraction:
"duplicating model outputs") were anchored on the session prompt, whose
capture instruction asked for the "verbatim" question and "the exact
prose you would have shown" in a "format-capture test". The approach
case now asks for the same question in the same file as the step's
output, without that wording. The other three plan-format cases keep
their unchanged prompt (no refusals in 72 trials).
* C2: /qa defines its outcome filename fields; QA report templates name the measured timing fields
Low judge samples in failing panels named both: qa workflow (37186854666)
'an undefined {user} in the outcome filename', and qa-only workflow
(37198445662) 'template asks for Duration / durationMs totals while
reporting.md mandates Probe budget and Guarded command time'. The qa
union stays under its 1.11 size ratio (93292 of 93294 bytes).
* 0.1: session ledger written by the claude -p, Agent SDK, PTY and Codex runners
Each armed model session appends one best-effort line to
$GSTACK_EVAL_DIR/session-ledger.jsonl: a stable session key, the timeout it
armed, monotonic elapsed time, its end reason (completed, session_timeout,
observer_timeout, api_error, refusal, aborted, error), one sanitized evidence
line and, for stream-JSON runners, a compact liveness summary (longest
in-request silence with nothing outstanding, last event kind, tool open at
end). Partial messages now stream for every claude -p session and for SDK
sessions without a custom query provider (runSharedInteractive opts in); their
rows feed only the summary, so transcripts keep their size unless
publicStreamDiagnostics is set. Structured refusal and API-error signals win
over phrases, and tool results are never read for them.
* 0.2: failure_cause beside failure_class, from runner facts and the session ledger
One pure failureCauseOf() beside failureClassOf() sets failure_cause
(contract > pre_turn_infra > api_error > refusal > provider_stall >
session_timeout > observer_timeout > assertion > unknown; no machine-set
'detector') with one bounded evidence line, and failureDetailOf() keeps a
Bun matcher's Expected/Received or a judge's failing dimensions in
failure_detail. Both record builders (classifyTrialShard, junitCensus) call
them and attach the trial's ledger sessions. failure_class, error and
TRIAL_ERROR_MAX are computed exactly as before; the new fields are optional,
bounded and validated only when present, and a frozen copy of the v1 reader
accepts every new record. A provider stall needs partial messages and
STALL_WINDOW_MS (120 s, eval-budgets.ts beside HEADROOM_WARN/FAIL) of
in-request silence with nothing outstanding; it stays a failure under
EVAL_POLICY v1.
* E4: fingerprint under a read-only .git via a private object dir owned by review-log
Codex's sandbox mounts .git read-only, so gstack-wtree's git add -A could not
write blobs and /review lost its start capture. gstack-review-log now creates
a private 0700 object directory under TMPDIR, passes it to gstack-wtree (which
writes new objects there with the common-dir object store, resolved before
exporting, as an alternate, so linked worktrees work), exposes it to the
bindReview/checkSharedLibsReuse consumers that read the tree and blobs, and
removes it on EXIT/INT/TERM after they finish. gstack-wtree prints one stderr
line and exits 1 on a failed write or unwritable TMPDIR instead of printing a
partial hash; review-log no longer swallows that line. Standalone gstack-wtree
keeps writing to the repository store so git diff between saved snapshots
(/ship) still works.
* 0.3: red lines name their values, evidence and an after-a-repair command; one published-text sanitizer
Each red line in the census report keeps its stored error and adds the
machine cause with its evidence, the failure_detail values (Expected/Received,
or each failing judge dimension with its mean, sample count and the sample
sentence naming it), the slice artifact and shard directory that hold the
transcript, a fetch command (eval:pass-rates --run <id> --case <id>), and
'after a repair:' with a command that selects only that case (E2E ids via
--case, judges via their selection and an exact test-name filter). A
session that completed and failed a check says so instead of 'assertion
(success)'. Report text that reaches a PR comment, issue or committed file
goes through scripts/lib/published-text.ts: control and bidi characters
stripped, credential-scan patterns redacted, mentions neutralized, HTML and
link brackets escaped inline, blocks in a fence longer than any backtick run,
length capped. Verdicts and collector verdict objects are unchanged.
* E2: one make-pdf resolver for readiness and every $P block on env-var hosts
make-pdf exported $P from its readiness block, which Codex and the other
env-var hosts lose between blocks. binaryAssignment gains 'make-pdf': the
MAKE-PDF SETUP readiness check and the runtime prelude both use it, so every
block that runs $P derives the same make-pdf/dist/pdf binary, with
MAKE_PDF_BIN (dev builds) winning when it is executable. Claude output keeps
its repo-local-first probe; only the false 'subsequent blocks' comment
changed. Root plus P=... fits the 400-byte prelude budget; the fixed
MAKE_PDF_OVERRIDE segment is the one exemption, named in runtime-root.ts and
the conformance test.
* 0.1: eval:pass-rates --headroom, --reds and --run report views; 85% headroom alarm in --gate
A new pure module, scripts/lib/eval-history.ts, reports per case the slowest
armed session against the timeout it armed (censored when it timed out,
insufficient below 3 samples, unknown without a session ledger, with the
case wall shown separately as an upper bound), the census critical path from
the slice jobs' timings, the verdict red ledger per census by failure class
and machine cause with shrunk per-case rates, and the all-green probability
(an approximation: the product of 1 - p_i, with an interval). --run <id>
lists one census's reds with their values and history and downloads only the
slice artifacts those reds name, reporting any over 64 MB instead of skipping
them. eval-flake-rank.ts only routes flags: --help works offline, an unknown
flag or case id exits 2 before any fetch, --case narrows the JSON and the
alarms, every view prints its run/branch scope, and --gate also alarms on a
session above HEADROOM_FAIL (85%) of its budget.
* E5: GSTACK_DESIGN_IMAGE_MODEL override; the live model check always tests the defaults
The image_generation tool model gains an env override validated against a
conservative gpt-image-<version> pattern; a bad value fails before any
request and the error names the variable. Pairing and rejection hints name
the tool model actually requested. design/scripts/live-model-check.ts now
ignores GSTACK_DESIGN_MODEL and GSTACK_DESIGN_IMAGE_MODEL (and says so), so
an override cannot mask a retirement of a default, and its exit 2 (no key)
says nothing was checked.
* C3: /deslop-shared-libs names a concrete risk for each recommendation, never 'none'
Both shared-libs-opportunity-judgment reds (37193478719 t3, 37195203538 t1)
failed adoption_and_risk because the top recommendation said 'Main risk:
none found inside the repo' / 'none in behavior'. The skill asked to 'name
the main risk or uncertainty' without saying what counts.
* B4: --case selects loop-registered cases; an unknown id exits before any process
caseSelection() resolves the one paid file that registers a case and how a
trial selects it: by its literal Bun test name, by the id a registration loop
names its test at runtime (coverage audits, TPA, journeys), or by running the
whole file when it registers no other case (plan-mode-no-op, whose test names
are not the id, so the old name pattern ran nothing). A whole-file case runs
one file shard per trial and is judged by shard status, as CI judges it.
--list prints the selection. Every E2E_TIERS id now resolves except
carve-section-loading, which fifteen files register (CI runs each as its own
shard) and which still fails with the owning files named.
* 0.3: after-a-repair runs the shard file when --case cannot resolve a case
A case that several files register (carve-section-loading) cannot be selected
by --case; its red line now prints the bun test command for the shard file
that went red, so every printed command selects exactly one case.
* C1: pre-register workflow-judge control bundles for qa, qa-only, review and setup-deploy
Positive (production bundle), negative (headings only) and one targeted
execution-blocking defect per judge, one 3-sample panel each, built with the
eval's own reader, prompt builder, options and floors.
* F: cost_known: a trial whose harness captured no billing says so; totals never hide it
cost_usd stays a finite number. Trial and eval records gain an optional
cost_known boolean (absent means known, so old records still parse; the
validator accepts only a boolean). Session-ledger rows record whether the
session's terminal result carried total_cost_usd; PTY and Codex sessions
never do, and SDK captures without a terminal result do not. A trial is
cost-known only when it has eval records and every record and session
captured billing, so JUnit-only PTY cases (125 of 298 trials in run
37198445662 recorded 0) now read cost unknown. The census headline prints the
known sum plus the unknown count, never a bare total over unknown costs, and
no total when every cost is unknown; the history import keeps the flag.
PTY transcripts are not priced: test/helpers/pricing.ts estimates exclude
cache writes, so they cannot give the billed cost.
* B1: replay corpora for the five repeat-offender cases
shared-libs-plan-callers, plan-ceo-section-loading, auto-decide-preserved,
plan-design-review-plan-mode and ship-docsync-late-result each get an
inventory of every wave census in scope (all artifacts still available)
and minimal, redacted detector or actor inputs replayed through the real
detector in the free suite: census passes, every distinct census red, and
derived negative controls labelled as derived. Fixed false reds stay as
expected passes with their history; the two genuine auto-decide product
failures (37176837432, 37179171083) stay refused. A contract test bounds
each entry at 64 KB, scans it for HIGH credentials, requires a pass and a
refusal per case, and reports entries whose series identity is stale.
Building the auto-decide corpus exposed a latent native detector misread:
a bold 'Mode application and rationale.' heading was parsed as a
withdrawn Mode field (37186854666). A field status now ends at its
sentence.
* B2: read approved stale-fill decisions from more valid ledger records
The structured ledger reader is authoritative before the prose fallback,
but it missed two valid approved records in passing censuses
(37186854666: a camelCase writeProfile subject; 37195203538: a
slash-joined 'D1 / STALE-FILL' row id), which then rested on the prose
detector that caused both historical reds. Both forms are accepted; an
unresolved row, an unrelated decision, the copied fixture plan and
non-write camelCase words stay refused.
* C1: record workflow-judge control results (negatives separated; qa and review miss a removed step)
* C2: record every low workflow-judge sample in the wave censuses with its disposition
* 0.3: published lines cannot break a fixed fence
evals.yml wraps collector-outcomes failures in a fixed three-backtick fence
for the PR comment, so those lines (and eval:pass-rates output meant for the
tracking issue) go through sanitizeFixedFenceLine: cleaned like every
published line, with backticks turned into quotes. Inline published text
also neutralizes backticks.
* A1: fit the post-child record wording under ship's 1.428 parity size cap
* E4: inline the Step 1 base refresh to fit review's 1.199 parity size cap
* B2: inventory of gate and periodic phrase-only detector checks
docs/evals/detector-inventory.md lists every pass/fail decision in gate
and periodic E2E cases that rests on phrases in model free text, with
its structured source (if any), whether the contract is user-visible,
and a disposition: already structured-first, convert candidate, phrase
witness required, or no structured source. Convert candidates are
ordered by census red history.
* 0.2: failureCauseOf takes the stall window instead of importing it; the PTY ledger never throws
eval-store.ts imported STALL_WINDOW_MS from eval-budgets.ts, which free tests
replace with mock.module stubs (review-finalization-budget, office-hours,
qa-fix-loop, qa-only), so loading eval-store failed there. failureCauseOf is
now dependency-free: the record builders pass STALL_WINDOW_MS. A PTY session
whose handle cannot show its text (test doubles) records no panel instead of
throwing from runPtySession's cleanup. The 37186854666 replay also runs
through classifyTrialShard.
* A2/auto-decide: the CEO handoff chat repeats the helper's printed line verbatim
Claude Code 2.1.284 collapses the helper's Bash output ('Ran 1 shell
command'), so the 'Auto-decided review mode → …' line reaches the user only
through chat. In lane D's auto-decide-preserved corpus every passing capture's
chat starts with that line, and the reds paraphrased it ('HOLD SCOPE was
auto-selected…', 'Mode decision done…', "I'm defaulting to HOLD SCOPE…").
The template now says verbatim and why.
* C2: record the judge panels rerun on the final ship and review wording
* integrate: touchfiles and module-size ratchet entries for the wave's new modules
GLOBAL_TOUCHFILES gains session-ledger.ts, published-text.ts and
auq-substance-panel.ts; bin/gstack-bun-version.sh joins every entry that
lists setup or the updater; auq-matrix stays a rule case with its C4
evidence beside E2E_KINDS; eval-history, published-text and session-ledger
are registered new modules.
* E1: BROWSE SETUP install hint pins the tested Bun (1.4.0)
* E2: quote "$P" in make-pdf commands so install paths with spaces work
* E1: /gstack-upgrade checks the incoming Bun floor before pulling
gstack_bun_incoming_hold (bin/gstack-bun-version.sh) is now the one
pre-advance check: the auto-updater calls it before its fast-forward and
/gstack-upgrade's git path calls it after fetching. Below the incoming
release's floor the upgrade stops with BUN_TOO_OLD and the checkout,
VERSION and installs stay untouched. Vendored installs were already safe:
setup refuses before writing and the swap restores the previous install.
* 0.3: only main posts or closes tracking issues; reports carry alarms, red ledger, headroom and all-green
A branch-dispatched periodic or marathon run writes its report to the step
summary and a report artifact; the upsert and close steps run only on
refs/heads/main. The periodic body now keeps every pass-rate alarm line (the
old grep kept only the header), adds the red ledger and session headroom read
with an explicit --branch, the all-green estimate and the triage guide link,
and routes report text through scripts/lib/published-text.ts. The marathon
lane gains the matching close-on-green step.
Migrated: evals-workflow-wiring pinned report-summary.md in the upsert step;
the body is now written by 'Write the census report'.
* C3: prosons neutral-posture detector ignores a negated "taste call"
Both plan-review-prosons-neutral-neg reds (37179171083 t3, 37193478719 t3)
recommended A and closed with "a coverage call, not a taste call"; the
old /taste call/i matched the negation. The detector moves to a helper so
the two census captures replay free as negatives, with positive controls.
* B2: credit the CEO mode handoff only when the chat shows its line
Claude Code 2.1.284 collapses the gstack-ceo-mode-handoff Bash result, so
the printed AUTO_DECIDE line is user-visible only when a later chat message
of the session begins with it. The native detector now requires that
witness; the helper's tool result locates it but no longer replaces it, and
a generic mode statement cannot stand in once a handoff ran. Census
37182865432 is relabelled a genuine miss (expected: refused): the chat said
"I'm defaulting to HOLD SCOPE…" and "Mode decided: HOLD SCOPE". Corpus
replay: 5 credited, 3 refused. ceo-mode-handoff tests migrated to the
witness rule (deliberate behavior change).
* test: baseline the wave's approved detector and harness replay tests
B1 corpora, B3, C3, C4 and D2 replay captures through paid-case detectors,
actors or fixtures (test code by design); each entry carries its reason.
* E6: live Codex multi-block periodic case in the host-run Codex job
test/codex-e2e-multiblock-live.test.ts runs ./setup --host codex once from the
job's checkout into a fresh HOME and CODEX_HOME (auth.json copied from the CI
login home), asserts codex debug prompt-input lists the gstack router as a
real file, seeds one learning, and runs /learn's Show recent block through
codex exec: that later block must root itself through its prelude, run the
installed gstack-learnings-search and print the seeded learning; no block may
report a missing install or an unresolved gstack helper path.
The file joins CODEX_CI_FILES so the planner gives it to the host-run Codex
job. paid-test-durations.json carries a 180 s packing estimate (refreshed by
--write-durations after the first census) so the plan keeps every slice
within max-parallel. Oracles live in test/helpers/codex-multiblock.ts with
free replays in test/codex-multiblock.test.ts. Registration (E2E_TOUCHFILES,
E2E_TIERS periodic, E2E_KINDS rule) is the integrator's.
* E5: design live-model check as a periodic case
test/skill-e2e-design-model-smoke.test.ts runs design/scripts/live-model-check.ts
(default models only) under a temp HOME. Without OPENAI_API_KEY the case is a
Bun skip that the census lists as SKIPPED with no coverage credit; with a key,
exit 0 passes and any other exit fails with the script output (exit 2 under a
present key fails as 'nothing was checked'). paid-test-durations.json carries
a 90 s packing estimate. Registration (E2E_TOUCHFILES, E2E_TIERS periodic,
E2E_KINDS rule) is the integrator's.
* A1: docsync parent points children at a saved interface file
The /ship docsync fixture saved its observation interface only in the
parent prompt and told the parent to copy it into every child prompt,
about 30-40 s of parent generation per session (lane S, 37198445662).
docsSessionOptions now writes it to <home>/fixture-interface.md (0600) and
the parent tells each child to Read that file first. Fault-actor prompts,
which prepare already writes to a file, are unchanged. Fixture-only.
* A3: a JUnit case's recorded wall is capped by its shard's wall
junitCensus summed a case's JUnit test times as duration_ms; for
plan-mode-no-op's concurrent PTY tests that is 346 s inside a 176 s shard
(the "432-613 s" history). The case wall headroom shows is now
min(sum, shard elapsed), a true upper bound. Session headroom already
used the ledger's per-session clocks.
* A3: refresh paid-test durations from census 37198445662
Recorded 2026-09-29 estimates under-packed gate slice 11 (17 files
estimated 1078 s ran 1353 s; plan-review-report 76 s -> 198 s,
plan-ceo-plan-mode 37 s -> 112 s, safety-codex-consult 83 s -> 150 s).
Refreshed both tiers with --report --write-durations from the latest
census's slice walls. Old plan re-costed with observed walls: worst slice
~677 s (slice 11); refreshed plan: every slice estimated <= 539 s.
* C2: Browser Quick gets a feasible budget; QA loop names its directory and stop path
Judge rationales on the qa-only panels (2 of the last 3) cited the 30 s
Browser Quick budget: homepage + 5 targets, each a checkpoint write and a
guarded command, left ~5 s per probe. Quick is now 180 s (method text: 3
minutes); the qa-only-no-fix paid check expects the same native deadline.
PROBE_DIR points at its source line, a stopped loop goes to the final
report steps (annotations, materialize, Markdown), and /review and /ship
say that after smoke expiry only plan checks rerun in revalidation.
* C2: name the flows judges found implicit in review resolvers
Structured outside review: MISSING COVERAGE asks no fix question and is
never reported clean (was "preserve the existing user decision flow").
/review's HIGH-impact plan gate: spawned or non-interactive runs report
REQUIREMENTS MISSING and continue, writing no TODOs. Value cards: the byte
clamp applies to card lines; JSON records you write keep full values.
* C1: a free test catches a removed workflow step, which the judges missed
C1 controls showed the qa workflow judge passing /qa without Phase 10 and
the review judge passing /review without Step 3. The new test pins every
step/phase heading of qa, qa-only, review, ship, plan-ceo-review,
plan-eng-review and setup-deploy (derived from the templates' step
headings) and names any that goes missing or out of order; new step
headings must be pinned. The judge rubric is unchanged (EVAL_POLICY).
* size: fit the wave's wording under ship, review and qa parity caps
Measured after all lane merges: ship -79, review -210, qa -68 bytes over.
Trimmed the added wording first (post-expiry rule, MISSING COVERAGE,
plan-gate default, value-card JSON, stop line) and folded /review's
redundant INFORMATIONAL sentence and findings bullet. ship 268,036 (room
8), review 130,118 (room 1). qa needed 11 more bytes: cap 1.11 -> 1.1102
with the measured 93,305 / 84,049 = 1.11013 and its reason beside it.
* docs: troubleshooting sections for the wave's new messages
Bun floor (setup, auto-update, /gstack-upgrade), /review stale base and
fingerprint temp-dir errors, the design image model override, the
unrecorded Chromium PID, and a contributor section for census report
lines (pass-rates flags, failure causes, headroom, cost unknown, overlay
v4, stale corpora, auq substance). A free test checks every
troubleshooting.md#anchor printed by setup or bin scripts exists.
* TODOS: reliability follow-ups deferrals; close the Bun floor, make-pdf $P and #2709 reap
* 0.4: 'A census went red' guide; docs for causes, session ledger, headroom and new messages
docs/evals/census-red.md walks one red line to a diagnosis: prerequisites, the
report fields, eval:pass-rates --run/--reds/--headroom with expected output,
what unknown/insufficient/unrecorded data means, free replay, and the single
'after a repair' paid run allowed by TESTING_INTERNALS. Linked from
CONTRIBUTING, TESTING_INTERNALS and both census reports.
TESTING_INTERNALS documents the failure_cause glossary (one row per
FAILURE_CAUSES value, diagnostic only under EVAL_POLICY v1), failure_detail,
sessions[], cost_known, the session ledger, the pass-rates views and the 85%
headroom alarm, main-only tracking issues, overlay contract v4, auq-matrix's
3-sample substance panel and the E5/E6 periodic cases. troubleshooting.md
gains #bun-too-old and an anchor per new user-facing message from lanes P, D
and F. README names GSTACK_DESIGN_IMAGE_MODEL. test/census-red-docs.test.ts
pins the glossary to FAILURE_CAUSES, parses every guide command and checks the
links.
* v1.91.20.0: CHANGELOG and version for the reliability follow-ups
* regen: SKILL.md and sections for all hosts, digest, context-budget fixture
bun run gen:skill-docs --host all; goldens and llms.txt re-rendered
unchanged; bun test/helpers/capture-context-budget.ts.
* integrate: register codex-multiblock-live and design-model-smoke (periodic, rule)
* 0.4: census report and eval:pass-rates views link the census-red guide
* test: cookie approval fixture reverses the Bun install-hint pin (E1)
The historical approved cookie-workflow prompt carried BUN_VERSION=1.3.10;
the regenerated install hint now pins 1.4.0, so the fixture's reversal
list gains that edit like the earlier waves' entries.
* test: release floor counts the two new periodic cases (E5, E6)
* fix: /freeze, /guard, /unfreeze and /investigate work on Codex and other env-var hosts
Templates quoted "$HOME/.claude/skills/gstack/...", which the env-var hosts'
rewrites turned into "$HOME/.agents/skills/gstack/..." on Codex (a path a
global install never has; same class on the other hosts), and setup's
runtime roots had no freeze/bin. Quoted runtime paths now use
{{RUNTIME_ROOT}}: $GSTACK_ROOT on env-var hosts (the fence gets the
runtime prelude), the unchanged $HOME path on Claude. _link_runtime_dists
links freeze/bin and the careful/bin helper it sources into every
env-var host's root. Also fixes the same guess in /benchmark-models,
/retro and /diagram. test/env-host-runtime-assets.test.ts fails on the
previous code: no fence guesses a $HOME/<dir>/skills/gstack path, every
$GSTACK_ROOT path a fence uses is in the host's runtime root, and the
installed Codex /freeze and /unfreeze fences run from a fresh shell.
Template-reading tests now read the Claude render or expect the
placeholder (deliberate change).
* CHANGELOG/TODOS: lane W's workflow, cases and guide; the env-var host freeze fix
* E6: a block that reads a SKILL.md is not a missing install
The first live codex-multiblock-live run passed its real check (the /learn
block found the install and printed the seeded learning) but the oracle
flagged a sed of the learn SKILL.md, whose prelude source quotes the
'gstack: no install found (tried $_r)' text. Only an emitted line that
starts with the message and names a real root counts now; free replay
negative added.
* test: qa-only fixture keeps its caller-set 30 s deadline
The qa-only-no-fix fixture tells the model to start the guard with 30 s
(a caller limit), so its deadline assertion's default stays 30 000 ms; the
product's Browser Quick default (180 s) does not apply to it. Reverts the
helper and its test to their base.
* C2: keep PROBE_DIR 'owned' (qa-functional-prompt pins it); qa measured 93,311
* test: paid file count and periodic floor include the two new periodic cases
* test: ship docs freshness pin checks meaning after A1's wording trim
The snapshot still hashes selected release paths (--select narrows them),
generated outputs and docs; the pin now checks that instead of the
pre-trim sentence.
* E3: reap the recorded Chromium only after the daemon has exited
With the PID now recorded (E3), browse stop reaped Chromium while the
daemon was still shutting down; the daemon read that as a crash, exited 1
and left its state file (chromium-profile-isolation's headless stop test
failed 3/3 locally, passes on base). The CLI waits up to 5 s for the daemon
to exit before reaping a survivor.
* regen: qa and qa-only exploratory sections (PROBE_DIR wording)
* B1: batching counter reads a target declared mid-line; replay corpus
GSTA-23's census 37228573062 failed plan-eng-multi-finding-batching with
reviewCount=0 after 975 s although D1-D10 were asked one per finding,
each with its saved decision-ledger record. The report wrote 'Report
destination requested by the user. Review target (fixed): `PLAN.md` ...'
and the counter read target fields only at a line start, so no brief
inherited the plan identity. A target field may now open any sentence of
its line; a mid-sentence mention or another file still gives no identity.
Floor (3), budget and kind unchanged. Corpus: 9 earlier passing censuses,
the red (now credited at D5), and a derived batching refusal; 975 s noted
in the headroom diagnosis.
* B1: shared-libs corpus labels its git path and covers the supported shape
Every census entry was recorded in CI's eval image (git 2.43), where
gstack-safe-git falls back because --no-lazy-fetch is rejected; each entry
now says so (git_path: fallback). Two synthetic supported-path entries
(HEAD and a history citation in the brief, as git >= 2.44 produces) keep
the 37193478719 pass and the 37195203538 refusal, and a test asserts both
shapes are covered and the actor reads no git-path-specific text.
* v1.91.21.0: move off v1.91.20.0 (claimed by #3031)
* test: settle nav-guard commands before asserting; make two new tests Windows-safe
navigation-guard: Bun's expect(promise).rejects slows each Playwright round-trip
to ~1s while it waits (50 page.evaluate: 43ms plain vs 46.6s inside the matcher),
so the click's 5s timeout expired before the guard saw the link-local request
(CI: 'click: Timeout 5000ms exceeded ... waiting for scheduled navigations').
Settle the command first, then assert on the refusal; a source check keeps it so.
env-host-runtime-assets: PROBE_FILES now matches either path separator.
detector-corpus ship-docsync-late-result: JSON-escape the <ROOT> substitution so a
Windows temp path's backslashes do not break JSON.parse.
* fix: CI run/job ids are not phone numbers; context-recovery fixture gets the full runtime bin
pii.phone.e164 redacted the run id in the published eval report's fetch command
(--run 37235771700), breaking the command it prints. Digit-only spans in a run/job
flag, gh run argument, actions/runs/ or /job/ URL segment, or run_id/job_id key are
now exempt; the same digits elsewhere still report.
context-recovery-artifacts: the fixture's ./bin/ lacked gstack-skill-start and
gstack-context-recovery (the preamble's runtime scripts since v1.71), so the agent
spent its 10 turns hunting for them and hit error_max_turns with no final message
(run 37235771700, 11 turns). Link the whole runtime bin; diagnostic rerun passed
in 6 turns.
* test: replay run 37237194905's late-result capture through the docs authority detector
The only failure was call 20's write to a mistyped shard root (3GanyY for 3GanyU),
which the model disclosed; the detector is correct and the A1 post-snapshot
--out under the fixture home stays permitted. Named red recorded in TODOS.
* test: keep docsync-authority in the Windows profile
The replay test's literal helper path tripped the Windows curator's shebang-spawn
pattern, dropping the whole file from the Windows profile (test-free-shards
invariant). The snapshot --out case is already covered by docsync-command-grammar,
so it is removed here, and the Linux-path replay skips on win32.
* fix: gstack-docs-candidate --help works; docs gate red lines name the denying tool failure
PR run 37239016858 ship-docsync-completion: the parent ran
'gstack-docs-candidate --help' (the helper's usage error points at --help, which
failed). That undeclared command made docsToolFailures non-empty, which denied
the atomic-replacement proof of the child's legitimate native Edit, so the red
line blamed the Edit's temp file. A replay of the capture proves the atomic
replace once --help is declared.
- the helper prints its usage for --help/-h and exits 0
- the fixture interface and command grammar admit the helper's bare --help
- the atomic-proof denial names its check and the tool failures behind it
(deny labels renumbered to their lines)
- replay with negative controls (undeclared command, temp left behind, renamed
elsewhere); diagnostic rerun of ship-docsync-completion passed
* test: retro-base-branch fixture uses main and a shallow report path
PR run 37241493377: the agent wrote retro-output.md one directory above the
requested <tmp>/retro-base/retro-output.md (it dropped the retro-base segment),
so the report check found nothing. The fixture also ran plain git init, which
makes master on CI's git while the prompt says the repo is on main; the agent
logged that mismatch as a learning. Init on main and use the temp dir itself as
the repo. Diagnostic rerun passed (18 turns, 203s).
* test: qa-quick sets a caller probe limit and drains past its session budget
C2 raised Browser Quick's native budget to 180 s. qa-quick is the only gate case
on the native budget, and its wall went 194 s (main periodic, pre-C2) -> 254 s and
274 s (this PR's runs) -> 300 s session timeout (run 37242691913): the agent now
probes every navigation target with a checkpoint, then needs ~80 s to report.
Like qa-only-no-fix, the eval now sets a caller probe limit (30 s), scopes the
fixture, and gives bun's outer timeout the session drain margin instead of racing
the 300 s session. The product's 180 s budget and the session budget are unchanged.
Diagnostic rerun passed in 195 s.
* review/ship: use main's post-expiry smoke sentence; v1.91.24.0
GSTA-23's measurement on the review workflow judge: main's 'Post-expiry smoke
rechecks are not-run.' passed 10/10; the C2 wording passed 6/10 and the
'..., even in 4c.' variant failed in PR run 37248650879. qa-probe-gates pins
main's sentence before step 4c.
* test: record the plan-design corpus test in the test-of-test baseline
#3031's ratchet now counts tests that slice a paid test file's source; the B1
plan-design-review-plan-mode replay corpus reads test/skill-e2e-design.test.ts
like its sibling corpora already listed.
* fix: --case no longer deadlocks importing the running CLI
runCaseDiagnosis loaded runPaidShards with a dynamic import of
scripts/test-paid-shards.ts; when that file is the entry it is still in its
top-level await, so the import never resolved and `--case` hung after its
first line (on main since #3031, found by the per-case measurement runs). The
CLI now passes its runner in; library callers keep the lazy import.
* test: name the in-flight file and the native error when bun dies before its summary (#3037)
* Bun 1.4.2 everywhere CI and setup pin the tested version
Bun 1.4.0 on Windows aborts about 3% of free-suite shards with native
bad-handle errors (main too; #3037); the same head on 1.4.2 had none in 180
shard runs. Every current 1.4.0 pin moves together, as bun-version-drift
requires: setup-bun in every workflow (CSO orchestration included), the eval
image (sha256 from bun-v1.4.2's SHASUMS256.txt, re-hashed locally), GitLab,
the Ubicloud free-suite setup, gstack-bun-version.sh's tested version,
engines.bun and @types/bun, the install hint in generated skills, the Dia and
cookie native-qualification pins, and docs and tests that state the current
pin. Recorded 1.4.0 observations, fixtures, CHANGELOG history and the CSO
images' oven/bun:1.3.10 base stay.
* fix: --case runs a single-case file whole
A session option naming the case (testName: 'office-hours-section-loading')
made caseSelection pick test-name mode, whose filter matched no Bun test, so
every trial would run hollow. A file that registers only this case now runs
whole before any name match is tried.
* fix(office-hours): reviewer returns a hash-bound receipt instead of echoing its verdict
The spec reviewer now Writes its verdict once, seals it with
gstack-office-hours-review seal (validates the saved JSON and prints
OFFICE_HOURS_VERDICT round=N sha256=<hash> path=<verdict>), and returns
only that line. check --receipt recomputes the hash and fails closed on a
missing, malformed, stale-round, wrong-path or mismatched receipt. The
saved file stays the sole findings inventory; validators bind the receipt
to the saved bytes and the judge reads those bytes.
* fix(office-hours): only blocking findings send a design back for another review round
Review findings carry severity blocking|minor (verdict schema v2). PASS when
no blocking findings remain; minor findings are recorded in Reviewer
Concerns and the Spec Review table, never a reason for another round.
CONVERGENCE needs a persisting blocking prior. A persisting or unverified
blocking prior cannot be relabeled minor. The 3-round cap is unchanged.
* fix(office-hours): rounds 2 and 3 are delta re-reviews of the captured design diff
prepare saves each reviewed design as round-N.design.md and gives later
reviewers the exact line diff since the previous round. A new later-round
blocking finding must cite (changed_text) an excerpt of a changed line;
seal, check and the validators verify it against the snapshots and fail
closed when a citation or snapshot is missing. Other new observations are
minor. Severity, anti-downgrade and CONVERGENCE rules are unchanged.
* test(office-hours): preservation judge accepts COMPLETED with recorded minor findings
* fix(office-hours): parent fixes only blocking findings; callers may set --max-rounds
Between rounds the parent edits for blocking findings only; minor findings
stay recorded. gstack-office-hours-review takes --max-rounds 1|2|3 (default
3) on every command; reaching it stops at MAX_ITERATIONS. The marathon
fixture sets 2 rounds so round 1 (full) and round 2 (delta) run live.
* docs(changelog): office-hours review loop changes in v1.91.28.0
* fix(cso): helper JSON keeps run URLs only in the validated workflow slot
The shared redactor now treats GitHub Actions run ids as CI metadata
(
|
||
|
Garry Tan
|
7fca42ad8b |
v1.91.12.0 v1.91.12.0: audit fix wave, ~11-minute paid eval lanes, eval reliability policy (#2999)
* test: delete test-infrastructure dead code (G) - exit-propagation drives the runner's real strict verdict (BunTestOutputClassifier + strictTestExitCode); delete the unused shardRunLooksTruncated predicate. - delete skill-coverage-matrix registry + its gate (nothing reads it; the floor already iterates skillCensus()). - delete touchfiles-facade export-parity tests (Bun fails missing imports at link time) and the duplicated E2E_TIERS tier-value test. - delete brain-cache-spec TRANSPORT_DEFAULT_POLICY, SKILL_RUN_RETENTION_DAYS and the now-unused BrainTrustPolicy type with their literal tests. AUTOPLAN_PREFLIGHT_BUDGET_BYTES stays: skill-preflight-budget enforces it against real resolver output. - delete audit-compliance's JSDoc-comment grep. * test: replace product tests that fake the product with real-boundary tests (F) - design: serve.test.ts drove an inline mirror server; now two tests run the real serve() on an ephemeral port (reload confinement, submit exit 0). - setup-gbrain: rollback + voyage tests execute the template-extracted init blocks (3 sites) instead of drifted local bash copies. - terminal-agent: internalHandler source greps replaced by a behavioral /internal/grant + /internal/revoke auth matrix (no/wrong/valid token). - /health: server-security-surface and the server-auth / security-audit-r2 / sidebar-tabs source greps fold into one liveness-only check on the real body; the L4 sidecar wiring gets a behavioral /pty-inject-scan test. - delete tautologies (browser-manager onDisconnect, memory-command #12), ios swiftui tap fixture self-check, memory-ingest put_page grep, detach source greps, sidebar-agent absence pins, dead-CSS pins + the dead CSS, security-audit-r2 Task 1 + the test-only meta-commands re-export, duplicate generated-SKILL.md checks. - make-pdf coverage-gaps cases move into their owner test files. * test: delete tests of dead eval code (A) - A1: the retired Eng lexical oracle (evaluateEngSeedCoverage, isEngSeedDecisionAUQ), the completion-handoff detector and the retained corpus had no paid caller since v1.87.6; delete their 26 replay files, ~2.6k helper LOC and fixtures, and the dead blocks in 8 mixed files (live hasNativePlanTerminal / batching assertions stay). - A2: dead viewport approvers in autoplan-artifact-permission and their 11 replay files + fixtures; recorder/launcher cases stay. - A3: never-wired oracles and seeders (autoplan-phase-order, eng-finding-fixture, ceo-paired-fixture, design-ui-scope, plan-skill-completion, pty-current-screen, required-reads, transcript-section-logger); plan-seed-submission now decodes through the production createPtyScreen; section manifests name their actual guard. - A4: zero-reference helper exports, plus execGit and invokeAndObserve found by the reachability pass. - 52 fixtures orphaned by the deletions; touchfile and selection-table entries for every deleted path. * test: clean up the paid eval lane (B1-B4, B6, B7) - B1: delete paid files that assert nothing or cannot pass meaningfully: skill-llm-eval-spec and skill-e2e-spec-execute (test.todo), gemini-e2e (+ gemini-session-runner; no gemini CLI in CI), ship-idempotency (red since v1.63), the two opus-4-7 *-sonnet overlay wrappers, conductor-prose (+ its source-evaluation replay), codex-e2e-plan-format; drop their keys, scripts and census rows. - B2: skill-llm-eval grades browse/sections/command-list.md with one union judge that also carries the baseline score pin; regression-vs-baseline deleted (paid run: pass, c4/c4/a4). - B3: memory-pipeline, ios-qa, ios-qa-swift-build and plan-tune-cathedral make no model calls; renamed out of the paid glob so they run on every PR. Swift builds need GSTACK_TEST_SWIFT=1; device stub deleted. - B4: codex-e2e*, outside-voice, aside and ios-device cannot run in the CI image; excluded from the weekly lane with a tracked re-entry condition. - B6: fold opus-47's negative routing controls into skill-routing-e2e journey-negatives (paid run: 3/3 unrouted) and delete the file. - B7: delete the never-green brain-privacy-gate eval; a free gstack-skill-start test now proves consent precedes artifacts egress. * test: retire the finding-count cluster and trim its helpers (C) - C0/C1: the five never-green evals (skill-e2e-autoplan-chain and skill-e2e-plan-{ceo,eng,design,devex}-finding-count) failed on harness and budget, never on skill behavior; delete them, their touchfile/tier ids, AUTOPLAN_CHAIN_BUDGET and the dedicated eighth periodic slice (--slices 7). - C2: delete the helper groups whose only paid consumers were those files (11 modules), trim claude-pty-runner and eng-seeded-coverage to the paid closure, and delete the free replay tests whose assertions exercised only that dead code (89 files, 135 orphaned fixtures). Blocks that used dead code only as input for a live subject keep their assertions: the multiSelect default moved to plan-review-decisions, runner PTY tests use inline caller policies, and the timer-safe budget checks moved to eng-finding-retry-budget. - The eight production-touching files stay except ceo-current-decision-record (its template read only feeds the retired counter). - CARVE_GUARDS.autoplan is behavioral 'none'; TODOS records the lost chain and per-finding cadence coverage with their re-entry tests. * test: fold per-incident replay series into their detector owners (D) Twelve detector families move into one owner test each: 73 incident files become describe blocks in ceo-section-loading-fixture (stale-fill race), model-overlays, coverage-audit-evidence, autoplan-phase-observer, native-auto-decide, outside-voice-evidence, eng-first-review, plan-count-completion, plan-count-file-permission, ceo-mode-option, plan-scope-selection and plan-count-prerequisite. Each block keeps its original code and fixture, so every case still runs; only tests asserting the incident file's own touchfile registration are dropped (41). Touchfile lists that named an incident now name its owner. * test: start the plan-count history PTY on its readiness marker (H) The fake CLI prints a startup marker and the runner waits for it instead of the fixed 8 s startup sleep (8.6 s -> 0.9 s locally). eng-semantic-terminal's sleeping registration cases went with C; plan-count-timeout keeps the fixed wait because it asserts deadline behavior. * test: derive paid touchfiles from each eval's static closure (E) touchfiles.test.ts now checks, per key, that the paid file's static test/helpers and test/fixtures closure (plus fixture paths it names in string literals) is covered, and names the file, path, chain and key to fix when it is not. Free *.test.ts files are no longer touchfiles, so editing a free replay test stops selecting paid evals: 950 entries removed, 653 real closure paths added. The hand-copied inventories go: periodic-fixture-selection, fake-impeccable-touchfiles and 45 per-file selection examples. Selection for the sample edits (plan-eng-review template, claude-pty-runner, plan-count-fixture, gstack-config) loses no case under either profile. CONTRIBUTING documents the rule and its lower bound. * test: skip hollow tier shards and census judges in the paid planner (B5) A paid file is now skipped for a tier lane only when every E2E id it registers is known statically and none has that tier; ids come from the touchfile registrations and literal testName/*IfSelected arguments, so a comment or skill path that quotes another id cannot unschedule it, and computed names keep today's scheduling. --list and the manifest show each skip as "skipped: no E2E_TIERS id has tier <tier>". The weekly gate census drops the LLM judges (--skip-judges); they still run in the periodic census and PR gate lanes. Gate lane 52 -> 42 files, census 41; periodic 77 -> 69. * test: run seven paid evals on the current default capture model (B8) skill-e2e-{auq-matrix,plan-format,qa-bugs,retro,workflow} pinned claude-opus-4-7 and skill-e2e-office-hours plus -brain-writeback pinned claude-sonnet-4-6; none tests a historical model, so they now capture with resolveEvalModel('capture'), and the free harness tests that execute these registrations receive the same resolver. The paid re-pin run passed all of them. skill-e2e-{design,office-hours-phase4,plan-prosons,plan} keep claude-opus-4-7: six of their cases failed on the default model (three timeouts, a missing report file, a format miss and a posture score of 3), so per the plan's fallback they keep their pins with a TODOS entry. The pre-spend estimate and drop threshold are in docs/test-audit-2026-09.md. * test: guard the reduced suite against new test-of-test files - test/test-of-test-ratchet.test.ts records the 228 free tests that import only test/ code and fails on a new one, naming the owner test to extend instead; a stale baseline entry fails with the remove instruction. - test/helpers/resolve-repo-path.ts is the one specifier/literal resolver for the ratchet and the touchfile closure invariant, with its own unit tests. - CONTRIBUTING "Test tiers" describes the paid-failure workflow (fix, then one row in the detector's owner test) and the ratchet; TEST_PORTFOLIO gains the detector -> owner-test table and no longer claims an Autoplan chain eval. - TODOS: automatic exclusion policy for chronically red periodic files (P3), the deferred native-completion table collapse, the unused CEO payment seeder; the PTY readiness item is narrowed to the paid runner. - docs/test-audit-2026-09.md collects the triage, security mapping, inventories, selection proof, behavior-commit decisions and retained false positives. * v1.91.8.0 test: smaller suite, derived paid selection, retired never-green evals Release metadata for the test-reduction branch: VERSION 1.91.8.0 (1.91.7.0 is claimed by #2983), CHANGELOG with the measured before/after table and a contributor section, durations re-recorded on Ubicloud standard-16 (857 files, 0 failures), the agents digest, CONTRIBUTING's after-measurement row, the B8 fallback TODOS entry, and the after metrics, kept-vs-plan notes, B8 run and census estimate in docs/test-audit-2026-09.md. * fix(ubicloud): skip retrieval globs that match nothing instead of reporting a failed pull * test: pin DISABLE_AUTOUPDATER in hermetic env and capture corrupt-seed warning Both EVALS_HERMETIC branches of buildHermeticEnv now carry DISABLE_AUTOUPDATER=1 (the allowlist scrubbed the workflow's copy, so every PTY screen showed the updater's npm-prefix failure). Per-test overrides still win. The corrupt durations-seed test now captures its expected warning and restores the console spy. * style(cso): format lib/cso TypeScript with pinned Prettier Mechanical reformat only. Minified transpile output is byte-identical for 21 of 22 files; witness.ts differs only in three regex flag orders (/mi -> /im), which JavaScript canonicalizes. Source-text assertions over lib/cso now compare whitespace-insensitively with the same tokens. * fix(cso): import join for compiled-launcher assertion witnesses Compiled installs always take the non-Bun branch, which called an unimported join and threw before any runtime-tested assertion could be witnessed. The child command selection is now a pure, platform-aware function; a missing sibling launcher fails with its expected path. * fix(browse): make connect --supervise actually respawn a crashed server The supervisor respawned with a block-scoped env that no longer existed, so every attempt threw and the loop gave up after five tries. The headed env is now one pure helper used by connect and respawn, the loop is an injectable runHeadedSupervisor with behavioral tests, failures name the daemon log and relaunch command, and connect's usage advertises --supervise. * test: one finite PR world for the shared-libs fixture; name dual-voice probe evidence The shared-libs shim served 2 PRs for pulls?state=all and endless full pages for state=open. gh pr list, pulls?state=open|all|closed (per_page/page, short last page, direction) and search/issues now page one deterministic table: PR 7, 600 older open PRs, PR 42 and 3 closed PRs, so five 100-item open-metadata pages still leave older open PRs unchecked. The Contents API lists pinned directories (the captured attempt got 404 for contents/ and contents/src while files resolved, then fell back to a raw host), unknown endpoints return 404 instead of repo metadata, and the read-only detector is unchanged. Free tests cover view agreement, the budget bound, gh/curl agreement and the empty world. Dual-voice outside-voice failures now report probeToolUseId, probeMode and the canonical-match result with the reason the probe output was rejected. * feat: require a zero-error product typecheck and a test type-debt ratchet Adds tsconfig.json (strict) over product code, fixes its remaining 90 diagnostics (type-only, interface corrections, and explicit narrowing), and adds a typecheck job to the required free-tests aggregate running bun run typecheck, the test-code ratchet (identity -> count baseline, fails on new, repeated, or unlocked fixed diagnostics), and the lib/cso format check. Reuses fixes from #2447 where they still applied. * test: follow the headed env helper and the typecheck gate in source-shape checks * fix(test): pin the package.json change kind in shared-input selection tests computePaidCaseSelection read the version-only exemption from git even when changed files were injected, so the shared-input test failed on main and on version-only branches. The exemption is now an optional input; the test pins a real package.json change and covers the version-only case. * test: judge plan-count completion on structured evidence, not wording Replaying run 36385945043's two Design attempts showed the existing routes rejected correct endings: attempt 1 at the typed-completion path field ('- Reviewed plan written to …' is not a 'Plan written to' line), attempt 2 at the leading-fence veto (its final message opens with the dashboard). nativePlanTerminalPreconditions is the structural prefix of hasNativePlanTerminal (behavior unchanged). structuredPlanCompletion adds, inside the existing nativeSummary branch: a complete report (Design binding for Design), a completed review-log row for the expected skill appended during this attempt under the child's GSTACK_HOME/project slug (resolved with bin/gstack-slug) and stamped with the fixture commit, timed between the report/last answer (second resolution) and the final native message, a final message with stop_reason end_turn (now carried on public transcript messages), and no visible question or permission prompt. Timeout summaries add idleFor and lastTerminalCandidate. Terminal and throw captures copy the plan file and review-log rows into the artifact directory; copies are best-effort and recorded in evidence-copy.json. Free regressions: both captured Design endings (trimmed fixture with provenance; report, row and end_turn reconstructed and labelled), the negative controls, and real-PTY completion/timeout runs through the real review logger. * test: structural Design count boundary; TODO proposals are not findings Replaying run 36385945043 through the Design count predicates: routing, focus and learnings setup was not recognized as setup, Issue 1 was counted pre-review in both attempts (the boundary fired on it), and attempt 2 counted the Font TODO proposal as a finding (review=4 and review=5 for five issues). The paid caller now starts review at the first answered native decision that is not setup (recognized packet, or setup header/question ID), a completion handoff, artifact rendering or a TODO proposal (the review's Add to TODOS.md / Skip / Build it now menu). TODO proposals are recorded as administrative extra decisions. The replay asserts each counted call: both attempts review=5 (Issues 1-5). isDesignCountFirstReview and its controls are unchanged. * test: CEO classifier throws name the question and matched predicates Replaying run 36385945043's FAN-1 and ERR-1 throws (ledger rows reconstructed from rendered diffs) through ceoPaymentFinding: the email obligation's row, subject, option and proposal predicates pass and the ELI10 explanation-defect predicate fails first ('lets that exception fly out', 'the error bubbles up'). Binding the defect to the named ledger row instead (the planned fix) was tried and reverted: scoped to the email seed it flips 30+ existing cf74 still-rejects replays, which require a vocabulary-free, ledger-bound email question to earn credit only through a complete saved comparison. With FAN-1's rendered currentDecision payload reconstructed, the recorded- decision path counts it, so the real saved plan (not uploaded) must have differed; failure artifacts now retain it. The classifier stays fail-closed and unchanged. Its throw now prints the header, the first 200 question characters and each obligation's predicate results. Free regressions with provenance and negative controls: an unrelated question, an email question whose row says it is already rescued, and a ledger ID whose row belongs to another seed. * chore: regenerate the test type-debt baseline on top of #2994 * fix(typecheck): strip the checkout root from ratchet diagnostic identities * fix(test): recognize ledger row-ID split candidates so collection stops at the last ACK Run 36385945043's split-overflow case asked all five candidate decisions by 8m55s, but the live candidate check required the question to open with "E1:" and every option to be a known disposition. The skill cited ledger row IDs ("D2.1 — R-E1: …") and offered "Hold, discuss first", so no candidate was recognized and the attempt ran the whole review (1302s). Identity now comes from the native header; the question must open with that candidate's ledger reference, name only that candidate, and offer exactly one include, defer and cut disposition. The selected answer must still be one of those three. The semantic evaluator and every existing negative control are unchanged; a trimmed capture from the run adds the positive case and four row-ID negative controls. * fix(test): stop the eng batching eval once its floor is proven The case's only verdict is reviewCount >= FLOOR (3). Run 36385945043 had three distinct acknowledged review decisions at 6m41s but kept answering until the ceiling (7) at 12m13s. The registration now passes the runner's existing isCollectionComplete stop once FLOOR non-setup, non-administrative review decisions are acknowledged; the floor check, ceiling, budget and counter are unchanged. A child-process registration test proves the stop predicate and that below-floor and timeout outcomes still fail. * test: add the non-blocking 'marathon' E2E tier Full start-to-finish flows move out of the blocking lanes. E2E_TIERS and E2ETier gain 'marathon'; describeE2ETier('marathon') is enabled only when EVALS_TIER=marathon, so the gate/PR and periodic lanes (and the gate census) never run those cases. The PR profile accepts marathon ids as scheduled elsewhere and defers them with their own reason, even on full fallback. * test: move the full office-hours workflow to marathon; add a periodic design-draft checkpoint The full startup workflow runs 1–3 real spec-review rounds (~280s each) and hit its 1200s capture in run 36385945043 at finalize. Review depth is the product's loop, so the case cannot fit a blocking lane without cutting rounds. It is now marathon tier with every assertion unchanged. skill-e2e-office-hours-design-draft.test.ts (periodic) runs the same fixed interview only through the Write that creates the design (269s in that run) and applies the full validator's design-draft checks, the required section reads and the launch/foreign-skill-read guards. validateOfficeHoursDesignDraft is extracted from validateOfficeHoursCompletion, which still applies it. Selection: office-hours-design-draft is registered periodic; the marathon-only file is already excluded from the gate and periodic plans by the B5 planner rule. Tier-alignment regexes and the valid-tier check accept 'marathon'. A type-only cast in plan-scope-selection.test.ts removes a diagnostic whose union print order made the ratchet identity unstable; baseline tightened. * test: supply the split-overflow fixture's HOLD SCOPE mode as a prerequisite The split actor always answered 0E's mode question with HOLD SCOPE. The skill skips that question on an explicit choice, so the fixture now states it and the attempt starts at the five candidate decisions (about 1.5 min earlier in run 36385945043). Candidates, actor policy, floor and semantic evaluation are unchanged; the fixture test pins the supplied choice. * test: start the eng batching eval with its setup prerequisites supplied Routing setup and cross-project learnings (D1/D2 in run 36385945043) are never counted and are not what the case measures. The registration now uses the runner's existing preconfiguredReviewActor so the attempt starts at the review; engSetupAUQ still vetoes any late setup question. The registration test pins the option. * test: count the design-draft paid file and defer marathon ids in PR selection pins The discovered paid-file census grows by one (skill-e2e-office-hours-design-draft). Full-fallback PR selection defers every non-gate id; the shared-input pins now expect periodic and marathon ids there. * fix(review): resolve the judged revalidation, setup-authority, plan-gate and findings-record ambiguities The census review workflow judge scored clarity/actionability 3 on both attempts: smoke-clock limits appeared to forbid post-repair revalidation, the caller deadline was undefined, 'ask for setup' conflicted with the report-only browser rule, fallback-sourced HIGH discrepancies had no gate decision, and the Step 5.8 record omitted adversarial findings. * fix(office-hours): load the builder section for every builder-mode reply Both census builder-wildness attempts answered a direct request for adjacent unlocks without reading phase-2b-builder-brainstorm.md, whose trigger read as applying only to the generative questions. * fix(sync-gbrain): define Step 4 helper args and one atomic write path Both census read-ready attempts spent turns reading the helper source to resolve <user-args>, inspecting fixture internals kept inside the repo, and reconciling 'Read + Edit' with the tmp+mv atomic write, then hit max turns before the verdict. * refactor(evals): share the import-closure walker and add the E2E shard reuse identity sourceDependencyClosure moves from the workflow-judge adapter into scripts/eval-input-cache.ts unchanged, so judge keys stay byte-identical. scripts/e2e-shard-reuse.ts builds the consumed-input identity of one PR-lane E2E shard (test import closure, every registered case's touchfiles, globals, runner/workflow/setup actions, child env pins, CI image, Claude CLI) and fails closed on anything unknown. Marathon joins the always-fresh purposes. * feat(evals): ~12-minute blocking paid lanes and a non-blocking marathon lane - Planner budget mode (--slice-budget S --jobs J): recorded per-tier wall times pack into as many ~9-minute executors as the work needs; the plan records per-slice estimates and the CI job timeout (supervised worst case + 20 min). evals.yml and evals-periodic.yml derive matrix size and timeout-minutes from it; max-parallel covers every slice at once. - Case shards: plan/design/review-army/shared-libs(-paths) run one registered case per process (<file>#<case id>, exact name pattern, exactly one case). - Retry rule: a timed-out attempt is a verdict. Only files whose every case budget is CAPTURE tier or shorter keep one retry; walls shrink to match. - Marathon tier: positive selection, excluded from gate/periodic planners, run by the new evals-marathon.yml (weekly + dispatch, fresh, own report). - PR-lane E2E reuse of verified first-attempt passes on identical inputs; the report rejects reuse outside the fast PR profile. - Duration seed from census run 36385945043, per tier and per case shard. * docs: blocking lane budget, marathon lane, retry policy and E2E reuse * chore(typecheck): lock in two fixed test diagnostics * fix(ci): drop a duplicated env/jobs block in evals-marathon.yml * test(ship-docsync): shard the doc-sync lifecycle by case and drop the duplicate dispatch-only case ship-docsync ran the same fixture and prompt as ship-docsync-completion and asserted a subset of it. The file now runs one case per process, so its lane wall is its longest case instead of half the sum of thirteen. * fix(evals): plan CI-unrunnable cases as excluded entries, not empty case shards design-review-fix drives the Aside browser and registers test.skip on Linux runners, so its case shard executed zero cases and failed the exact-one-case check in proof census 36597762183 (eval-slices 6). CASE_CI_EXCLUDE (reason + tracking, beside PERIODIC_CI_EXCLUDE) now turns such cases into excluded manifest entries that --list and the manifest surface; every planned case shard still must execute exactly its case. * docs(todos): list the case-level Aside exclusion with the CI-unrunnable evals * fix(plan-ceo-review): restore experience-first expansion framing, require the mode handoff, skip pacing menus Census 36597762183: both mode-routing runs logged provenance and moved on without the mandated handoff chat; the EXPANSION run asked an unauthorized batch/narrow pacing menu instead of the first per-addition question; the expansion-energy proposals led with the spec because v1.87.6.0 dropped 'lead with the felt experience'. The HOLD review detector also rejected a decision whose grounding line named no plan file although the owned source Read binds it. * test(outside-plan-disabled): bind quoted prior-record values by their sentence, not phrase order The parent obeyed the off switch and twice named the seeded completed record as pre-existing, once with the quotation after its owner and once with slash separators; the order-specific stripper counted both as current completion. Timestamp, location, current-claim and value-match controls still reject. * test(outside-plan-disabled): compare named record timestamps as instants; negated authorship is not a current claim The repair rerun named the seeded record by its ISO second (2026-09-29T16:58:52Z vs .727Z) and said 'I did not write'; both were misread as a foreign timestamp and a current write. * test(ceo-section-loading): recognize an arrow-ordered stale-fill execution by event roles The census review traced the seeded race as 'R1 miss -> R1 store read (v1) -> W commit v2 -> W cache.delete -> W fulfills -> R1 cache.set(v1) -> R2 (begun after W) hits v1', but the in-flight gate only accepted race vocabulary or fixed sentence shapes. Order, actor, version and dismissal mutations still fail. * test(design-floor): answer the seed-declared all-seven 0D focus menu while it is pending The actor declares 'Design: review all seven dimensions', but its picker reused designReviewSetupAUQ, which only matches already-answered calls (and a narrower header/label set), so the pending D1 focus menu was never answered and the case waited out its 609 s deadline. The skill's Step 0D requires asking; the fixture now answers it. * test(ceo-mode-routing): accept the skill-mandated Note form and Recommendation reason as HOLD posture HOLD Defer/Keep briefs must use 'Note: options differ in kind' (preamble), but the answered-HOLD path demanded a Completeness score, rejected a one-line Net with a semicolon, and read posture only from ELI10. The rerun's brief applied HOLD SCOPE in its Recommendation reason. Revert the ineffective 'always'/'handoff chat' wording: two runs still skipped the mode handoff. * test(qa-bugs): keep claude-opus-4-7 after qa-b6-static stalled on the default model qa-b6-static timed out on claude-fable-5-1 in census 36597762183 and in one of two targeted reruns. Both times the stream stopped mid-message with no pending tool, right after the model found the disabled submit button, and stayed silent until the 300 s deadline. Per the B8 fallback, re-pin with a TODOS entry; budgets and retries are unchanged. A rerun on opus-4-7 passed (125 s, 5/5 detected). * test(evals): add E2E_KINDS, BEHAVIOR_WHY, EVAL_POLICY and CASE_QUARANTINE skeletons Every E2E_TIERS and LLM_JUDGE_TOUCHFILES key starts as 'rule'; BEHAVIOR_WHY and CASE_QUARANTINE start empty. EVAL_POLICY pre-registers the approved panel (3, majority 2), quarantine entry 0.95/10 and exit 0.97/10, 10% cap, 8-weekly-run expiry, Fisher drift alarm and one INFRA re-dispatch. * test(evals): add trial records, panelVerdict, expectContract and trial-outcomes JSONL EvalTestEntry gains case_id, kind, trial, panel, failure_class and policy_version, stamped from the runner's TRIAL_ENV on isolated trial shards. panelVerdict() is the single verdict function (INCOMPLETE on missing or duplicate trials, contract veto at any count, quarantine hard-break rule, INFRA/INCOMPLETE machine classification). expectContract() records failure_class 'contract' on the collector entry and a sidecar before throwing. trial-outcomes JSONL has a fail-closed writer and a data-only reader. * test(evals): pin the fail-closed rule-shard gate through the real --report path Synthetic slice artifacts for rule fail, timeout, missing slice, unreported entry, hollow, never-started, collector failure and wrong-slice reports all exit red before the panel-verdict gate change lands. * test(evals): retire every paid automatic retry Paid evals never retry (approved 2026-09-29): delete SHORT_CASE_RETRY_FILES and retriesWithinCaseCap, drop the retry fields from the registered wall rows (walls now cover one run plus reserve), make retriesForFiles return 0, pass --retry 0 explicitly, and drop --retry 1 from the package.json paid scripts. Add the eval:pass-rates alias. Tests that pinned the old retry allowance are updated as a policy change; review-finalization-budget now proves late-result recording under the production zero-retry arguments. * test(llm-judge): sample every judge as a pre-registered 3-sample panel Each of the 24 skill-llm-eval judges now draws EVAL_POLICY.judge.samples independent samples of the same prompt concurrently inside the unchanged JUDGE_MS budget. Numeric dimensions gate on the per-dimension panel mean against the unchanged threshold; booleans (would_browse, consistent) on a strict majority. An erroring sample fails the whole panel and is never resampled; a refusal is an unscored panel only when every sample refused. callJudge's 429 backoff stays: it is transport before any model output. The workflow-judge cache stores and validates only complete panels, and its identity now records the panel and zero file retries. Harness tests that pinned one provider call per case now pin the panel size. * test(evals): classify every live case and re-select a case when its kind changes E2E_KINDS: rule by default (191 E2E ids), 22 behavior cases whose verdict is a live model choice with an acceptable sub-100% per-trial rate, each with a BEHAVIOR_WHY tolerance, and 25 judge entries (the 24 workflow judges plus the fixed-fixture llm-judge-recommendation rubric check). Contract-shaped cases (ask-before-decide, plan-mode no-writes, mandated steps, secrets, the batching floor) stay rule. Behavior requires a known literal registration and an exact Bun test name so the case runs as its own trial shard. Map-diff selection now diffs E2E_KINDS and BEHAVIOR_WHY per key, and a base revision without them selects every key, so a kind flip runs the panel it introduces. test/eval-kinds.test.ts enforces coverage, tolerances, isolatability and the reviewed counts, printing the literal to add. * feat(evals): per-case pass rates with Wilson intervals, identity series and quarantine policy scripts/eval-flake-rank.ts becomes eval:pass-rates (eval:flake-rank stays an alias, and the legacy aggregate stays exported). It reads eval-store's trial-outcomes JSONL from the last N completed evals-periodic runs on this branch and main (gh, downloading only the trial-outcomes artifact, cached and size-capped, parsed as data), plus local eval dirs, and prints per-case per-trial pass rates with 95% Wilson intervals. A series is a case's own touchfiles minus GLOBAL_TOUCHFILES (caseSeriesIdentities, for the report job to stamp), per model, CLI version and policy version. Labels: INCONCLUSIVE, BROKEN, FLAKY, FAILING, PASSING. --backfill imports legacy slice artifacts as pre-policy trials (first attempt only, attributed by registry id, never guessed) for display only. --gate fails with ACTION REQUIRED on post-policy evidence only: drift below the quarantine entry rule, a rule case behaving like behavior, a one-sided Fisher drop against the previous identity (Holm-controlled), and quarantine entries that met their exit rule, expired after 8 weekly runs, broke the 10% tier cap or are invalid. CASE_QUARANTINE entries now carry a failureClass (detector, harness or model-latency); a product defect has no class and is never quarantined. The policy test pins EVAL_POLICY's approved constants. * feat(eval-pass-rates): attribute legacy records by the exact slug of their display name * ci(image): pin Claude Code 2.1.284 so the eval model is recognized 2.1.251 logs [claude-code:unrecognized_model] for claude-fable-5-1, the eval capture/judge default. 2.1.284 does not. The gate PTY smoke subset (plan-ceo/plan-devex plan-mode, plan-mode-no-op) parses on the new TUI; plan-design-review-plan-mode passed at 293 s on 2.1.284 and timed out at 300 s on 2.1.251 on the same tree. * test(eng-batching): grade the floor once the review report is complete A completed GSTACK REVIEW REPORT ends the review, so the review-question count is final there. Run 36606688266 wrote its report at 1,248 s and closed the session at 1,318 s; the case now stops collection and applies the unchanged floor at the report instead of waiting out the session. No budget changes. * test(eng-batching): bind unsourced native briefs through the report's target Run 36606688266 asked ten separate native review questions (D1-D9 bound to ledger records R1-R9) and failed reviewCount=0 < FLOOR=3: its briefs named the plan by title instead of citing PLAN.md, its report declared 'Review target (fixed): PLAN.md' under '# Engineering review: <plan>', and it kept an unfenced copy of the plan's own H1. The named-source route now accepts those spellings and non-inline ledger briefs. The same replay rejects a foreign, mixed, duplicate or missing target, another plan's title or copied H1, a brief naming another plan or file, a mismatched saved brief, and re-asks. The run-36597762183 capture still counts 3. * fix(plan-design-review): treat a designer with no API key as unavailable Both proof runs (36597762183, 36606688266) printed DESIGN_READY, hit 'No OpenAI API key found' on the first $D variants call, then hand-built HTML/CSS wireframes, screenshots and a comparison board for ~195-245 s before the first review question; the second run timed out at 600 s. A failed first generation now takes the existing text-only path, and the skill forbids substituting hand-built mockups. * fix(deslop-shared-libs): read related sources together within the turn limit Run 36606688266's opportunity audit read sixteen sources one per turn and stopped at error_max_turns; the passing run 36597762183 read the same files in three batched commands. The skill now says turns are bounded and asks for parallel reads or one read-only command per step. * test(ceo-mode-routing): submit a mode review that scrolled past the viewport Run 36606688266 bundled routing, learnings and the mode choice into one native call. Its review panel was taller than the terminal, so the tab bar scrolled off, ceoModeSubmissionInput returned null for 240 s and HOLD SCOPE was never submitted ('no posture match'). With no bar on screen the viewport must still end at the focused Submit prompt, and the accumulated screen text supplies the one complete panel, authenticated exactly as before. Replay controls reject another mode, an unoffered answer, an altered question, a quoted panel, trailing output, a moved cursor and an answered or changed call. * docs(evals): document the pre-registered verdict policy, quarantine, pass-rate history and arithmetic AGENTS.md replaces the retry rule with the approved policy text (no retries; kind fixes trials; no added trials, samples or dispatches after a result; quarantine by CASE_QUARANTINE only; one INFRA/INCOMPLETE re-dispatch) and notes that a pre-registered fixed panel is not rejudging. CONTRIBUTING gains the kind rules, the judge panel, eval:pass-rates and an 'Add a paid eval' checklist. TESTING_INTERNALS describes verdicts, quarantine, history and the arithmetic, including the rule term: 1 trial vs 2-of-3 red rates at p = 0.99/0.95/0.90/0.70/0.30 and lane all-green probabilities for the current 191 rule / 22 behavior / 25 judge registry. * feat(evals): trial planner, slice exit split and panel-verdict report Planner: behavior and quarantined cases become panels of isolated trial shards (<file>#<id>~t<N>) bound by EVALS_SELECTION_JSON=[id] and the exact test name; the file shard excludes them by name. Trials of one case never share a slice, result slugs are unique, panels are validated whole, unknown registrations throw, and the planner prints a capacity preflight. Executor: each trial shard gets its TRIAL_ENV identity and a trial record (outcome, failure class, cause, cost); every shard writes a JUnit report. The slice exit now means execution completeness: a failed rule shard or a trial without a record reds the runner, a failed trial does not. Report: panelVerdict() decides every panel of the first run attempt (later attempts are reported, never replacing it); rule shards keep the unchanged fail-closed checks; collector records all count (no last-attempt wins); census runs enforce the quarantine cap and expiry. It writes collector-outcomes v2, trial-outcomes.jsonl (trials plus JUnit rule/judge cases), report-summary.md, and one headline + failure block with rerun commands, and flags INFRA/INCOMPLETE-only reds for the one re-dispatch. The fail-open suite gains the panel cases: behavior 1/3 red, 2/3 green with its failed trial shown, missing trial INCOMPLETE, contract at 2/3 red, quarantined 1/3 green, 0/3 and contract red, missing slice red, and a later attempt never replacing the first. * chore(evals): refresh paid duration seeds from proof runs 36597762183 and 36606688266 Both tiers, merged in run order (the later run wins). Notable: split-overflow 1332s -> 504s, section-loading 604s -> 342s, mode-routing 575s -> 444s; multi-finding-batching 734s -> 1318s (its red path in run 36606688266). * feat(evals): stamp trial series identities and fit panels to the live registry - scripts/eval-trial-series.ts stamps series_identity (eval-flake-rank's caseSeriesIdentities) on a report's trial-outcomes JSONL as its own step, keeping the history tool out of the paid runner's closure; TrialOutcomeRecord gains the optional series_identity field. - Slice-count plans let a registered trial spill into an ordinary lane when its siblings hold every long lane, so panels never share a runner. - Re-audited test-selection.ts (Stream B added the E2E_KINDS/BEHAVIOR_WHY map-diff; no new module loading) and repinned its hash. - Detach and release floors now count trial shards (66 periodic trials in 22 panels): periodic floor 33,821s, still under eval:bg:periodic's 67,380s. - Coordination fixtures supply the executor's trial records. * ci(evals): attempt-scoped artifacts, verdict-v2 PR comment, weekly pass-rate gate and one INFRA re-dispatch - Slice, census and marathon artifacts carry -a<run_attempt>; reports download them per artifact (no merge), so records never overwrite and a re-run never replaces the first attempt's verdict. - Planners pass --max-parallel for the capacity preflight (24/16 unchanged: the refreshed periodic plan needs 24 slices, the gate census 12). - PR comment: jq-only job reads collector-outcomes v2 (headline, sanitized failure block); the group_by(.name)|last recomputation is gone. - Reports stamp series identities, upload trial-outcomes-* for history, and shard logs upload always (a failed trial no longer reds its runner). - Weekly report: headline + failure block of both lanes in the issue body, the eval:pass-rates --gate step (fails closed without history), close the issue on a green run, and UC-E1: when every red is machine-classified INFRA/INCOMPLETE, one re-dispatch as a new run in its own concurrency group (redispatch_of), both runs reported. * feat(evals): planner-side whole-panel reuse and negative receipts The planner job restores this PR's receipt store once and ships a single filtered set with the plan: a pass or panel receipt with a same-or-newer FAIL for its input identity is dropped, and a panel receipt ships only as a whole PASS panel (re-verified with panelVerdict) from one run. Executors read only that set (no per-slice cache restore or save), so every trial of a panel sees the same receipts; a trial reuses its own record from the panel receipt, keeping a split PASS's failed trial. Trial identities drop the trial index (run-scoped) and bind the panel policy. Executed shards carry their input identity; the report turns a whole fresh PASS panel into a panel receipt and a FAIL panel or failed rule shard into a negative receipt, and marks a panel that mixes reused and fresh trials INCOMPLETE. The report job merges plan, slice and report receipts (newest per file) and saves one store per run. Also fixes two TS2352 casts in browse/test/dia-macos-qualification.test.ts whose diagnostic text drifted with program order (baseline locked, fix only). * feat(evals): --case/--trials local diagnosis and panels in local sharded runs bun run scripts/test-paid-shards.ts --case <id> [--trials N] runs N independent trials of one case through the CI panel runner (trial shards, TRIAL_ENV identity, name-pattern isolation) and prints its panelVerdict(); N defaults to the case's policy panel and CI never reads it. The local sharded path (test:gate:sharded, test:periodic:sharded) now plans the same trial shards and exclusions as CI and exits on execution completeness plus panel verdicts. * test(pty): grant an owned Create pane whose title row is cropped The targeted batching rerun on Claude Code 2.1.284 left its first report Write unanswered for 1,372 s and timed out: the viewport began at the pane's relative file row and rule, with the 'Create file' title cropped above, so the preview parser rejected the file row as foreign. That row must now resolve to the owned path and is skipped before the unchanged line-by-line preview match. Replay controls reject another file, another directory and an edited preview row. * fix(evals): tsx-safe generics in eval-flake-rank, legacy artifact names, no-retry wall docs * test(evals): record the read-only and detector-row invariants as contracts shared-libs-opportunity-judgment and review-design-lite are behavior cases: their recommendation and checklist judgments may vary, but the read-only invariant (commands, provider requests, fixture bytes, hooks, state) and the deterministic fake-engine detector rows are contracts. Both now go through expectContract, so any failure vetoes the panel. * test(judges): sample the recommendation rubric as a panel; never re-ask armJudge llm-judge-recommendation is a judge case: each fixture now draws a 3-sample judgePanel, gates reason_substance on the panel mean and the present/commits/has_because checks on a 2-of-3 majority, thresholds unchanged. armJudge no longer re-asks on a malformed verdict; it is a failed sample, as the judge policy requires. * test(evals): record a pre-turn API or CLI failure as infra recordE2E sets failure_class 'infra' on a failed session whose runner reports error_api, timeout_startup, error_output_stream or a non-zero CLI exit with zero turns and no assistant event. A model refusal, a timeout after model work, max turns, or an explicit caller pass/class keeps its ordinary classification. * test: pin every-record outcome counts and the twelve doc-sync callbacks * test(eng-batching): read the report target as a field, not a spelling The next targeted rerun (Claude Code 2.1.284) again asked eleven separate native questions and again counted zero: its briefs named no plan and its report declared '- **Review target (fixed):** `/abs/PLAN.md`' under '# Eng Review — PLAN.md: <plan>'. An unsourced brief now inherits the one current target field that names a PLAN.md file, whatever its list or emphasis markup; its ledger record still supplies the cited finding and must reproduce the brief exactly. A brief that names its plan must still match the report title. Replays of all three captures count 9, 9 and 3; controls reject a foreign, duplicate or missing target and an archived title. * fix(evals): --case list mode and name precheck; case-shard qa-callers; refresh batching and design-with-ui seeds * chore(release): v1.91.9.0 * test: settle the post-response composer before seeding; give the TPA recorder adapter its infra helper submitPlanSeed accepted a stale empty composer when the transcript recorded end_turn before the CLI repainted (late-repaint-typed-current fails 5/5 on the old helper, passes 5/5 now). The TPA recording fixture extracted recordE2E without isPreTurnInfraFailure, so every failed case threw before recording. * test(autoplan-dual-voice): unwrap Claude Code 2.1.284 subagent hand-back frames; accept read-only probe diagnostics; record before asserting Census run 36626737820: the native CEO report arrived framed and indented, so its INPUT line never matched, and the model's exact probe plus two variable echoes was not canonical. A column-zero line inside a frame, command substitution, backticks, redirects, assignments, CODEX_MODE echoes and output line-count mismatches stay rejected. The failure now records before asserting. * ci(image): keep Claude Code 2.1.251; test(ceo-mode-routing): keep HOLD's own deferrals in scope before assessing its rigor decision 2.1.284 enables per-turn effort for claude-fable-5-1: in gate census 36626737820, 66 of 84 sessions ran longer than on 2.1.251 (+20% session time, +32% thinking tokens) and 11 cases timed out on unchanged budgets. HOLD SCOPE's 0G step asks its own defer/keep menu; the actor answered it Defer and the assessment then judged that scope question as the rigor decision. The actor now answers that menu Keep and assesses the next one. * test: attribute quoted prior-record field lists, state the judge reason bound in its schema, move split-overflow to marathon Census 36629958451 reds: - outside-plan-disabled-no-fallback: the model quoted the pre-existing record as a parenthesized field list with its exact timestamp; attribution now requires that exact timestamp and the record's own field values. - plan-devex-peer-comparison-classification: the judge correctly returned missing but wrote a 1069-character reason, voiding the judgment; structured outputs cannot enforce maxLength, so the bound is stated on the field. - plan-ceo-split-overflow ran 504-1188 s as one PTY flow and set the periodic lane's wall clock; it now runs weekly in the marathon lane. * test: supply holdDeferKeepIndex to the CEO routing mocks and follow split-overflow into the marathon lane The registered-callback fixtures mock ceo-mode-option and lacked the new export; the split fixtures asserted the periodic tier; the registered-budget check looked for split-overflow only in the periodic manifest. * fix(qa): checkpoint receipts print the report link for their exploration file qa-functional-webhook-report failed in two of three censuses because the report linked .qa-evidence/NNN capture folders as "checkpoints" and never linked exploration-NNN.json. The checkpoint receipt now prints link: [checkpoint NNN](exploration-NNN.json), and the functional report template says capture folders are not checkpoints. * docs: final census numbers in the v1.91.9.0 entry; file the paid-eval follow-ups * ci(evals): name the PR-comment loop's unused fields so shellcheck passes (SC2034) * fix(plan-ceo-review): tighten expansion pacing wording to fit the skeleton cap after the main merge The merged skeleton measured 80,166 bytes against its unchanged 80,150 cap. Same instructions: ask separately for each addition, in turn, with no pacing menu; lead each proposal with the felt experience, then shape, effort and impact. * fix(eval-pass-rates): match trial-outcome files by basename so Windows backslash paths are read * fix(evals): repair proof-run reds in design-consultation, document-release, design and QA fixtures - design-consultation Phase 1 asks one brief that confirms context and decides research; the confirm-only first question scored substance 2. - document-release defines ship-owned inputs, exact steps and the JSON result, and drops stale spawned-from-/ship text (judge actionability 3.67 -> 4/4/4). - plan-design-with-ui accepts the Step 0D focus menu the same way the shared picker does ("focus on specific ones?"). - plan-design-review plan-mode saves in three Edits instead of one final Write. - QA functional annotations ask for the full 40-character revision. - Outside-disabled attribution judges quoted prior-record data by its exact timestamp or a dated, pre-existing-record sentence; four captured phrasings replay clean and current claims still fail. - --case can select autoplan-dual-voice by its literal test name. * test(design): revert the three-Edit plan-mode flow A focused paid run still timed out at 300 s: the first three passes alone took 150 s of thinking. The case stays a named timeout red rather than cutting review depth. * test: accept 'review mode = X' auto-decide declarations and parenthetical scope exclusions in the shared-libs actor auto-decide-preserved: the product auto-decided HOLD SCOPE and said "Decision: review mode = HOLD SCOPE"; the grammar knew only "is" and ":". shared-libs-plan-callers: the recommended option said "(no hardening)" and the actor read "hardening" as an expansion. Both replay the captured text, keep negative controls, and passed focused paid runs. * fix(review): pass Review Army checklists by path, run research alongside dispatch, always probe the design detector; state review-log invocation and statuses in the caller fixture - review-army-perf-n-plus-one: the parent copied full checklists into agent prompts and ran web research before dispatch (290 s on a 12-line diff); 212 s now. - review-design-lite: 5 of 6 captured trials reported the detector absent without probing; the probe is mandatory and its first line is reported, and the contract credits only fake-engine rule ids the checklist never names. - review-exploratory-small-cli: the fixture never gave review-log's direct invocation or status vocabulary; the model ran it through bun and wrote status "blocked". The prompt states both and the validator rejects out-of-vocabulary review statuses. Each case passed a focused paid run after repair. * docs(changelog): proof-run product fixes * fix(ship): always run the design-lite detector probe; test(shared-libs): credit a failed first file view and deferred-reuse Skip wording - /ship design-lite: the probe is mandatory and any non-ready first line is stated, matching /review (5 of 6 captured /review trials had skipped it). - shared-libs-pr-coverage: the first PR 42 page-1 read printed only a jq error, so the one refetch is a legitimate recovery, charged to the same budget. - shared-libs-review-prior-coverage: the Skip option said a future review can "reuse it once snapshot coverage holds"; a conditional tail on the recorded decision is not product work. Captured-text regressions and negative controls. * fix(ship,qa,document-release): repair proof-run regressions and fixture gaps - ship-docsync-completion: yesterday's audit-scope result dropped the section's status, so /ship spliced one in; the section now opens with **Status:**. - ship-docsync-missing-asset: a missing section or old Ship-owned mode blocks before launch. - ship-docsync-late-result: the invocation record says prepare already saves the candidate selection (no extra Read; budget unchanged). - qa exploratory: await the method Reads before the first probe. - qa-callers fixture: quote the real review-log record template; allow the git log command plan-completion prescribes. - qa functional observer: a receipt caught mid-link(2) is checked at stop instead of failing with ENOENT (reproduced from CI). Each repaired case passed a focused paid run. * ci(image): pin Claude Code 2.1.284, the version users run Request-body capture shows both 2.1.251 and 2.1.284 send effort "high" to claude-fable-5-1; 2.1.284 adds the model's own profile. The slower 2.1.284 census was mostly API latency: its SDK-only judges were 25% slower too. Nine previously slow cases pass on 2.1.284 within unchanged budgets. * test: one owner per case id, a structural devex 0B setup rule, and correct design/gbrain actors - plan-design-review-plan-mode was registered by two files; the PTY smoke is now plan-design-review-plan-mode-smoke, and a registry test requires one owner per case in case-sharded files. - plan-devex-finding-floor: the template's 0B narrative-confirmation question is classified as setup structurally instead of timing out a Haiku assessor. - setup-gbrain-remote: the actor accepted 'skip' on the MCP-registration question the test asserts; it now accepts that question and declines others. - design-review-plugin-handoff: the fake engine cited a file absent from the fixture repo and index.html linked a missing styles.css. Captured-question regressions with negative controls; each case passed a focused paid run. * test: PTY harness handles clipped reviews and bundled setup tabs; AUQ judge uses structured output; design-consultation carve declines optional outside voices - ceo mode routing: a Submit review taller than the viewport, a setup tab bundled after the mode tab, and a clip through the mode question each hung or misread the run; the native answer is still verified after Submit. - judgeRecommendation requests a 1-5 enum schema; a malformed Haiku reply had scored substance 0 for a 4/5 brief. Judge failures now propagate. - carve section-loading for design-consultation declines the optional outside voices (a supported path) and treats DESIGN.md as the report; timeout unchanged. The Step 0E handoff defect is not fixed (0/15 samples across four wordings, none shipped) and is filed in TODOS. * test: fold the design-consultation completion replay into carve-section-sharding (test-of-test ratchet) * docs(todos): record the pre-push hook shard-order hang * test(qa-callers): disable git auto maintenance in the fixture repo (same guard as shared-libs; from #3002) * test(office-hours-attempt): the fake judge SDK response carries stop_reason like the real API (structured judge requires end_turn) * fix(qa): the caller STOP line says to await the method Reads before any probe ship-exploratory-plan-checks: the model read exploratory.md and sent a capture in the same response, before seeing the section's own await rule. * fix(qa): number the qa value-bar questions from 1 and say reproduced bugs already answer the first two * fix(qa): define evidence.json where it is built, point the preparation gate at the next section, name measured command durations in the report template Recurring qa/qa-only workflow-judge complaints in CI (clarity/actionability 3.33). * fix(plan-eng-review,review): a disallowed question tool is not headless; report kept tests only when some were skipped * fix(plan-eng-review): keep the headless-rule contract phrases adjacent * fix(evals): cut path variance at its measured sources - gstack-qa-evidence capture prints startedAt/completedAt/durationMs and, for --deadline captures, remainingMs; the functional report takes durations from them. The section clock notice asks for one clock read up front instead of one after every checkpoint (QA runs spent 7-14% of tool calls on date -u). - ship plan-completion: skip the audit dispatch when discovery already found no plan (the dispatch-vs-skip conflict produced an optional 60-100 s subagent). - materialize/checkpoint validation errors state the expected schema, so a rejected annotations file is fixable in one call instead of blocking the phase. - session-runner counts turns from the transcript when a run times out, so timeouts stop reporting 'turn 0'. * fix(evals): count timeout turns only from object transcript events * test(qa-callers): deterministic child transport, completion-time handoff reads, compact phase report The exploratory caller cases exist to prove the caller starts and bounds exploratory QA. Their native adversarial reviewer (review) and plan audit (ship plan-checks) now come from recorded child outputs instead of a live subagent, handoff freshness reads are required before completion records rather than every bookkeeping log, and the phase report is compact. Measured: 194-257 s per case against 208-284 s before, no subagent calls. * test(ship-docsync): seed fault cases at their gate instead of replaying attempt 1 The post-dispatch fault cases (missing-marker, launch-failure, timeout-unsettled, late-result, stale-before, stale-after, recovery) now start from a fixture-owned attempt 1: the real actor prepares and dispatches it, its verbatim output is saved once, and the invocation journal carries its pre-dispatch entry with the child asset hashes. The model resumes at Parent processing with a trimmed read list, inspect named as the authoritative repository observation, and recovery's intermediate checkpoint folded into the next attempt's pre-dispatch entry. Assertions count only parent-issued transport events and require a read of the saved attempt-1 output; missing-asset and the legacy failure case keep the full model-driven first attempt, and their prompts are byte-identical. * test(ship-docsync): name the seeded read list and cap journal/report length The first seeded stale-before run spent calls locating documentation.md (two ls sweeps), reading through cat and re-Reading the record before Edit, and ~40 s composing 1.5-2.2 KB entries and report. Name every seeded read path, ask for native Read, and bound entry/report length. * test(ship-docsync): trim the seeded parent's measured model time Measured on the seeded runs: one read the 78 KB ship/SKILL.md, the post-child freshness comparison spent 18-32 s of thinking over full inspect contents, and the final response restated the report (~1.1 KB). Say the phase excerpt stands in for ship/SKILL.md, compare hashes first and read content only for changed paths, and end with one status line. * feat(qa-evidence): enforce the checkpoint sequence and fill report bookkeeping in code - capture refuses to run another probe until a checkpoint anchored on the latest complete capture names this capture as its next command, and every complete capture prints that requirement. - materialize fills revision, runtime, cwd and learning (checkpoints whose next native command differs) when omitted and prints the reportLinks the report must include; the QA section shrinks accordingly. * test(qa-callers): hand the caller phase its invocation-start observations and review token; fix(next-version): fetch without auto maintenance - Every caller case receives the diff, status, log, untracked list, HEAD and an already-captured review start token, so the phase spends its budget on the contract under test instead of re-running setup reads. - gstack-next-version's fetches pass --no-auto-maintenance. On git 2.55 a completed fetch forks detached maintenance in the caller's repository; the free suite's live smoke test ran it inside the CI checkout, and every shard-12 pre-push hook hang so far followed a completed smoke fetch. * feat(deslop-shared-libs): route every Git read through bin/gstack-safe-git The skill made the model retype a long safe-Git prefix on each call and a dropped flag failed shared-libs-read-only. bin/gstack-safe-git applies the fixed env + flag prefix, adds --no-ext-diff --no-textconv to log/show/diff, allows diff only between two explicit object IDs and ls-files only in the NUL-delimited overlay form, and refuses every other shape with one line naming the allowed forms. The template now points at the installed helper (host global runtime via {{SAFE_GIT}}) and drops the prose it enforces. Fixtures resolve the helper to this checkout, the git shim records the safety environment, and isGuardedGitRequest requires the complete prefix (env included) for every repository read. * test(shared-libs): tee to a discard device is not a file write Paid shared-libs-opportunity-judgment t1 on 1213b01 failed read-only on '... | tee /dev/null | sha256sum'. The detector flagged any tee operand while the same devices are allowed for redirection. tee now fails only when an operand is a real file; tee to a file, -a file and -- -a stay violations. * fix(qa-evidence,observer): reject placeholder metadata and replay-only learning; declare the docs atomic-write target - materialize measures revision, runtime and cwd itself and rejects supplied values that differ (CI run wrote revision "HEAD" and runtime "bun"), and refuses learning checkpoints that replay the same probe, naming the fix. - The docs write observer treats Claude Code's atomic temp for the authorized doc target as transient, so a temp renamed before its per-file watch no longer marks the observation incomplete (ship-docsync-completion flake). Per-file monitoring outside declared targets stays fail-closed. * test(qa-functional): fix mode requires only the happy scenario from the model (carried byte-identical from #3002 183b01f4..3e6074b4) verifyQANativeRegression already reruns all eight webhook scenarios on the repaired source, so the model-side eight-scenario requirement in fix mode duplicated harness coverage and pushed qa-functional-webhook-fix past its budget. qa-only still requires every scenario. * fix(deslop-shared-libs): probe the audited repository with -C <repo> A CI run probed safe-git from the session directory above the target repo, so the capability probe never touched the repository and the run fell back to the API without a local attempt. The probe (and any call from elsewhere) now names the audited repository. * test(qa-deadline): never attach a reader to the full-pipe fixture's stdout The full-pipe receipt test attached a 'data' listener (flowing mode) and then paused; on CI the reader could drain the 2 MB write before the pause, so the receipt write never blocked and the helper exited 0 in ~126 ms. The stdout pipe now stays unread until the assertion, which is what the test means to model. * feat(qa): helpers answer --help, and the QA eval interfaces declare it Approved by Garry: asking gstack-qa-evidence or gstack-qa-deadline for usage is read-only, so both helpers print usage and exit 0 on --help (the evidence usage now names the annotation shape), and the functional and caller command allowlists accept exactly 'bun <path>/bin/gstack-qa-{evidence,deadline} --help'. Two CI runs failed only on that call. * fix(qa): after an input change, a probe is affected unless shown otherwise CI late-input run finished in time but revalidated only the happy probe after the locale input changed and reported the stale adverse probe green. The revalidation step now treats any probe not shown to be unaffected as affected. * test(shared-libs): seed the lifecycle replay's first Step 3 pass instead of replaying it shared-libs-review-lifecycle ran ~88% of its 300 s session budget (12-run census median 265 s, 4/24 sessions timed out). The fixture now executes pass 1's Step 3 once with the real logger and Git: a real unused REVIEW_START, then the diff, inventories, attributes/config/index flags, gstack-review-read output and every file's bytes and sha256, saved to one observation. The model resumes at Step 4 with an exact four-file first read, the observation named as the authoritative pass-1 repository read, one post-fix verification, an explicit pass-2 read list and a twelve-line summary. Pass 2 still runs its own --start, diff, reads, fingerprint and stage actor before --finish. The actor scope now states that a current settled final-pass actor result supplies the replaced QA/adversarial prerequisites and that the no-credit disclosure is a reporting label: one r1 session persisted completed:false from that ambiguity. New assertions: the final binding never uses the seeded token's start or tree, and the observation was read; free controls finish the seeded token (binding changed) and omit the observation read, and both fail. * test(shared-libs): trim the resumed review replays' setup and report Every sibling review session (revalidation, path-eligibility, index-flags, prior-coverage) loaded qa/sections/exploratory.md and often scope.md although its QA and native adversarial results are supplied synthetic inputs, then spent a second request on shared-code-reuse.md and base metadata. The resumed scope now states that the supplied results replace Step 4's QA method loading; the revalidation contract names one first response (workflow, checklist, finding, prerequisites, shared-code-reuse.md, base metadata) and caps the summary at twelve lines. Receipt order, direct source reads, the checker, the question and final persistence are unchanged. * fix(review): define what a Step 5c Skip option says Step 5c named "B) Skip" without saying what its description may claim. Two CI captures (path-eligibility on |
||
|
Garry TanandOpenAI Codex
|
06ed920a97 |
v1.89.0.0 feat: add shared-code extraction audit (#2925)
* feat: bind shared-code review advice to source and branch * feat: add shared-code extraction audit and scoped review checks * test: recognize complete source reads and explicit coverage legends * chore: bump version and changelog (v1.88.0.0) Co-Authored-By: OpenAI Codex <noreply@openai.com> * test: capture native review questions and retain public evidence Capture the actual first public native question with strict ownership and display matching. Preserve terminal failures and raw evidence, and retain SDK completion checks. * test: recognize verified review evidence and complete fixtures Recognize complete source and diagram evidence, concrete design and developer-experience decisions, and the complete planted scenario contracts. Preserve negative controls and grading thresholds. * fix: preserve decision brief structure in native questions Keep the required pros-and-cons heading and final Net field in native question text. Regenerate host outputs and document the release and evaluation repairs. Co-Authored-By: OpenAI Codex <noreply@openai.com> * docs: update project documentation for v1.88.0.0 Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix: correct eval retry accounting and ship workflow gates * fix: capture native eval evidence and stabilize CI fixtures * fix: keep shared-code eval skips read-only Choose explicit no-change answers instead of mixed fix/preservation options. Reuse the bounded revalidation prompt for path fixtures so required review metadata is available without repeated discovery. Preserve source checks, retry limits, and failed native terminal outcomes. Add captured-question and callback regressions, plus evaluation selection coverage for the affected fixtures. --------- Co-authored-by: OpenAI Codex <noreply@openai.com> |