mirror of
https://github.com/garrytan/gstack.git
synced 2026-10-07 11:56:59 +02:00
7fca42ad8b6c707b8a38f579f72bf3c4f7de6d85
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 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 TanandClaude Opus 4.7
|
ce5fbfa99f |
v1.52.0.0 feat(plan-tune): explicit consent + first-run setup wizard for contributors (#1741)
* feat(plan-tune): explicit-consent surface + setup gate for question_tuning Step 0 grows two implicit gates that run before user-intent routing: - Consent gate: question_tuning=false + no marker → offer opt-in (contributor-specific copy variant) - Setup gate: question_tuning=true + declared empty + no marker → run 5-Q wizard Markers (~/.gstack/.question-tuning-prompted, ~/.gstack/.declared-setup-prompted) ensure each user is asked at most once. The Enable+setup section split into "Consent + opt-in" (with contributor framing) and standalone "5-Q setup" reachable from both the consent flow and the setup gate. Also aligns the calibration gate across three docs (V0 said 90+ days, TODOS said 2+ weeks, binary uses 7 days). The fix distinguishes: - Display gate (sample_size>=20, skills>=3, question_ids>=8, days_span>=7): for rendering inferred values in /plan-tune output - Promotion gate (90+ days stable across 3+ skills): for shipping E1 behavior-adapting defaults TODOS.md E1 card updated to reference 90+ days, plus Codex's substrate risk note: generated skill prose is agent-compliance-based, so E1 ships as advisory annotations on AskUserQuestion recommendations, not silent AUTO_DECIDE. Tests can verify templates contain right reads but can't prove agents obey them. Per /plan-eng-review + Codex outside-voice 2026-05-26. Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * chore: bump version and changelog (v1.49.0.0) Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com> * feat(bins): honor GSTACK_STATE_ROOT override for test isolation Plan-tune cathedral T1 (per D16 / Codex outside voice). The 3 bins that back /plan-tune (question-log, question-preference, developer-profile) previously ignored GSTACK_STATE_ROOT, so tests that tried to point state at a tempdir via that env var silently wrote to the real ~/.gstack. Make STATE_ROOT take precedence over GSTACK_HOME so the cathedral's E2E + unit tests can isolate cleanly without sledgehammering HOME. Order of precedence: GSTACK_STATE_ROOT > GSTACK_HOME > $HOME/.gstack Matches the existing gstack-paths emission order. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(plan-tune): regression coverage for v1.49 consent + setup gates Plan-tune cathedral T2 + part of T1 follow-up (Codex IRON RULE — regressions get tests). v1.49 shipped two prose-driven implicit gates inside plan-tune Step 0 (consent, setup) with zero test coverage. The cathedral refactors that template heavily; without tests, silent breakage is possible. Three regression families plus a static template assertion: 1. Consent gate fires under qt=false + no marker; goes silent on marker write or qt=true flip. 2. Setup gate fires under qt=true + empty declared + no marker; goes silent when declared populates, marker is written, or qt is still false. 3. Marker idempotency: gates stay silent across 5 re-invocations after a single decline/bail. Markers honored independently. 4. Static template assertion: gate language can't be silently deleted without breaking a test. Also extends gstack-config to honor GSTACK_STATE_ROOT (it was the last bin still ignoring it — caught while writing the tests; without this, tests would silently mutate the user's real config.yaml). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(spikes): Claude hook mutation + Codex session format Plan-tune cathedral T4 (per D5/D10). Two Phase 1 design spikes that downstream tasks (T3, T5, T6, T8, T9) depend on. claude-code-hook-mutation.md - Confirms PreToolUse allow + updatedInput is supported and is the right mechanism for substituting an auto-decided answer. - Pins stdin/stdout JSON schemas with field-by-field reference. - Documents matcher regex syntax for "(AskUserQuestion|mcp__.*__AskUserQuestion)" so Conductor's MCP-routed AUQ is covered. - Captures parallel-hook merge order caveat and our settings.json snippet. codex-session-format.md - Maps the on-disk ~/.codex/sessions/<date>/rollout-*.jsonl schema by event type (response_item 76%, event_msg 19%, turn_context, session_meta). - Critical finding: Codex has NO AskUserQuestion tool. Gstack AUQ-shaped Decision Briefs surface as agent_message text; answer is the next user_message. Two-tier recovery: marker-first (D18), then pattern fallback for hash-only logging. - Confirms logs_2.sqlite is internal telemetry, not session content. - Lists open questions to answer during T9 implementation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(settings-hook): schema-aware PreToolUse/PostToolUse registration Plan-tune cathedral T3 (per D4 + Codex correction). The previous bin only knew SessionStart and dedup'd on the hardcoded `gstack-session-update` substring. The cathedral needs PreToolUse + PostToolUse hooks registered side-by-side with the user's own hooks, with explicit consent UX, backups, and rollback. New subcommands: - add-event --event <SessionStart|PreToolUse|PostToolUse|...> --command <cmd> --source <tag> [--matcher <re>] [--timeout <s>] - remove-source --source <tag> # removes all entries tagged by source - diff-event ... # preview without mutating - rollback # restore latest backup - list-sources # audit gstack-tagged hooks Multi-source dedup via a new `_gstack_source` field on each hook entry (Claude Code preserves unknown fields). Source tag lets plan-tune-cathedral register PreToolUse + PostToolUse without colliding with the existing SessionStart wiring, and lets remove-source clean up cleanly during gstack-uninstall. Backups written automatically to settings.json.bak.<ts> before any mutation, with a .bak-latest pointer the rollback subcommand reads. Existing legacy `add <cmd>` / `remove <cmd>` shape preserved verbatim so setup --team and gstack-uninstall keep working unchanged. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(hooks): PostToolUse capture hook for AskUserQuestion Plan-tune cathedral T5. Closes the substrate hole that motivated this entire branch: agent-compliance-only logging produced zero events in weeks of dogfood. PostToolUse hook captures every AUQ fire deterministically. What ships: - hosts/claude/hooks/question-log-hook.ts — TS hook that reads Claude Code's hook stdin, walks tool_input.questions[*], extracts user choice + recommended option from tool_response, spawns gstack-question-log per question. - hosts/claude/hooks/question-log-hook — bash shim Claude Code's hook runner invokes; execs bun against the .ts file. - Marker-first question_id extraction (D18 progressive markers): <gstack-qid:foo-bar> stripped from question text, used as the id. Hash fallback hook-<sha1[:10]> for unmarked questions (observed-only, never used as preference key — D18 hash drift mitigation). - (recommended) label parsing for the user_choice/recommended fields, with refuse-on-ambiguous when two labels are present (D2 safety). - Free-text capture: source=auq-other + free_text field when user picks Other and types (Layer 8 dream cycle input). - Matcher covers both native AskUserQuestion and mcp__*__AskUserQuestion (Codex/Conductor catch from outside voice review). - Crash safety: always exits 0; errors land in ~/.gstack/hook-errors.log so the user's session is never blocked by a hook failure. gstack-question-log extended to: - Accept `source` field (default 'agent', new values: hook, auq-other, auto-decided, codex-import-marker, codex-import-pattern). - Accept `tool_use_id` (<=128 chars) for dedup. - Composite dedup on (source, tool_use_id) across the last 100 lines — protects against hook + preamble both firing on the same tool call (D3 belt+suspenders). - Async fire `gstack-developer-profile --derive` after each successful write so inferred.sample_size actually grows (D17 — without this, the cathedral's "before 0, after >0" metric never moves). - GSTACK_QUESTION_LOG_NO_DERIVE=1 escape hatch for tests. 9 new unit tests covering capture, marker extraction, MCP variant, free-text, dedup, ambiguous-recommended safety, crash paths. All pass plus the existing 88 tests across related files. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(hooks): PreToolUse enforcement hook for AskUserQuestion preferences Plan-tune cathedral T6 — the keystone that makes never-ask actually bind. Today preferences are agent-convention (silently ignored). This hook enforces them via Claude Code's hook protocol: when a never-ask preference matches an AUQ that is two-way + has a marker + has a clear recommendation, the hook returns permissionDecision: "deny" with permissionDecisionReason naming the auto-decided option. The agent obeys the rejection feedback and proceeds with the recommended option without re-firing AUQ. Decision tree (per question): - marker absent → defer (D18: hash IDs are observed-only) - one-way door → defer (safety override — never auto-decide one-way) - always-ask preference → defer - no preference set → defer - ambiguous recommendation (two (recommended) labels OR no parseable rec) → defer (D2 refuse-on-ambiguous) - never-ask / ask-only-for-one-way + two-way + clean rec → deny+reason Preference precedence per D8: project-local (~/.gstack/projects/<slug>/question-preferences.json) wins, global (~/.gstack/global-question-preferences.json) is fallback. Why deny+reason instead of allow+updatedInput: AskUserQuestion's updatedInput shape for "pre-resolve this question" isn't structurally pinned in Claude Code docs (T4 spike open question). deny with a reason that names the auto-decided option is the conservative + reliable v1 — the model receives the rejection, reads the recommended option from the reason, proceeds without re-prompting. Swap to allow+updatedInput once the AUQ input shape is verified against real Claude Code. Since deny prevents PostToolUse from firing, this hook logs the auto-decided event itself via gstack-question-log (source=auto-decided) so /plan-tune's Recent auto-decisions surface picks it up. Also writes a session marker ~/.gstack/sessions/<id>/.auto-decided-<tool_use_id> for coordination when the AUQ-shape switch lands. Multi-question AUQ: enforcement is all-or-nothing per call. If any question in the batch isn't eligible (no marker, no preference, ambiguous rec, etc.), the whole call defers so the user still gets to answer the rest normally. Registry lookup: cheap regex extraction from scripts/question-registry.ts (reading + bun-importing the TS file from a hook is too slow). Door type defaults to two-way for unregistered. Matcher covers both native AskUserQuestion and mcp__*__AskUserQuestion (Conductor disables native — Codex outside-voice catch). 15 unit tests cover defer paths, enforcement, one-way safety override, ambiguous-rec refuse, precedence (project wins, global fallback, project-overrides-global), MCP matcher, auto-decided event logging, session marker writing, crash safety. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(scripts): declared-annotation helper + autonomy signal_key wiring Plan-tune cathedral T7. Adds the helper that lets skills inject one-line plain-English annotations on AUQ recommendations based on the user's declared profile — read-only, advisory-only, per TODOS.md E1 substrate-risk guidance (no AUTO_DECIDE off inferred). scripts/declared-annotation.ts - getDeclaredAnnotation(signal_key) → annotation | null - primaryDimensionFor(signal_key) → Dimension | null - Signature uses kebab signal_key per D2/Codex correction (registry uses hyphens; profile dimensions use underscores; helper maps internally). - Bands: >= 0.7 high, <= 0.3 low, else null. Middle band stays silent. - Per-dimension plain-English phrasing: 5 dimensions × 2 bands = 10 phrases. - Reads ~/.gstack/developer-profile.json (honors GSTACK_STATE_ROOT). scripts/psychographic-signals.ts - New signal_key 'decision-autonomy' that maps user_choice → autonomy dimension nudges. This was the missing signal for the 'autonomy' dimension — without it, the cathedral could annotate four of five declared dimensions but autonomy stayed silent. scripts/question-registry.ts - Add signal_key: 'decision-autonomy' to land-and-deploy-merge-confirm and land-and-deploy-rollback. These are the highest-leverage autonomy questions in the surface — "let me decide" vs "go ahead" is exactly what the dimension captures. 13 unit tests cover the helper's full contract (unknown keys, missing profile, middle-band null, both band thresholds, all five dimensions rendering distinct phrases). Existing 47 plan-tune.test.ts tests still pass after the registry + signal-map enrichment. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(setup): install plan-tune cathedral hooks with explicit consent UX Plan-tune cathedral T8. Wires the new PostToolUse capture hook and PreToolUse enforcement hook into ~/.claude/settings.json via the schema-aware gstack-settings-hook (T3) — respecting D4's "never mutate settings.json silently" boundary and the Codex outside-voice warning. Behavior at setup time: - Idempotency: if list-sources already shows 'plan-tune-cathedral', no-op with a one-line note. - Marker present (previously declined): no-op, no re-prompt. - Interactive terminal: print rationale + diff preview from settings-hook, rollback command, and prompt y/N. On accept, register both hooks (PostToolUse and PreToolUse) with --source plan-tune-cathedral. On decline, touch ~/.gstack/.plan-tune-hooks-prompted so we don't re-ask. - Non-interactive (CI / scripted): no prompt; print the two exact commands the user would need to install manually. - --no-team teardown also removes the plan-tune hooks via remove-source. gstack-uninstall extended to clean up plan-tune-cathedral hooks alongside the existing SessionStart cleanup. Listed as a separate "plan-tune cathedral hooks" line in the REMOVED summary when it fires. No new test file — coverage from T3's gstack-settings-hook-schema-aware tests proves the underlying bin behavior; setup-level integration is verified manually (re-running ./setup is cheap and the prompt makes it obvious whether install happened). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(bin): gstack-codex-session-import — structured Codex transcript parser Plan-tune cathedral T9. Backfills question-log.jsonl from Codex sessions since Codex has no AskUserQuestion tool (per docs/spikes/codex-session-format.md) and gstack AUQ-shaped Decision Briefs show up as agent_message prose. Walks ~/.codex/sessions/<date>/rollout-*.jsonl, matches each agent_message that contains either a <gstack-qid:foo-bar> marker or a D-numbered Decision Brief header, then pairs it with the next user_message for the answer. Two-tier recovery per D5: - marker present → source=codex-import-marker, stable question_id - no marker but D-shape detected → source=codex-import-pattern with hash-only question_id (never used as preference key per D18) Subcommands: gstack-codex-session-import # latest session gstack-codex-session-import <file> # explicit path gstack-codex-session-import --since <iso> # all sessions newer than User-choice extraction handles A/B/C letter responses and prose responses that start with the option label. Recommended option parsed via the "(recommended)" label suffix (same convention as Layer 2). Each extracted event written via gstack-question-log, so source tagging, dedup, and async derive all apply uniformly. spawnSync uses the cwd from session_meta so gstack-slug buckets events into the project the user was actually working in, not the importer's cwd. 7 unit tests cover marker path, pattern fallback, multiple briefs in sequence, missing user_message, numeric/letter user response forms, empty-sessions-dir handling. Smoke-tested against a real ~/.codex/sessions/ file from earlier today — returns IMPORTED: 0 because that session was autonomous (no AUQ-shaped prose), proving the bin doesn't false-positive on unrelated agent_message events. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(bin): gstack-distill-free-text — Layer 8 dream cycle distiller Plan-tune cathedral T10. Reads auq-other free-text events from this project's question-log.jsonl, calls Claude via the Anthropic SDK to extract structured proposals (preference candidates, declared-profile nudges, memory nuggets), writes them to distillation-proposals.json for the user to review via /plan-tune (never autonomous — every apply requires explicit Y). Subcommands: gstack-distill-free-text # sync distill gstack-distill-free-text --background # detach + return PID gstack-distill-free-text --dry-run # emit prompt + events, no API call gstack-distill-free-text --status # run history + cost-to-date D7 rate cap: 3 distills per slug per day. Reads ~/.gstack/distill-cost.jsonl for the count, exits with RATE_CAPPED when limit hit. Cost log lines tagged by slug so sibling projects don't share the cap. Yesterday runs don't count. D6 API auth: Anthropic SDK direct, fail-loud on missing ANTHROPIC_API_KEY with explicit message that distill is a separate billing surface from the interactive Claude Code session. Uses claude-haiku-4-5 for cost (~$0.001/ 1k input, $0.005/1k output) — sufficient for structured extraction. D14 execution context: --background spawns detached (nohup) so auto-trigger during /ship doesn't add 30s of pause; results surface on next /plan-tune. Source events get distilled_at:<ts> stamped on them after the run so they don't re-propose on the next distill. Match by ts + question_id. Cost-log line per run includes: slug, proposals_count, rejected_low_confidence, input_tokens, output_tokens, cost_usd_est. /plan-tune stats reads this to show "$X estimated, N runs this month" per Layer 4 surface. 10 unit tests cover --status, rate cap (3/day, yesterday-not-counted, other-slug-not-counted), no-log/no-free-text paths, --dry-run, missing API key, --background spawn. The actual SDK call is exercised by the T16 E2E test (uses real key, ~$0.001 per run). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(bin): gstack-distill-apply — apply distillation proposals with gbrain tag Plan-tune cathedral T11. Bin that applies a single user-approved proposal from distillation-proposals.json to the right surface: - memory-nugget → appended to ~/.gstack/free-text-memory.json (durable local source-of-truth; gbrain is mirror when configured). - preference → routed through gstack-question-preference --write with source=plan-tune (clears the user-origin gate). - declared-nudge → atomic update to developer-profile.json declared dim, small=0.05, medium=0.10, large=0.15, clamped to [0, 1]. Why a separate bin (not inline in the skill template): /plan-tune's apply step needs to be invokable from any host (Claude, Codex, etc) and must write to multiple state files atomically. A bin centralizes the schema + clamp logic; the skill template just calls it after user Y. gbrain coordination: --gbrain-published true marks the nugget so /plan-tune stats can show "12 nuggets, 8 mirrored to gbrain". The skill template invokes mcp__gbrain__put_page / extract_facts / add_tag in the same turn (those are MCP tools, not CLI-callable) before calling this bin. Local file remains canonical so the PreToolUse hook injection path (T12) doesn't depend on gbrain availability. Subcommands: gstack-distill-apply --list # show pending proposals gstack-distill-apply --proposal <N> # apply, file fallback gstack-distill-apply --proposal <N> --gbrain-published true Applied proposals get applied_at + gbrain_published stamped on them so re-running --list shows only unconsumed ones. 11 unit tests cover --list (all three kinds + quotes), memory-nugget append + non-clobber, preference routing through the gate-respecting bin, declared-nudge math (medium=0.10, small=0.05, large=0.15, clamp at [0,1]), proposal mark-applied with gbrain flag, and error paths (bad index, missing --proposal). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(hooks): Layer 8 memory injection via per-session cache Plan-tune cathedral T12. Extends the PreToolUse hook to inject matching free-text-memory.json nuggets into AskUserQuestion responses, giving the agent + user the distilled context from past 'Other' answers right when the related question fires. Per-session cache (D13 perf): first read of free-text-memory.json writes ~/.gstack/sessions/<id>/memory-cache.json. Subsequent hooks on the same session take the cached path. Invalidation is by file-missing: when the canonical file changes (via gstack-distill-apply), the per-session cache either reflects the staler view for the rest of the session or the session restarts and the cache rebuilds. Cheap, correct enough for v1. Matching logic: - Walk this AUQ batch's questions, extract marker question_ids. - Look up signal_key in scripts/question-registry.ts. - Collect nuggets whose applies_to_signal_keys include any of the matched signal_keys. - Cap to 3 most-recent (by applied_at) so the additionalContext stays short. - Surface as additionalContext on the hookSpecificOutput response. Memory + enforcement interact cleanly: the same hook can both surface nuggets AND deny the tool when a never-ask preference matches. Memory context isn't doubled in the deny reason — the auto-decided option name in the deny path is sufficient signal. 6 new tests cover injection on defer, no-match silence, 3-most-recent cap, memory-alongside-deny enforcement, cache file write-through, empty-canonical graceful degradation. Existing 15 preference-hook tests still green. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(plan-tune): SKILL.md surfaces for cathedral T13 Plan-tune cathedral T13. Rewires plan-tune/SKILL.md.tmpl to expose the new cathedral surfaces: Step 0 routing: - Implicit gate #3 (dream-cycle): fires when distillation-proposals.json has unapplied proposals. Marker is per-proposal applied_at so re-firing naturally skips already-handled items. - Added user-intent route for "dream cycle" / "distill" / "what have I been free-texting". - Power-user shortcuts: distill, dream, audit. Stats: - Host-aware source breakdown (SOURCE_HOOK, SOURCE_AGENT, SOURCE_AUTO_DECIDED, SOURCE_CODEX_IMPORT_*, SOURCE_AUQ_OTHER). - MARKED percentage so D18 progressive-markers progress is visible. - Distill cost-to-date via gstack-distill-free-text --status. Recent auto-decisions: - Last 10 source=auto-decided events with question_id + user_choice. Lets the user spot-check enforcement and flip via always-ask. Audit unmarked questions: - Top N hash-only ids by frequency. Surfaces next candidates for the D18 marker retrofit. Dream cycle review + manual distill: - Walks unapplied proposals via AskUserQuestion (one per call), routes accepts through gstack-distill-apply with --gbrain-published flag. Skill template invokes mcp__gbrain__put_page when MCP is available; local file remains source-of-truth. Regenerated SKILL.md via `bun run gen:skill-docs`. All 60 plan-tune tests still green. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(preamble): inject <gstack-qid:...> marker convention into question-tuning resolver Plan-tune cathedral T14. Per D18 progressive markers, the PreToolUse enforcement hook only fires when the AUQ question text contains a <gstack-qid:foo-bar> marker the hook can extract. Without a marker, the hook logs the fire as observed-only and skips enforcement (hash IDs drift with prose so they're never used as preference keys). The high-leverage retrofit point is the preamble's Question Tuning section, not 10 individual skill templates. Updating scripts/resolvers/question-tuning.ts adds the marker convention to every tier-≥2 skill in one change — agents running ANY of the 30+ tier-≥2 skills now embed the marker by default when the question matches a registered question_id. Two convention additions in the preamble: 1. "Embed the question_id as a marker (<gstack-qid:{id}>) somewhere in the rendered question." With explanation that the marker is the only path for the PreToolUse hook to enforce preferences. 2. "Embed the option recommendation via the (recommended) label suffix on exactly one option per AUQ." Documents the D2 parser contract: label first, prose fallback, refuse-on-ambiguous. Net cost: ~700 bytes added to the preamble per generated skill. Plan-review preamble budget ratcheted from 39000 → 40000 (test/gen-skill-docs.test.ts) with a comment explaining the cathedral T14 expansion is load-bearing. Regenerated 42 SKILL.md files via `bun run gen:skill-docs`. The token ceiling warning on ship/SKILL.md (~41K tokens) is pre-existing; this PR doesn't change ship's preamble materially. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(ship): plan-tune discoverability nudge after first successful ship Plan-tune cathedral T15 (the ship-side surface; the setup-side surface shipped in T8 with explicit hook-install consent UX). Adds Step 21 to ship/SKILL.md.tmpl: after Step 20 (persist metrics) succeeds, surface /plan-tune once per machine via a marker-gated single-line nudge. Behavior: - If ~/.gstack/.plan-tune-nudge-shown exists → no-op. - If question_tuning is already true → no-op (user already on board). - Otherwise: print one nudge line, touch marker. The nudge mentions both the observational substrate AND the hook-installed auto-decide enforcement so users know what they get when they opt in. Non-blocking — never asks a question, doesn't gate ship completion. To re-show: rm ~/.gstack/.plan-tune-nudge-shown before next ship. Setup-side discoverability shipped in T8 via the hook install prompt (explicit consent + diff preview + backup). Together these two surfaces cover first-install AND first-ship moments — the user discovers plan-tune organically rather than needing to know /plan-tune exists. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(plan-tune): 5 cathedral E2E scenarios + touchfile registration Plan-tune cathedral T16 (per D12 — all 5 in gate tier). One consolidated file with five describeIfSelected scenarios, each selectable by its own touchfile entry so they only run when the relevant code changes (or EVALS_ALL=1 forces all): plan-tune-hook-capture — PostToolUse hook fires → question-log fills plan-tune-enforcement — never-ask + marker + 2-way → deny+reason + auto-decided event logged plan-tune-annotation — declared profile + memory nugget → additionalContext surfaced on defer plan-tune-codex-import — synthetic JSONL → import bin → log with source=codex-import-marker plan-tune-dream-cycle — apply proposal → re-fire question → memory injected via additionalContext Each scenario fixtures an isolated git repo + bins + scripts + hooks under tmp, then exercises the cathedral chain end-to-end against real on-disk binaries (no mocks at the bin layer). GSTACK_STATE_ROOT keeps the user's real ~/.gstack untouched. These five complement the existing unit tests by proving the full sub-process chain works (not just individual functions in isolation). They DON'T spawn claude -p because the cathedral's substrate behavior is deterministic — agent compliance is no longer the variable. The existing test/skill-e2e-plan-tune.test.ts (plan-tune-inspect) still covers the LLM-driven intent-routing behavior. Cost: each scenario runs in ~1s with $0 because no claude -p invocations. Touchfile-gated, so they only run on PRs that touch cathedral code. Also fixes a bug found by the E2E: question-log-hook didn't pass the incoming tool call's cwd to spawnSync when invoking gstack-question-log, so the bin used the hook process's cwd (the repo root) instead of the session's cwd. Result: log writes landed in the wrong project bucket. Fix mirrors the same cwd-passing pattern from question-preference-hook. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: bump VERSION to 1.50.0.0 + plan-tune cathedral CHANGELOG Plan-tune cathedral T17. Bumps VERSION 1.49.0.0 → 1.50.0.0 (MINOR per CLAUDE.md scale-aware rule: this is substantial new capability — 8 layers, ~3000 LOC, 96 new tests, deterministic substrate + dream-cycle distillation). CHANGELOG entry follows the release-summary format from CLAUDE.md: - Two-line bold headline naming what changed for users (deterministic capture, binding preferences, free-text memory loop) - Lead paragraph: before/after framed concretely (zero events captured → every fire, agent-honored → hook-enforced, declared profile → injected context, regex backfill → structured JSONL parser) - Two tables: metric deltas + layer/where-it-lives. Real numbers (96 tests, ~$0.01 per distill, 3/day cap), no AI vocabulary, no em dashes. - "What this means for solo builders" close: ties dream cycle to the compounding loop and points to ./setup as the on-ramp. - Itemized Added/Changed/For contributors sections list every layer's surfaces with file paths. Also: - Refreshed test/fixtures/golden/{claude,codex,factory}-ship-SKILL.md to match the regenerated ship templates (Step 21 nudge added). - Rebased plan-tune entry in parity-baseline-v1.47.0.0.json from 51717 → 64017 bytes with a baseline_note explaining the cathedral T13 expansion. Documents that the new Dream cycle, Recent auto-decisions, Audit unmarked, Dream cycle review/distill sections are load-bearing, not bloat. Without the rebase, the size-budget gate fails — and the cathedral's whole point is making /plan-tune do more, not less. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: bump VERSION 1.50.0.0 → 1.52.0.0 (queue collision with #1742) CI version gate caught: PR #1742 (garrytan/upgrade-gstack-gbrain-v1) already claims v1.50.0.0 and #1751 (garrytan/browser-memory-leak) claims v1.51.0.0. gstack-next-version util recommends v1.52.0.0 as the next free slot. Updates: - VERSION 1.50.0.0 → 1.52.0.0 - package.json version sync - CHANGELOG.md header + metric table label - parity-baseline-v1.47.0.0.json baseline_note reference No content changes; pure slot rebase per the queue. The cathedral scope (8 layers, 96 tests) and CHANGELOG narrative stay identical — same ship, different release number. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: cap audit — remove distill rate cap, loosen size/budget gates Plan-tune cathedral follow-up. The 3/day distill cap was theatrical: at ~$0.01 per Haiku call, even a runaway loop firing every minute would cost ~$14/day, and free-text events are rare enough that the natural input rate self-limits to 1-2 fires/day. Count caps don't protect against runaway bugs (which fire 1000x/second, not 4 times/day) but DO punish heavy users who'd legitimately distill multiple times during a busy week. Removed: 3/day rate cap on bin/gstack-distill-free-text. --status output swapped from "TODAY: N / 3" to "TODAY: N run(s), $X" so users see what they're spending instead of how close they are to a meaningless count. Loosened (caps that exist for real-runaway protection, not normal scope): - EVALS_BUDGET_HARD_CAP_GATE $25 → $200/run - EVALS_BUDGET_HARD_CAP_PERIODIC $70 → $500/run - EVALS_BUDGET_HARD_CAP $30 → $300/run (umbrella fallback) - GSTACK_SIZE_BUDGET_RATIO 1.05 → 1.50 per-skill ratio - plan-review preamble byte budget 40K → 60K Principle: caps exist to catch obvious bugs (infinite retry, model price change, prompt blowup), not to gate legitimate scope growth. Set high enough that real growth never trips them, only bug territory does. Adjusted defaults are 4-8× historical worst case, leaving ample headroom for the next 12 months of legitimate expansion. Tests updated: distill-free-text removes the 3-test rate-cap describe block in favor of "no rate cap" assertion that 10 runs/day pass. Other budget tests still pass because they were never near the old ceilings. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com> |
||
|
|
0a803f9e81 |
feat: gstack v1 — simpler prompts + real LOC receipts (v1.0.0.0) (#1039)
* docs: add design doc for /plan-tune v1 (observational substrate) Canonical record of the /plan-tune v1 design: typed question registry, per-question explicit preferences, inline tune: feedback with user-origin gate, dual-track profile (declared + inferred separately), and plain-English inspection skill. Captures every decision with pros/cons, what's deferred to v2 with explicit acceptance criteria, and what was rejected entirely. Codex review drove a substantial scope rollback from the initial CEO EXPANSION plan. 15+ legitimate findings (substrate claim was false without a typed registry; E4/E6/clamp logical contradiction; profile poisoning attack surface; LANDED preamble side effect; implementation order) shaped the final shape. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: typed question registry for /plan-tune v1 foundation scripts/question-registry.ts declares 53 recurring AskUserQuestion categories across 15 skills (ship, review, office-hours, plan-ceo-review, plan-eng-review, plan-design-review, plan-devex-review, qa, investigate, land-and-deploy, cso, gstack-upgrade, preamble, plan-tune, autoplan). Each entry has: stable kebab-case id, skill owner, category (approval | clarification | routing | cherry-pick | feedback-loop), door_type (one-way | two-way), optional stable option keys, optional psychographic signal_key, and a one-line description. 12 of 53 are one-way doors (destructive ops, architecture/data forks, security/compliance). These are ALWAYS asked regardless of user preference. Helpers: getQuestion(id), getOneWayDoorIds(), getAllRegisteredIds(), getRegistryStats(). No binary or resolver wiring yet — this is the schema substrate the rest of /plan-tune builds on. Ad-hoc question_ids (not registered) still log but skip psychographic signal attribution. Future /plan-tune skill surfaces frequently-firing ad-hoc ids as candidates for registry promotion. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test: registry schema + safety + coverage tests (gate tier) 20 tests validating the question registry: Schema (7 tests): - Every entry has required fields - All ids are kebab-case and start with their skill name - No duplicate ids - Categories are from the allowed set - door_type is one-way | two-way - Options arrays are well-formed - Descriptions are short and single-line Helpers (5 tests): - getQuestion returns entry for known id, undefined for unknown - getOneWayDoorIds includes destructive questions, excludes two-way - getAllRegisteredIds count matches QUESTIONS keys - getRegistryStats totals are internally consistent One-way door safety (2 tests): - Every critical question (test failure, SQL safety, LLM trust boundary, security scan, merge confirm, rollback, fix apply, premise revise, arch finding, privacy gate, user challenge) is declared one-way - At least 10 one-way doors exist (catches regression if declarations are accidentally dropped) Registry breadth (3 tests): - 11 high-volume skills each have >= 1 registered question - Preamble one-time prompts are registered - /plan-tune's own questions are registered Signal map references (1 test): - signal_key values are typed kebab-case strings Template coverage (2 tests, informational): - AskUserQuestion usage across templates is non-trivial (>20) - Registry spans >= 10 skills 20 pass, 0 fail. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: one-way door classifier (belt-and-suspenders safety fallback) scripts/one-way-doors.ts — secondary keyword-pattern classifier that catches destructive questions even when the registry doesn't have an entry for them. The registry's door_type field (from scripts/question-registry.ts) is the PRIMARY safety gate. This classifier is the fallback for ad-hoc question_ids that agents generate at runtime. Classification priority: 1. Registry lookup by question_id → use declared door_type 2. Skill:category fallback (cso:approval, land-and-deploy:approval) 3. Keyword pattern match against question_summary 4. Default: treat as two-way (safer to log the miss than auto-decide unsafely) Covers 21 destructive patterns across: - File system (rm -rf, delete, wipe, purge, truncate) - Database (drop table/database/schema, delete from) - Git/VCS (force-push, reset --hard, checkout --, branch -D) - Deploy/infra (kubectl delete, terraform destroy, rollback) - Credentials (revoke/reset/rotate API key|token|secret|password) - Architecture (breaking change, schema migration, data model change) 7 new tests in test/plan-tune.test.ts covering: registry-first lookup, unknown-id fallthrough, keyword matching on destructive phrasings including embedded filler words ("rotate the API key"), skill-category fallback, benign questions defaulting to two-way, pattern-list non-empty. 27 pass, 0 fail. 1270 expect() calls. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: psychographic signal map + builder archetypes scripts/psychographic-signals.ts — hand-crafted {signal_key, user_choice} → {dimension, delta} map. Version 0.1.0. Conservative deltas (±0.03 to ±0.06 per event). Covers 9 signal keys: scope-appetite, architecture-care, code-quality-care, test-discipline, detail-preference, design-care, devex-care, distribution-care, session-mode. Helpers: applySignal() mutates running totals, newDimensionTotals() creates empty starting state, normalizeToDimensionValue() sigmoid-clamps accumulated delta to [0,1] (0 → 0.5 neutral), validateRegistrySignalKeys() checks that every signal_key in the registry has a SIGNAL_MAP entry. In v1 the signal map is used ONLY to compute inferred dimension values for /plan-tune inspection output. No skill behavior adapts to these signals until v2. scripts/archetypes.ts — 8 named archetypes + Polymath fallback: - Cathedral Builder (boil-the-ocean + architecture-first) - Ship-It Pragmatist (small scope + fast) - Deep Craft (detail-verbose + principled) - Taste Maker (intuitive, overrides recommendations) - Solo Operator (high-autonomy, delegates) - Consultant (hands-on, consulted on everything) - Wedge Hunter (narrow scope aggressively) - Builder-Coach (balanced steering) - Polymath (fallback when no archetype matches) matchArchetype() uses L2 distance scaled by tightness, with a 0.55 threshold below which we return Polymath. v1 ships the model stable; v2 narrative/vibe commands wire it into user-facing output. 14 new tests: signal map consistency vs registry, applySignal behavior for known/unknown keys, normalization bounds, archetype schema validity, name uniqueness, matchArchetype correctness for each reference profile, Polymath fallback for outliers. 41 pass, 0 fail total in test/plan-tune.test.ts. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: bin/gstack-question-log — append validated AskUserQuestion events Append-only JSONL log at ~/.gstack/projects/{SLUG}/question-log.jsonl. Schema: {skill, question_id, question_summary, category?, door_type?, options_count?, user_choice, recommended?, followed_recommendation?, session_id?, ts} Validates: - skill is kebab-case - question_id is kebab-case, <= 64 chars - question_summary non-empty, <= 200 chars, newlines flattened - category is one of approval/clarification/routing/cherry-pick/feedback-loop - door_type is one-way or two-way - options_count is integer in [1, 26] - user_choice non-empty string, <= 64 chars Injection defense on question_summary rejects the same patterns as gstack-learnings-log (ignore previous instructions, system:, override:, do not report, etc). followed_recommendation is auto-computed when both user_choice and recommended are present. ts auto-injected as ISO 8601 if missing. 21 tests covering: valid payloads, full field preservation, auto-followed computation, appending, long-summary truncation, newline flattening, invalid JSON, missing fields, bad case, oversized ids, invalid enum values, out-of-range options_count, and 6 injection attack patterns. 21 pass, 0 fail, 43 expect() calls. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: bin/gstack-developer-profile — unified profile with migration bin/gstack-developer-profile supersedes bin/gstack-builder-profile. The old binary becomes a one-line legacy shim delegating to --read for /office-hours backward compat. Subcommands: --read legacy KEY:VALUE output (tier, session_count, etc) --migrate folds ~/.gstack/builder-profile.jsonl into ~/.gstack/developer-profile.json. Atomic (temp + rename), idempotent (no-op when target exists or source absent), archives source as .migrated-YYYY-MM-DD-HHMMSS --derive recomputes inferred dimensions from question-log.jsonl using the signal map in scripts/psychographic-signals.ts --profile full profile JSON --gap declared vs inferred diff JSON --trace <dim> event-level trace of what contributed to a dimension --check-mismatch flags dimensions where declared and inferred disagree by > 0.3 (requires >= 10 events first) --vibe archetype name + description from scripts/archetypes.ts --narrative (v2 stub) Auto-migration on first read: if legacy file exists and new file doesn't, migrate before reading. Creates a neutral (all-0.5) stub if nothing exists. Unified schema (see docs/designs/PLAN_TUNING_V0.md §Architecture): {identity, declared, inferred: {values, sample_size, diversity}, gap, overrides, sessions, signals_accumulated, schema_version} 25 new tests across subcommand behaviors: - --read defaults + stub creation - --migrate: 3 sessions preserved with signal tallies, idempotency, archival - Tier calculation: welcome_back / regular / inner_circle boundaries - --derive: neutral-when-empty, upward nudge on 'expand', downward on 'reduce', recomputable (same input → same output), ad-hoc unregistered ids ignored - --trace: contributing events, empty for untouched dims, error without arg - --gap: empty when no declared, correctly computed otherwise - --vibe: returns archetype name + description - --check-mismatch: threshold behavior, 10+ sample requirement - Unknown subcommand errors 25 pass, 0 fail, 60 expect() calls. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: bin/gstack-question-preference — explicit preferences + user-origin gate Subcommands: --check <id> → ASK_NORMALLY | AUTO_DECIDE (decides if a registered question should be auto-decided by the agent) --write '{…}' → set a preference (requires user-origin source) --read → dump preferences JSON --clear [id] → clear one or all --stats → short counts summary Preference values: always-ask | never-ask | ask-only-for-one-way. Stored at ~/.gstack/projects/{SLUG}/question-preferences.json. Safety contract (the core of Codex finding #16, profile-poisoning defense from docs/designs/PLAN_TUNING_V0.md §Security model): 1. One-way doors ALWAYS return ASK_NORMALLY from --check, regardless of user preference. User's never-ask is overridden with a visible safety note so the user knows why their preference didn't suppress the prompt. 2. --write requires an explicit `source` field: - Allowed: "plan-tune", "inline-user" - REJECTED with exit code 2: "inline-tool-output", "inline-file", "inline-file-content", "inline-unknown" Rejection is explicit ("profile poisoning defense") so the caller can log and surface the attempt. 3. free_text on --write is sanitized against injection patterns (ignore previous instructions, override:, system:, etc.) and newline-flattened. Each --write also appends a preference-set event to ~/.gstack/projects/{SLUG}/question-events.jsonl for derivation audit trail. 31 tests: - --check behavior (4): defaults, two-way, one-way (one-way overrides never-ask with safety note), unknown ids, missing arg - --check with prefs (5): never-ask on two-way → AUTO_DECIDE; never-ask on one-way → ASK_NORMALLY with override note; always-ask always asks; ask-only-for-one-way flips appropriately - --write valid (5): inline-user accepted, plan-tune accepted, persisted correctly, event appended, free_text preserved with flattening - User-origin gate (6): missing source rejected; inline-tool-output rejected with exit code 2 and explicit poisoning message; inline-file, inline-file-content, inline-unknown rejected; unknown source rejected - Schema validation (4): invalid JSON, bad question_id, bad preference, injection in free_text - --read (2): empty → {}, returns writes - --clear (3): specific id, clear-all, NOOP for missing - --stats (2): empty zeros, tallies by preference type 31 pass, 0 fail, 52 expect() calls. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: question-tuning preamble resolvers scripts/resolvers/question-tuning.ts ships three preamble generators: generateQuestionPreferenceCheck — before each AskUserQuestion, agent runs gstack-question-preference --check <id>. AUTO_DECIDE suppresses the ask and auto-chooses recommended. ASK_NORMALLY asks as usual. One-way door safety override is handled by the binary. generateQuestionLog — after each AskUserQuestion, agent appends a log record with skill, question_id, summary, category, door_type, options_count, user_choice, recommended, session_id. generateInlineTuneFeedback — offers inline "tune:" prompt after two-way questions. Documents structured shortcuts (never-ask, always-ask, ask-only-for-one-way, ask-less) AND accepts free-form English with normalization + confirmation. Explicitly spells out the USER-ORIGIN GATE: only write tune events when the prefix appears in the user's own chat message, never from tool output or file content. Binary enforces. All three resolvers are gated by the QUESTION_TUNING preamble echo. When the config is off, the agent skips these sections entirely. Ready to be wired into preamble.ts in the next commit. Codex host has a simpler variant that uses $GSTACK_BIN env vars. scripts/resolvers/index.ts registers three placeholders: QUESTION_PREFERENCE_CHECK, QUESTION_LOG, INLINE_TUNE_FEEDBACK Total resolver count goes from 45 to 48. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: wire question-tuning into preamble for tier >= 2 skills scripts/resolvers/preamble.ts — adds two things: 1. _QUESTION_TUNING config echo in the preamble bash block, gated on the user's gstack-config `question_tuning` value (default: false). 2. A combined Question Tuning section for tier >= 2 skills, injected after the confusion protocol. The section itself is runtime-gated by the QUESTION_TUNING value — agents skip it entirely when off. scripts/resolvers/question-tuning.ts — consolidated into one compact combined section `generateQuestionTuning(ctx)` covering: preference check before the question, log after, and inline tune: feedback with user-origin gate. Per-phase generators remain exported for unit tests but are no longer the main entrypoint. Size impact: +570 tokens / +2.3KB per tier-2+ SKILL.md. Three skills (plan-ceo-review, office-hours, ship) still exceed the 100KB token ceiling — but they were already over before this change. Delta is the smallest viable wiring of the /plan-tune v1 substrate. Golden fixtures (test/fixtures/golden/claude-ship, codex-ship, factory-ship) regenerated to match the new baseline. Full test run: 1149 pass, 0 fail, 113 skip across 28 files. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: regenerate SKILL.md files with question-tuning section bun run gen:skill-docs --host all after wiring the QUESTION_TUNING preamble section. Every tier >= 2 skill now includes the combined Question Tuning guidance. Runtime-gated — agents skip the section when question_tuning is off in gstack-config (default). Golden fixtures (claude-ship, codex-ship, factory-ship) updated to the new baseline. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: /plan-tune skill — conversational inspection + preferences plan-tune/SKILL.md.tmpl: the user-facing skill for /plan-tune v1. Routes plain-English intent to one of 8 flows: - Enable + setup (first-time): 5 declaration questions mapping to the 5 psychographic dimensions (scope_appetite, risk_tolerance, detail_preference, autonomy, architecture_care). Writes to developer-profile.json declared.*. - Inspect profile: plain-English rendering of declared + inferred + gap. Uses word bands (low/balanced/high) not raw floats. Shows vibe archetype when calibration gate is met. - Review question log: top-20 question frequencies with follow/override counts. Highlights override-heavy questions as candidates for never-ask. - Set a preference: normalizes "stop asking me about X" → never-ask, etc. Confirms ambiguous phrasings before writing via gstack-question-preference. - Edit declared profile: interprets free-form ("more boil-the-ocean") and CONFIRMS before mutating declared.* (trust boundary per Codex #15). - Show gap: declared vs inferred diff with plain-English severity bands (close / drift / mismatch). Never auto-updates declared from the gap. - Stats: preference counts + diversity/calibration status. - Enable / disable: gstack-config set question_tuning true|false. Design constraints enforced: - Plain English everywhere. No CLI subcommand syntax required. Shortcuts (`profile`, `vibe`, `stats`, `setup`) exist but optional. - user-origin gate on tune: writes. source: "plan-tune" for user-invoked /plan-tune; source: "inline-user" for inline tune: from other skills. - One-way doors override never-ask (safety, surfaced to user). - No behavior adaptation in v1 — this skill inspects and configures only. Generates plan-tune/SKILL.md at ~11.6k tokens, well under the 100KB ceiling. Generated for all hosts via `bun run gen:skill-docs --host all`. Full free test suite: 1149 pass, 0 fail, 113 skip across 28 files. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test: end-to-end pipeline + preamble injection coverage Added 6 tests to test/plan-tune.test.ts: Preamble injection (3 tests): - tier 2+ includes Question Tuning section with preference check, log, and user-origin gate language ('profile-poisoning defense', 'inline-user') - tier 1 does NOT include the prose section (QUESTION_TUNING bash echo still fires since it's in the bash block all tiers share) - codex host swaps binDir references to $GSTACK_BIN End-to-end pipeline (3 tests) — real binaries working together, not mocks: - Log 5 expand choices → --derive → profile shows scope_appetite > 0.5 (full log → registry lookup → signal map → normalization round-trip) - --write source: inline-tool-output rejected; --read confirms no pref was persisted (the profile-poisoning defense actually works end-to-end) - Migrate a 3-session legacy file; confirm legacy gstack-builder-profile shim still returns SESSION_COUNT: 3, TIER: welcome_back, CROSS_PROJECT: true test/plan-tune.test.ts now has 47 tests total. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test: E2E test for /plan-tune plain-English inspection flow (gate tier) test/skill-e2e-plan-tune.test.ts — verifies /plan-tune correctly routes plain-English intent ("review the questions I've been asked") to the Review question log section without requiring CLI subcommand syntax. Seeds a synthetic question-log.jsonl with 3 entries exercising: - override behavior (user chose expand over recommended selective) - one-way door respect (user followed ship-test-failure-triage recommendation) - two-way override (user skipped recommended changelog polish) Invokes the skill via `claude -p` and asserts: - Agent surfaces >= 2 of 3 logged question_ids in output - Agent notices override/skip behavior from the log - Exit reason is success or error_max_turns (not agent-crash) Gate-tier because the core v1 DX promise is plain-English intent routing. If it requires memorized subcommands or breaks on natural language, that's a regression of the defining feature. Registered in test/helpers/touchfiles.ts with dependencies: - plan-tune/** (skill template + generated md) - scripts/question-registry.ts (required for log lookup) - scripts/psychographic-signals.ts, scripts/one-way-doors.ts (derive path) - bin/gstack-question-log, gstack-question-preference, gstack-developer-profile Skipped when EVALS_ENABLED is not set; runs on `bun run test:evals`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: bump version and changelog (v0.19.0.0) — /plan-tune v1 Ships /plan-tune as observational substrate: typed question registry, dual-track developer profile (declared + inferred), explicit per-question preferences with user-origin gate, inline tune: feedback across every tier >= 2 skill, unified developer-profile.json with migration from builder-profile.jsonl. Scope rolled back from initial CEO EXPANSION plan after outside-voice review (Codex). 6 deferrals tracked as P0 TODOs with explicit acceptance criteria: E1 substrate wiring, E3 narrative/vibe, E4 blind-spot coach, E5 LANDED celebration, E6 auto-adjustment, E7 psychographic auto-decide. See docs/designs/PLAN_TUNING_V0.md for the full design record. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(ci): harden Dockerfile.ci against transient Ubuntu mirror failures The CI image build failed with: E: Failed to fetch http://archive.ubuntu.com/ubuntu/pool/main/... Connection failed [IP: 91.189.92.22 80] ERROR: process "/bin/sh -c apt-get update && apt-get install ..." did not complete successfully: exit code: 100 archive.ubuntu.com periodically returns "connection refused" on individual regional mirrors. Without retry logic a single failed fetch nukes the whole Docker build. Three defenses, layered: 1. /etc/apt/apt.conf.d/80-retries — apt fetches each package up to 5 times with a 30s timeout. Handles per-package flakes. 2. Shell-loop retry around the whole apt-get step (x3, 10s sleep) — handles the case where apt-get update itself can't reach any mirror. 3. --retry 5 --retry-delay 5 --retry-connrefused on all curl fetches (bun install script, GitHub CLI keyring, NodeSource setup script). Applied to every apt-get and curl call in the Dockerfile. No behavior change on happy path — only kicks in when mirrors blip. Fixes the build-image job that was blocking CI on the /plan-tune PR. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: add PLAN_TUNING_V1 + PACING_UPDATES_V0 design docs Captures the V1 design (ELI10 writing + LOC reframe) in docs/designs/PLAN_TUNING_V1.md and the extracted V1.1 pacing-overhaul plan in docs/designs/PACING_UPDATES_V0.md. V1 scope was reduced from the original bundled pacing + writing-style plan after three engineering-review passes revealed structural gaps in the pacing workstream that couldn't be closed via plan-text editing. TODOS.md P0 entry links to V1.1. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: curated jargon list for V1 writing-style glossing Repo-owned list of ~50 high-frequency technical terms (idempotent, race condition, N+1, backpressure, etc.) that gstack glosses on first use in tier-≥2 skill output. Baked into generated SKILL.md prose at gen-skill-docs time. Terms not on this list are assumed plain-English enough. Contributions via PR. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(preamble): V1 Writing Style section + EXPLAIN_LEVEL echo + migration prompt Adds a new Writing Style section to tier-≥2 preamble output composing with the existing AskUserQuestion Format section. Six rules: jargon glossed on first use per skill invocation (from scripts/jargon-list.json), outcome- framed questions, short sentences, decisions close with user impact, gloss-on-first-use even if user pasted term, user-turn override for "be terse" requests. Baked conditionally (skip if EXPLAIN_LEVEL: terse). Adds EXPLAIN_LEVEL preamble echo using \${binDir} (host-portable matching V0 QUESTION_TUNING pattern). Adds WRITING_STYLE_PENDING echo reading a flag file written by the V0→V1 upgrade migration; on first post-upgrade skill run, the agent fires a one-time AskUserQuestion offering terse mode. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(gstack-config): validate explain_level + document in header Adds explain_level: default|terse to the annotated config header with a one-line description. Whitelists valid values; on set of an unknown value, prints a specific warning ("explain_level '\$VALUE' not recognized. Valid values: default, terse. Using default.") and writes the default value. Matches V1 preamble's EXPLAIN_LEVEL echo expectation. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: V1 upgrade migration — writing-style opt-out prompt New migration script following existing v0.15.2.0.sh / v0.16.2.0.sh pattern. Writes a .writing-style-prompt-pending flag file on first run post-upgrade. The preamble's migration-prompt block reads the flag and fires a one-time AskUserQuestion offering the user a choice between the new default writing style and restoring V0 prose via \`gstack-config set explain_level terse\`. Idempotent via flag files; if the user has already set explain_level explicitly, counts as answered and skips. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: LOC reframe tooling — throughput comparison + README updater + scc installer Three new scripts: - scripts/garry-output-comparison.ts — enumerates Garry-authored commits in 2013 + 2026 on public repos, extracts ADDED lines from git diff, classifies as logical SLOC via scc --stdin (regex fallback if scc missing). Writes docs/throughput-2013-vs-2026.json with per-language breakdown + explicit caveats (public repos only, commit-style drift, private-work exclusion). - scripts/update-readme-throughput.ts — reads the JSON if present, replaces the README's <!-- GSTACK-THROUGHPUT-PLACEHOLDER --> anchor with the computed multiple (preserving the anchor for future runs). If JSON missing, writes GSTACK-THROUGHPUT-PENDING marker that CI rejects — forcing the build to run before commit. - scripts/setup-scc.sh — standalone OS-detecting installer for scc. Not a package.json dependency (95% of users never run throughput). Brew on macOS, apt on Linux, GitHub releases link on Windows. Two-string anchor pattern (PLACEHOLDER vs PENDING) prevents the pipeline from destroying its own update path. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(retro): surface logical SLOC + weighted commits above raw LOC V1 reorders the /retro summary table to lead with features shipped, then commits + weighted commits (commits × files-touched capped at 20), then PRs merged, then logical SLOC added as the primary code-volume metric. Raw LOC stays present but is demoted to context. Rationale inline in the template: ten lines of a good fix is not less shipping than ten thousand lines of scaffold. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs(v1): README hero reframe + writing-style + CHANGELOG + version bump to 1.0.0.0 README.md: - Hero removes "600,000+ lines of production code" framing; replaces with the computed 2013-vs-2026 pro-rata multiple (via <!-- GSTACK-THROUGHPUT-PLACEHOLDER --> anchor, filled by the update-readme-throughput build step). - Hiring callout: "ship real products at AI-coding speed" instead of "10K+ LOC/day." - New Writing Style section (~80 words) between Quick start and Install: "v1 prompts = simpler" framing, outcome-language example, terse-mode opt-out, pointer to /plan-tune. CLAUDE.md: one-paragraph Writing style (V1) note under project conventions, linking to preamble resolver + V1 design docs. CHANGELOG.md: V1 entry on top of v0.19.0.0 with user-facing narrative (what changes, how to opt out, for-contributors notes). Mentions scope reduction — pacing overhaul ships in V1.1. CONTRIBUTING.md: one-paragraph note on jargon-list.json maintenance (PR to add/remove terms; regenerate via gen:skill-docs). VERSION + package.json: bump to 1.0.0.0. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: regenerate SKILL.md files + golden fixtures for V1 Mechanical regeneration from the updated templates in prior commits: - Writing Style section now appears in tier-≥2 skill output. - EXPLAIN_LEVEL + WRITING_STYLE_PENDING echoes in preamble bash. - V1 migration-prompt block fires conditionally on first upgrade. - Jargon list inlined into preamble prose at gen time. - Retro template's logical SLOC + weighted commits order applied. Regenerated for all 8 hosts via bun run gen:skill-docs --host all. Golden ship-skill fixtures refreshed from regenerated outputs. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test: V1 gate coverage — writing-style resolver + config + jargon + migration + dormancy Six new gate-tier test files: - test/writing-style-resolver.test.ts — asserts Writing Style section is injected into tier-≥2 preamble, all 6 rules present, jargon list inlined, terse-mode gate condition present, Codex output uses \$GSTACK_BIN (not ~/.claude/), tier-1 does NOT get the section, migration-prompt block present. - test/explain-level-config.test.ts — gstack-config set/get round-trip for default + terse, unknown-value warns + defaults to default, header documents the key, round-trip across set→set→get. - test/jargon-list.test.ts — shape + ~50 terms + no duplicates (case-insensitive) + includes canonical high-signal terms. - test/v0-dormancy.test.ts — 5D dimension names + archetype names forbidden in default-mode tier-≥2 SKILL.md output, except for plan-tune and office-hours where they're load-bearing. - test/readme-throughput.test.ts — script replaces anchor with number on happy path, writes PENDING marker when JSON missing, CI gate asserts committed README contains no PENDING string. - test/upgrade-migration-v1.test.ts — fresh run writes pending flag, idempotent after user-answered, pre-existing explain_level counts as answered. All 95 V1 test-expect() calls pass. Full suite: 0 failures. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: compute real 2013-vs-2026 throughput multiple (130.2×) Ran scripts/garry-output-comparison.ts across all 15 public garrytan/* repos. Aggregated results into docs/throughput-2013-vs-2026.json and ran scripts/update-readme-throughput.ts to replace the README placeholder. 2013 public activity: 2 commits, 2,384 logical lines added across 1 week, in 1 repo (zurb-foundation-wysihtml5 upstream contribution). 2026 public activity: 279 commits, 310,484 logical lines added across 17 active weeks, in 3 repos (gbrain, gstack, resend_robot). Multiples (public repos only, apples-to-apples): - Logical SLOC: 130.2× - Commits per active week: 8.2× - Raw lines added: 134.4× Private work at both eras (2013 Bookface at YC, Posterous-era code, 2026 internal tools) is excluded from this comparison. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat: 207× throughput multiple (with private repos + Bookface) Re-ran scripts/garry-output-comparison.ts across all 41 repos under garrytan/* (15 public + 26 private), including Bookface (YC's internal social network, 2013-era work). 2013 activity: 71 commits, 5,143 logical lines, 4 active repos (bookface, delicounter, tandong, zurb-foundation-wysihtml5) 2026 activity: 350 commits, 1,064,818 logical lines, 15 active repos (gbrain, gstack, gbrowser, tax-app, kumo, tenjin, autoemail, kitsune, easy-chromium-compiles, conductor-playground, garryslist-agent, baku, gstack-website, resend_robot, garryslist-brain) Multiples: - Logical SLOC: 207× (up from 130.2× when including private work) - Raw lines: 223× - Commits/active-week: 3.4× Stopped committing docs/throughput-2013-vs-2026.json — analysis is a local artifact, not repo state. Added docs/throughput-*.json to .gitignore. Full markdown analysis at ~/throughput-analysis-2026-04-18.md (local-only). README multiple is now hardcoded; re-run the script and edit manually when you want to refresh it. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: run rate vs year-to-date throughput comparison Two separate numbers in the README hero: - Run rate: ~700× (9,859 logical lines/day in 2026 vs 14/day in 2013) - Year-to-date: 207× (2026 through April 18 already exceeds 2013 full year by 207×) Previous "207× pro-rata" framing mixed full-year 2013 vs partial-year 2026. Run rate is the apples-to-apples normalization; YTD is the "already produced" total. Both are honest; both are compelling; they measure different things. Analysis at ~/throughput-analysis-2026-04-18.md (local-only). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(throughput): script natively computes to-date + run-rate multiples Enhanced scripts/garry-output-comparison.ts so both calculations come out of a single run instead of being reassembled ad-hoc in bash: PerYearResult now includes: - days_elapsed — 365 for past years, day-of-year for current - is_partial — flags the current (in-progress) year - per_day_rate — logical/raw/commits normalized by calendar day - annualized_projection — per_day_rate × 365 Output JSON's `multiples` now has two sibling blocks: - multiples.to_date — raw volume ratios (2026-YTD / 2013-full-year) - multiples.run_rate — per-day pace ratios (apples-to-apples) Back-compat: multiples.logical_lines_added still aliases to_date for older consumers reading the JSON. Updated README hero to cite both (picking up brain/* repo that was missed in the earlier aggregation pass): 2026 run rate: ~880× my 2013 pace (12,382 vs 14 logical lines/day) 2026 YTD: 260× the entire 2013 year Stderr summary now prints both multiples at the end of each run. Full analysis at ~/throughput-analysis-2026-04-18.md (local-only). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: ON_THE_LOC_CONTROVERSY methodology post + README link Long-form response to the "LOC is a meaningless vanity metric" critique. Covers: - The three branches of the LOC critique and which are right - Why logical SLOC (NCLOC) beats raw LOC as the honest measurement - Full method: author-scoped git diff, regex-classified added lines, aggregated across 41 public + private garrytan/* repos - Both calculations: to-date (260x) and run-rate (879x) - Steelman of the critics (greenfield-vs-maintenance, survivorship bias, quality-adjusted productivity, time-to-first-user) - Reproduction instructions Linked from README hero via a blockquote directly below the number. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * exclude: tax-app from throughput analysis (import-dominated history) tax-app's history is one commit of 104K logical lines — an initial import of a codebase, not authored work. Removing it to keep the comparison honest. Changes: - scripts/garry-output-comparison.ts: added EXCLUDED_REPOS constant with tax-app + a one-line rationale. The script now skips excluded repos with a stderr note and deletes any stale output JSON so aggregation loops don't pick up pre-exclusion numbers. - README hero: updated to 810× run rate + 240× YTD (were 880×/260×). Wording updated to "40 public + private repos ... after excluding repos dominated by imported code." - docs/ON_THE_LOC_CONTROVERSY.md: updated all numbers, added an "Exclusions" paragraph explaining tax-app, removed tax-app from the "shipped not WIP" example list. New numbers (2026 through day 108, without tax-app): - To-date: 240× logical SLOC (1,233,062 vs 5,143) - Run rate: 810× per-day pace (11,417 vs 14 logical/day) - Annualized: ~4.2M logical lines projected Future re-runs automatically skip tax-app. Add more exclusions to EXCLUDED_REPOS at the top of the script with a one-line rationale. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix: correct tax-app exclusion rationale tax-app is a demo app I built for an upcoming YC channel video, not an "import-dominated history" as the previous commit claimed. Excluded because it's not production shipping work, not because of an import commit. Updated rationale in scripts/garry-output-comparison.ts's EXCLUDED_REPOS constant, in docs/ON_THE_LOC_CONTROVERSY.md's method section + conclusion, and in the README hero wording ("one demo repo" vs the earlier "repos dominated by imported code"). Numbers unchanged — the exclusion itself is the same, just the reason. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: harden ON_THE_LOC_CONTROVERSY against Cramer + neckbeard critiques Reframes the thesis as "engineers can fly now" (amplification, not replacement) and fortifies the soft spots critics will attack. Added: - Flight-thesis opener: pilot vs walker, leverage not replacement. - Second deflation layer for AI verbosity (on top of NCLOC). Headline moves from 810x to 408x after generous 2x AI-boilerplate cut, with explicit sensitivity analysis showing the number is still large under pessimistic priors (5x → 162x, 10x → 81x, 100x impossible). - Weekly distribution check (kills "you had one burst week" attack). - Revert rate (2.0%) and post-merge fix rate (6.3%) with OSS comparables (K8s/Rails/Django band). Addresses "where are your error rates" directly. - Named production adoption signals (gstack 1000+ installs, gbrain beta, resend_robot paying API) with explicit concession that "shipped != used at scale" for most of the corpus. - Harder steelman: 5 specific concessions with quantified pivot points (e.g., "if 2013 baseline was 3.5x higher, 810x → 228x, still high"). Removed factual error: Posterous acquisition paragraph (Garry had already left Posterous by 2011, so the "Twitter bought our private repos" excuse for the 2013 corpus gap doesn't apply). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: update gstack/gbrain adoption numbers in LOC controversy post gstack: "1,000+ distinct project installations" → "tens of thousands of daily active users" (telemetry-reported, community tier, opt-in). gbrain: "small set of beta testers" → "hundreds of beta testers running it live." Both are the accurate current numbers. The concession paragraph below (about shipped != adopted at scale for the long-tail repos) still reads correctly since it's about the corpus as a whole, not gstack/gbrain specifically. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: reframe reproducibility note as OSS breakout flex "You'd need access to my private repos" → "Bookface and Posthaven are private, but gstack and gbrain are open-sourced with tens of thousands of GitHub stars and tens of thousands of confirmed regular users, among the most-used OSS projects in the world that didn't exist three months ago." Keeps the `gh repo list` command at the end for the actual reproducibility instruction. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * Rewrite LOC controversy post - Lead with concession (LOC is garbage, do the math anyway) - Preempt 14 lines/day meme with historical baselines (Brooks, Jones, McConnell) - Remove 'neckbeard' language throughout - Add slop-scan story (Ben Vinegar, 5.24 → 1.96, 62% cut) - David Cramer GUnit joke - Add testing philosophy section (the real unlock) - ASCII weekly distribution chart - gstack telemetry section with real numbers (15K installs, 305K invocations, 95.2% success) - Top skills usage chart - Pick-your-priors paragraph moved earlier (the killer) - Sharper close: run the script, show me your numbers * docs: four precision fixes on LOC controversy post 1. Citation fix. Kernighan didn't say anything about LOC-as-metric (that's the famous "aircraft building by weight" quote, commonly misattributed but actually Bill Gates). Replaced "Kernighan implied it before that" with the real Dijkstra quote ("lines produced" vs "lines spent" from EWD1036, with direct link) + the Gates quote. Verified via web search. 2. Slop-scan direction clarified. "(highest on his benchmark)" was ambiguous — could read as a brag. Now: "Higher score = more slop. He ran it on gstack and we scored 5.24, the worst he'd measured at the time." Then the 62% cut lands as an actual win. 3. Prose/chart skill-usage ordering now matches. Added /plan-eng-review (28,014) to the prose list so it doesn't conflict with the chart below it. 4. Cut the "David — I owe you one / GUnit" insider joke. Most readers won't connect Cramer → Sentry → GUnit naming. Ends the slop-scan paragraph on the stronger line: "Run `bun test` and watch 2,000+ tests pass." Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: tighten four LOC post citations to match primary sources 1. Bill Gates quote: flagged as folklore-grade. Was "Bill Gates put it more memorably" (firm attribution). Now "The old line (widely attributed to Bill Gates, sourcing murky) puts it more memorably." The quote stands; honesty about attribution avoids the same misattribution trap we just fixed for Kernighan. 2. Capers Jones: "15-50 across thousands of projects" → "roughly 16-38 LOC/day across thousands of projects" — matches his actual published measurements (which also report as 325-750 LOC/month). 3. Steve McConnell: "10-50 for finished, tested, delivered code" was folklore. Replaced with his actual project-size-dependent range from Code Complete: "20-125 LOC/day for small projects (10K LOC) down to 1.5-25 for large projects (10M LOC) — it's size-dependent, not a single number." 4. Revert rate comparison: "Kubernetes, Rails, and Django historically run 1.5-3%" was unsourced. Replaced with "mature OSS codebases typically run 1-3%" + "run the same command on whatever you consider the bar and compare." No false specificity about which repos. Net: every quantitative citation in the post now matches primary-source figures or is explicitly flagged as folklore. Neckbeards can verify. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: drop Writing style section from README Was sitting in prime real estate between Quick start and Install — internal implementation detail, not something users need up-front. Existing coverage is enough: - Upgrade migration prompt notifies users on first post-upgrade run - CLAUDE.md has the contributor note - docs/designs/PLAN_TUNING_V1.md has the full design Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: collapse team-mode setup into one paste-and-go command Step 2 was three separate code blocks: setup --team, then team-init, then git add/commit. Mirrors Step 1's style now — one shell one-liner that does all three. Subshell (cd && ./setup --team) keeps the user in their repo pwd so team-init + git commit land in the right place. "Swap required for optional" moved to a one-liner below. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * docs: move full-clone footnote from README to CONTRIBUTING The "Contributing or need full history?" note is for contributors, not for someone following the README install flow. Moved into CONTRIBUTING's Quick start section where it fits next to the existing clone command, with a tip to upgrade an existing shallow clone via \`git fetch --unshallow\`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: root <root@localhost> |