Repository navigation
v1.91.276.0 feat(preamble): tier 2+ skills load the shared preamble once per session (GSTA-96) - #3434
Merged
Merged
Conversation
… (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ion (GSTA-96) Every tier 2+ SKILL.md inlined the same ~23 KB preamble. It now renders once per host into sections/preamble.md beside each tier 2+ SKILL.md; the SKILL.md keeps a pointer that asks for the Read only when the file's id is not already in context. Only sections byte-identical to the shared render move, so every skill keeps exactly its pre-hoist paragraphs (verified across all 41 skills). Parity caps for ten skills raised to their measured unions (owner-approved, Garry 2026-10-25: "do it"); 63 size pins lowered and 26 deleted. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
|
😎 This pull request was merged. |
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> # Conflicts: # CHANGELOG.md
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> # Conflicts: # CHANGELOG.md
…mplate's identity in the PR planner (GSTA-96) screen-summary failed on #3434 with 'PR profile requires full validation' for nine skills whose only paid coverage is periodic: their new sections/preamble.md had no dependency identity, while the SKILL.md beside it inherits its template's. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…the hoist resolver, not a section template (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…odel-overlays guards the coverage half) The completeness section shrank by 3 characters (5 bytes), matching the 89 pins lowered by 3. The default-copy test pinned only the minimal-code half; an expectMentions check now fails if the full-coverage or never-excuses-a-gap clause is dropped. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> # Conflicts: # CHANGELOG.md # test/fixtures/skill-md-size-allowlist.json
…qual coverage) The default-copy check now also fails if 'on equal coverage' is dropped from the smaller-diff rule (which would tell models to prefer the smaller diff unconditionally) or if the no-impossible-state-checks clause is removed. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
The [Unreleased] Tests bullet now says the default-copy test checks both halves of the rule (full coverage of the requirement; smaller diff only at equal coverage; no impossible-state checks; never excuses a gap), matching 88e3cc0 and 44eacaf. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ore code Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…pe; tests bind each completeness rule) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Adds test/preamble-hoist.test.ts (tier 1 unchanged; tier 2+ SKILL.md plus sections/preamble.md reproduces the pre-hoist preamble; pointer position and id) and two gen-skill-docs-checks cases (one shared preamble per host, beside every pointer and named by its id; preamble artifact kind). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
- setup: follow a symlinked sections/ dir in the carved-pointer status check - touchfiles: select preamble-hoist.ts alongside every preamble.ts entry - section-manifest: exempt sections/preamble.md only for skills whose SKILL.md carries the shared-preamble pointer - skill-size-budget: growth ceiling measures SKILL.md plus its shared preamble - host-conformance: tolerate a missing sections/ dir, as setup does - skill-validation: read bin/gstack-skill-start directly - skill-fixture: error messages and docblocks name the actual boundary (harness-version non-behavioral note updated) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ep 9.4, pass 2) The growth ceiling now measures SKILL.md plus the shared preamble, the same measure as the shrink floor; the docblock said SKILL.md only. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
….md restates the new one Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…(ship Step 9.4) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…wording Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> # Conflicts: # CHANGELOG.md # test/fixtures/skill-md-size-allowlist.json
plan-ceo-review's parity cap moves 1.0999 -> 1.1006 for the approved COMPLETENESS IS COVERAGE bullet (Garry 2026-10-25: "do it", parity caps to measured size; applied by GStack Minder). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…rule quote (ship Step 7) host-conformance: --status still finds a broken bare sections/ pointer when a carved Claude skill's installed sections/ is a directory symlink (fails at the pre-repair 0e5d11f). gen-skill-docs: the house rule review/specialists/ simplification.md quotes from ETHOS.md still appears in ETHOS.md. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…first-run intro say complete coverage, minimal code (GSTA-96) The pointer skipped the shared file when any text carrying its id was in context, so a page or file with that heading could stand in for the anti-spoofing rules (security, red team and adversarial review, ship run 2). It now requires the skill's own earlier Read. The digest's Boil the Ocean bullet and the first-run intro still said 'AI makes completeness cheap, so do the complete thing' (docs audit blocker); both now carry Garry's 'complete coverage, minimal code' (2026-10-25: '1/ yes fix this'). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ean wording (ship Step 7) preamble-hoist: on every host, tier and render option the pointer allows the skip only when the skill's own Read is in context. agents-digest and gstack-skill-start: the digest bullet and the lake intro say complete coverage, minimal code, never the retired phrasing. Each fails at 9af36d0. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…id-only one (ship Step 9.4) The preamble-hoist.ts docblock and ARCHITECTURE.md still said the shared preamble is re-read only when its id is absent; since 2acf1a4 the pointer requires the skill's own earlier Read of the file. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ng tests (ship Step 13) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…head cannot change a release output (GSTA-96) CI 'Check VERSION is not stale vs queue' classified #3434 as stamped because it touched agents-digest/gstack-AGENTS.md (a release_outputs path), while regenerating only the generator fails the digest freshness test. The first-run intro rewording stays. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
4 tasks done
CHANGELOG additive. #3472's companion fixtureDocs test removes ship's generated sections/preamble.md from its copied tree before asserting that a skill without one gets none: under the hoist every generated tier 2+ skill, ship included, has one, so the train-14 assertion at :249 named a skill that does have it. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
garrytan
added a commit
that referenced
this pull request
Oct 26, 2026
Conflicts: CHANGELOG (dedupe entry kept above #3434's; its ship-cap sentence dropped because #3434 carries the cap), scripts/resolvers/preamble-hoist.ts (dedupe's added-subsection support plus #3434's empty-variant guard: no changed and no added subsections returns null), office-hours/SKILL.md regenerated. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Rule 19 grant (train 15 QUEUE_TRAIN_CONFLICT). CHANGELOG additive. test/fixtures/context-budget.json by hand: canary keeps the hoisted ceiling (main did not change it; the conflict was main's adjacent new captain row); captain (new on main, now hoisted) and five loaded-skill ceilings lowered to a fresh capture (captain 12011 -> 6551, autoplan, office-hours, plan-ceo-review, plan-devex-review, plan-eng-review by 2-22). No ceiling raised. Generated files and ship goldens regenerated. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…hoisted) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…preamble-hoist Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> # Conflicts: # ARCHITECTURE.md # CHANGELOG.md # docs/PROJECT_STRUCTURE.md # test/fixtures/context-budget.json # test/fixtures/skill-md-size-allowlist.json # test/helpers/carve-guards.ts # test/helpers/touchfiles-data.ts
garrytan
added a commit
that referenced
this pull request
Oct 27, 2026
…wo context-budget rows above main (GSTA-96, per GStack Minder) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…preamble-hoist Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> # Conflicts: # CHANGELOG.md # test/fixtures/context-budget.json
garrytan
added a commit
that referenced
this pull request
Oct 27, 2026
…s after its v1.91.265.0 merge (GSTA-96, per GStack Minder) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
This was referenced Oct 27, 2026
… install (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…vesting wording (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…preamble-hoist Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ts sections path mapped (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…preamble-hoist Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…ncated (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…preamble-hoist Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
… base=origin/main@f8d0ab0 policy=base origin/main@f8d0ab033914:.gstack/ship-policy.json) ship, next in queue at 1.91.276.0 STAMP_VERSION: 1.91.276.0 claims: ignored (stamp-at-merge) stage version: done (wrote 1.91.276.0) stage release_tool: done (bun scripts/gen-agents-digest.ts) stage changelog: done (Unreleased -> 1.91.276.0) stage stamp_paths: done (1 mention(s) in 1 file(s) (1.91.270.0 -> 1.91.276.0)) stage mirror: skipped (no mirror in policy) stage apply: done (5 file(s)) FILES: CHANGELOG.md VERSION agents-digest/gstack-AGENTS.md lib/ship-policy.ts package.json MERGE_CONDITION: sha=234227fc48b5200519474e53c33c152135723372 next: NEW_VERSION=1.91.276.0; commit and push only the stamp diff above (FILES), with no edit of your own: the train's verify accepts the CHANGELOG only as the base's plus this PR's entry re-headed; merge with the expected head 234227f and recompute on a stale queue. GSTACK_RESULT: skill=restamp status=complete run=/workspace/gstack/.gstack/tmp/restamp-journal.json Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
capy-ai Bot
added a commit
that referenced
this pull request
Oct 27, 2026
…UserQuestion subsections (GSTA-96) (#3453) * fix(preamble): completeness means complete coverage with minimal code (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * feat(preamble): tier 2+ skills load the shared preamble once per session (GSTA-96) Every tier 2+ SKILL.md inlined the same ~23 KB preamble. It now renders once per host into sections/preamble.md beside each tier 2+ SKILL.md; the SKILL.md keeps a pointer that asks for the Read only when the file's id is not already in context. Only sections byte-identical to the shared render move, so every skill keeps exactly its pre-hoist paragraphs (verified across all 41 skills). Parity caps for ten skills raised to their measured unions (owner-approved, Garry 2026-10-25: "do it"); 63 size pins lowered and 26 deleted. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix(paid-select): a generated sections/preamble.md takes its skill template's identity in the PR planner (GSTA-96) screen-summary failed on #3434 with 'PR profile requires full validation' for nine skills whose only paid coverage is periodic: their new sections/preamble.md had no dependency identity, while the SKILL.md beside it inherits its template's. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test(section-manifest): the shared sections/preamble.md renders from the hoist resolver, not a section template (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: pre-landing review fixes (CHANGELOG says 3 characters shorter; model-overlays guards the coverage half) The completeness section shrank by 3 characters (5 bytes), matching the 89 pins lowered by 3. The default-copy test pinned only the minimal-code half; an expectMentions check now fails if the full-coverage or never-excuses-a-gap clause is dropped. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: pre-landing review fixes (model-overlays binds smaller diff to equal coverage) The default-copy check now also fails if 'on equal coverage' is dropped from the smaller-diff rule (which would tell models to prefer the smaller diff unconditionally) or if the no-impossible-state-checks clause is removed. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG lists the model-overlays coverage checks The [Unreleased] Tests bullet now says the default-copy test checks both halves of the rule (full coverage of the requirement; smaller diff only at equal coverage; no impossible-state checks; never excuses a gap), matching 88e3cc0 and 44eacaf. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: plan-ceo-review and ETHOS say completeness means coverage, not more code Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: pre-landing review fixes (CEO bullet bounds code by accepted scope; tests bind each completeness rule) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: cover the shared-preamble hoist (ship Step 7) Adds test/preamble-hoist.test.ts (tier 1 unchanged; tier 2+ SKILL.md plus sections/preamble.md reproduces the pre-hoist preamble; pointer position and id) and two gen-skill-docs-checks cases (one shared preamble per host, beside every pointer and named by its id; preamble artifact kind). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: pre-landing review fixes (ship Step 9.4) - setup: follow a symlinked sections/ dir in the carved-pointer status check - touchfiles: select preamble-hoist.ts alongside every preamble.ts entry - section-manifest: exempt sections/preamble.md only for skills whose SKILL.md carries the shared-preamble pointer - skill-size-budget: growth ceiling measures SKILL.md plus its shared preamble - host-conformance: tolerate a missing sections/ dir, as setup does - skill-validation: read bin/gstack-skill-start directly - skill-fixture: error messages and docblocks name the actual boundary (harness-version non-behavioral note updated) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: describe the size-budget growth check's startup measure (ship Step 9.4, pass 2) The growth ceiling now measures SKILL.md plus the shared preamble, the same measure as the shrink floor; the docblock said SKILL.md only. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG says simplification.md quotes the old rule and CLAUDE.md restates the new one Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG lists the hoist tests and the growth-ceiling measure (ship Step 9.4) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: cover the symlinked sections/ status check and the ETHOS house-rule quote (ship Step 7) host-conformance: --status still finds a broken bare sections/ pointer when a carved Claude skill's installed sections/ is a directory symlink (fails at the pre-repair 0e5d11f). gen-skill-docs: the house rule review/specialists/ simplification.md quotes from ETHOS.md still appears in ETHOS.md. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: describe the shared preamble file (ship Step 14.5 docs audit) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: the shared-preamble skip needs the skill's own Read; digest and first-run intro say complete coverage, minimal code (GSTA-96) The pointer skipped the shared file when any text carrying its id was in context, so a page or file with that heading could stand in for the anti-spoofing rules (security, red team and adversarial review, ship run 2). It now requires the skill's own earlier Read. The digest's Boil the Ocean bullet and the first-run intro still said 'AI makes completeness cheap, so do the complete thing' (docs audit blocker); both now carry Garry's 'complete coverage, minimal code' (2026-10-25: '1/ yes fix this'). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: pin the shared-preamble provenance skip and the new Boil the Ocean wording (ship Step 7) preamble-hoist: on every host, tier and render option the pointer allows the skip only when the skill's own Read is in context. agents-digest and gstack-skill-start: the digest bullet and the lake intro say complete coverage, minimal code, never the retired phrasing. Each fails at 9af36d0. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: describe the provenance skip where the docs still described the id-only one (ship Step 9.4) The preamble-hoist.ts docblock and ARCHITECTURE.md still said the shared preamble is re-read only when its id is absent; since 2acf1a4 the pointer requires the skill's own earlier Read of the file. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG lists the provenance-pointer and Boil the Ocean wording tests (ship Step 13) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * revert(digest): keep the agents digest at main's bytes; an unstamped head cannot change a release output (GSTA-96) CI 'Check VERSION is not stale vs queue' classified #3434 as stamped because it touched agents-digest/gstack-AGENTS.md (a release_outputs path), while regenerating only the generator fails the digest freshness test. The first-run intro rewording stays. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * feat(preamble): office-hours' Q<N> form is its own subsection, so it stops copying the shared AskUserQuestion subsections (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: pin the added-subsection note and readSkillEffective's changed-plus-added order (ship Step 7, GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: variantSubsections and readSkillEffective docblocks name added subsections; office-hours cap history ends 'Before: 1.2279' (ship Step 9.4) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG says the Q<N> lead-in gained the Layout line text instead of claiming no wording changed (ship Step 11) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG gives office-hours' measured SKILL.md drop, 6,333 bytes (ship Step 13) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: CEO completeness posture check covers the accepted-scope bound Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: CHANGELOG says plan-ceo-review grows about 110 characters and each skill loses 3 from the section (cherry-picked from #3425's 25eec73; the digest/lake-intro line is dropped because #3434 rewords the intro and its own bullet names the digest deferral) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: voice judge reads the hoisted preamble; register ETHOS.md on its reader cases Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: refresh the claude ship golden the main merge took from main (ship Step 5 preflight) The 1a4dcaf merge resolved test/fixtures/golden/claude-ship-SKILL.md to main's pre-hoist bytes. The golden is a copy of ship/SKILL.md; regenerated with bun scripts/gen-ship-goldens.ts (pregate regen: REGEN_STALE). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: pr-dependencies' unplaceable-file example uses SECURITY.md now that ETHOS.md is mapped 6183ca8 registered ETHOS.md on its reader cases in touchfiles-data.ts, so a PR touching only ETHOS.md plans in pr mode and no longer restores the full gate. The fail-closed test and the real-unknown label check now use SECURITY.md, which is tracked and unmapped; the assertions are unchanged. CHANGELOG names the third test file. (CI free shard 7 at 7c21720; remote pregate shard 6.) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: host path-leak checks read the moved shared preamble too; name SHARED_PREAMBLE_FILE (ship review pass 2) The external-host path-leak assertions in gen-skill-docs.test.ts (Codex, Factory) and host-config.test.ts (Copilot, Antigravity) read SKILL.md alone, so a Claude path that moved into sections/preamble.md with the hoist went unchecked. They now read the skill's startup text through readSkillStartup. workflow-judge-input and host-conformance use the SHARED_PREAMBLE_FILE constant instead of a bare 'preamble.md' literal, and preamble.ts's header describes the hoisted composition. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG token savings name the per-skill skip and the autoplan follow-up (ship review pass 2) The pointer's skip is bound to the skill's own copy of sections/preamble.md, so "every later one saves about 5,800" holds per skill, not per session, and /autoplan's skip list and the inline /office-hours loader do not yet skip the pointer section. The entry now says so instead of claiming the full /autoplan saving. Adding the skip entry grows size-pinned skills, so it is a follow-up. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: corpus budget measures startup bytes; carve-safety counts shared preamble as uncarved (ship review pass 2) skill-size-budget's growth ceiling and shrink floor already measured SKILL.md plus sections/preamble.md, but the total-corpus check still summed hoisted SKILL.md bytes against the pre-hoist baseline, hiding about 0.9 MB of headroom. It now sums the same startup bytes (3,114,885 -> 2,590,484, x0.832); the unreachable SKILL.md-missing fallback is gone. auq-format-always-loaded's carve-safety guard treated any sections/ dir as carved, so every skill with only the shared preamble ran it and its sanity check passed on preamble.md. It now uses setup --status's rule: a section other than the shared preamble. Those skills keep the per-skill checks. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: remaining SKILL.md-only negatives read the shared preamble too (ship review pass 3) The parameterized external-host path-leak smoke test, the two Codex .claude path checks and both vestigial-handshake negatives read SKILL.md alone, so text that moved into sections/preamble.md went unchecked. They now read readSkillStartup. preamble.ts's header no longer says tier 1 renders through generatePreamble(). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: ratchet skeleton caps and eager context ceilings down after the hoist (ship review pass 3) The hoist moved about 22 KB out of every tier 2+ SKILL.md, which left each hoisted maxSkeletonBytes cap and each eagerPerInvocation ceiling roughly 22 KB / 5-6K tokens slack. The 20 hoisted skeleton caps now sit at the post-hoist skeleton plus their pre-hoist headroom, and 42 context-budget ceilings take the recaptured (lower) values. Nothing was raised: the four ceilings the recapture would have raised keep their old values. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: preamble-hoist's section splitter closes a fence only on a bare delimiter (ship Step 11) Inside an open fence, a line such as "```bash" is content in CommonMark, but splitSections treated it as the closing fence, so a later "## " line in the same block could split a section mid-code-block and hoist half of it. A closing delimiter now needs nothing but whitespace after it. Every host's render is unchanged (gen-skill-docs --dry-run FRESH on all twelve hosts). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: skill-startup.ts is a global, pinned harness file like skill-fixture.ts (ship review pass 3) skill-fixture.ts is a global touchfile pinned in scripts/harness-version.json because it reshapes the skill content most E2E suites read, and it now builds that content through test/helpers/skill-startup.ts. That helper was in neither list, so a later edit to it would change fixture inputs without the harness-version guard and select only 27 keys. It is now global and pinned, recorded non-behavioral (its fixtures are byte-identical to the pre-hoist output). ARCHITECTURE.md and the CHANGELOG stop calling the pointer id a content id: it hashes the shared text before host rewrites. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG matches the merged skip lists and the baseline diff (ship review, run 4) The headline still called the autoplan and inline /office-hours skip entries a follow-up, but the main merge added them; it now says the loaded skills skip their shared Read. The #3451 sentence is dropped, because that fix shipped in v1.91.237.0 and is recorded there. The typecheck baseline line says the TS2769 entry was re-keyed for the 'preamble' artifact kind with its count unchanged, not fixed and lowered. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: pin the pointer entry in the loader skip lists; drop a carve-safety tautology (ship review, run 4) A new INVOKE_SKILL test checks that plan-ceo-review's inline /office-hours skip list and autoplan's section skip list both name the shared-preamble pointer heading, and that office-hours and the four plan reviews carry it. The carve-safety "sanity" assertion repeated the predicate discoverSkills() now uses to set sectionsDir, so it could never fail; it is removed. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * fix: gstack-doctor finds the Codex model patch in the hoisted shared preamble (ship Step 11, run 4) The doctor's patch row grepped only the Codex review SKILL.md for the "Model-Specific Behavioral Patch (" heading. The hoist moves that section into sections/preamble.md, so the required row (for ship, review, the plan reviews, office-hours, investigate and more) would fail on every correct post-hoist Codex install and block install.sh's "gstack: ok". It now checks the SKILL.md and then its shared preamble and names the file it found. A new doctor test covers a hoisted render. preamble-hoist's variantSubsections also returns null when no subsection differs, instead of emitting a replacing heading with no subsections. Every host's render is unchanged. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test(docsync): fixtureDocs installs the generated sections/preamble.md (hunk taken from #3472) A generated skill that loads the shared preamble names sections/preamble.md as a required asset before skill start, and a spawned document-release child blocks on its absence, so docsync-spawned stayed red on every #3434 screen. fixtureDocs now copies that file when the generated skill has it. This single hunk is taken unchanged from gstack#3472 (head 2e3f7ba, GSTA-107), which trains ahead of #3434, so the train merge sees identical lines. Nothing else from #3472 is included. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG notes the docsync fixture hunk taken from #3472 Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test: pin readSkillEffective's added-subsection change in harness-version.json as non-behavioral (CI quick-bun-default) skill-startup.ts changed in this PR so readSkillEffective appends a variant's added ### subsections. Its output is byte-identical to the pinned helper for every tier 2+ render without added subsections (735 across all hosts) and for office-hours' pre-change SKILL.md, so no pass-rate series resets. CHANGELOG Tests line updated. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG and the office-hours cap note give the post-merge measure (6,328 bytes; union 160,952 -> 154,624) (ship Step 11, #3453) The 6,333-byte drop and the 160,989 -> 154,653 union were measured at 5db328f, before main was merged; against this PR's base b727a6d the numbers are 6,328 bytes and 160,952 -> 154,624. The cap stays 1.1796. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: round the post-merge office-hours ratio correctly (154,624 / 131,110 = 1.1793) (ship review pass 2, #3453) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs: CHANGELOG names the post-merge context-budget ratchet (captain hoisted) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): correct #3434's post-merge allowlist count and the two context-budget rows above main (GSTA-96, per GStack Minder) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): restate the corrections to #3434's post-merge bullets after its v1.91.265.0 merge (GSTA-96, per GStack Minder) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test(auq): planted skill fixtures resolve the shared preamble like an install (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test(plan-format): mode menu delivers its question without output-harvesting wording (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): record the v1.91.270.0 merge values (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): office-hours context-budget ceiling after the v1.91.270.0 merge (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * test(auq): parallel-capture contract expects the planted skill with its sections path mapped (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): correct the v1.91.270.0 recapture wording for hoisted rows (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): record the v1.91.273.0 merge re-measurements (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): restore the released entries the previous commit truncated (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> * docs(changelog): end two sentences the prose gate flags (GSTA-96) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com> --------- Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
This was referenced Oct 27, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Includes #3425's completeness rewording: #3425 was closed without merging, and its commits are part of this branch.
What changed
review/SKILL.md startup, before and after. The shared preamble moves to
sections/preamble.md, which a session reads once. Each SKILL.md keeps a short pointer plus the sections that name the skill.Every tier 2+ SKILL.md used to inline the same ~23 KB preamble: AskUserQuestion format, voice, writing style, completeness, completion protocol and so on. So each invocation re-read text the conversation already held. That text now renders once per host into
sections/preamble.mdbeside each tier 2+ SKILL.md.The SKILL.md's new Shared preamble (load first) section names the file and its id (a hash of the install root and the shared text before host rewrites). It asks for the Read unless the skill's own earlier Read of that file, headed by that id, is still in context, so other text that merely carries the id cannot stand in for the shared rules (a /ship review fix). A section or subsection with a shared heading replaces the shared one for that skill.
scripts/resolvers/preamble-hoist.tsmoves a section only when this skill's render of it is byte-identical to the shared render. Sections that name the skill stay inline. They are the start command, Question Tuning, Repo Ownership, Search Before Building and Telemetry. Text without its own heading and skill-specific variants also stay inline. A variant that keeps the shared###headings carries only the subsections that differ. Tier 1 skills are unchanged.extractSkillBody,extractSkillSections) produces byte-identical output before and after.gen-skill-docsemits the file for every host as a newpreambleartifact kind, with the same host, install-root and prelude rewrites as SKILL.md.setupalready installs every skill'ssections/, and overlay and per-install renders carry the file the same way; the install tests cover both.setup --statusnow treats a baresections/<file>pointer as carved only when the skill installs a section other thanpreamble.md.Tokens (Anthropic count_tokens, claude-sonnet-4-5; before = #3425's tree)
The shared file is 5,953 tokens.
Measured at 4da2eba, before the main merges. The skip is tied to each skill's own copy of the shared file, so the saving applies to repeat invocations of the same skill; the first invocation of each different skill pays the Read. One /autoplan run reads fewer tokens because its review-skill methodology loads no longer repeat the preamble, and /autoplan and the inline /office-hours loader skip each loaded skill's pointer section. A CEO+Eng run goes from 110,676 to 103,259 tokens (−7,417). A run over all four phases goes from 191,871 to 173,930 (−17,941, −9.4%). That replaces the autoplan-specific change this thread first planned.
office-hours and plan-eng-review cost more on first load because they keep their own versions of some shared sections. A follow-up PR, already approved, dedupes the copies that match the shared text.
Approvals
fixtureDocshunk of #3472 into this PR so docsync-spawned can passubicloudrow at its capture, 3,594. Sources:test/helpers/capture-context-budget.tsat merge base b004c06, d402d44, main 6c5b2d4 and HEADdesign-motionkeeps main's 10,409;eval-plan4,396 → 4,842 (this PR's ceiling plus #3493's measured growth, 3,996 → 4,442 tokens);gstackback to main's 5,326;alwaysOnTotal7,546 → 7,542. Sources:test/helpers/capture-context-budget.tsbuildRatchetBill at the merge base 6c5b2d4, 62ec2e6, main a50cf58 and HEADTests and gate integrity
test/helpers/carve-guards.tsraises the caps for ship (1.6768), review, autoplan, spec, land-and-deploy, design-consultation, plan-devex-review, plan-ceo-review, office-hours and plan-eng-review to their measured unions. Each cap carries its reason.maxSkeletonBytescaps drop to the new skeleton plus each one's pre-hoist headroom (ship 81,600 → 59,771). 42eagerPerInvocationceilings intest/fixtures/context-budget.jsonratchet down; the 4 the recapture would have raised keep their old values. The size allowlist is re-pinned at the merged head, and no pin is raised against main or our previous head (the merge offsets main's +14 characters in plan-eng-review by dropping two shared-only skip-list entries).test/skill-size-budget.test.tsmeasure SKILL.md plus its shared file (corpus 3,114,885 → 2,590,484 bytes). Tests that check preamble rules, host path leaks, the vestigial handshake, E2E fixture extractors, judge bundles and install checks read SKILL.md together with its shared file throughtest/helpers/skill-startup.ts.test/helpers/skill-startup.tsis a global touchfile pinned inscripts/harness-version.json(non-behavioral: fixtures are byte-identical), becauseskill-fixture.tsbuilds its fixtures through it.gstack-doctor --check's requiredpatchrow now finds the Codex model patch insections/preamble.md, where the hoist moves it (otherwise every hoisted Codex install would failinstall.sh --for ship); a doctor test covers a hoisted render. The hoist's section splitter closes a fence only on a bare delimiter, and an all-whitespace variant stays whole. A new test pins the pointer entry in both loader skip lists.test/helpers/docsync-fixture.ts'sfixtureDocscopies the generatedsections/preamble.mdinto the document-release and ship fixtures (one hunk, unchanged, from v1.91.249.0 fix(evals): repair five paid-case flakes at their causes (GSTA-107 daily flaky pass) #3472 at 2e3f7ba, which trains ahead of this PR).scripts/compare-pr-version.ts(offline) reports this head unstamped: VERSION and package.json equal the merge base (1.91.247.0), the regenerated digest matches generation at that VERSION (v1.91.237.0 fix(version-gate): a release output regenerated at the base VERSION is not a stamp #3451), and the CHANGELOG change is confined to [Unreleased]./ship (crew rules 22 and 23), unattended, at bfe904e
Shipping with these named items (crew rule 23), not fixed here:
sections/unattended.mdnames a file the host install doesn't ship.setup --statusdecides carved status from files on disk; runtime-root copies of office-hours and plan-design-review hold only SKILL.md; the pointer id repeats across hosts; no generator guard against body headings that collide with shared ones or asections/preamble.md.tmpl; stalesections/preamble.mdin a skill that drops to tier 1 is not caught;paid-select.tsandsetuphard-codepreamble.md; the generator'sSKILL_BYTE_CEILINGmeasures SKILL.md only; "rules below" wording in the start block; whole-###variant duplication;readSkillEffectivecopies the variant note's wording; the helper splitters still use the old fence rule; the doctor fallback is not tied to the pointer and names only SKILL.md on failure; the removed skip-list entries matter only for a future tier 1 INVOKE_SKILL target.Re-READY after train 14 (b727a6d)
Train 14 dropped this PR on #3472's companion test (
test/ship-document-release-dispatch.test.ts:249): under the hoist the generated ship skill also hassections/preamble.md, so "nothing when the skill has none" named a skill that has one. After #3472 merged (v1.91.249.0), main was merged in under GStack Minder's rule 19 grant (CHANGELOG additive, no other conflict), and the test now removes ship's copy from its tree before asserting it gets none; the fixture design (install exactly when the generated skill has one) is unchanged. Focused: ship-document-release-dispatch plus the docsync authority, lifecycle and grammar files, 292 pass / 0 fail. Remote free tier at tree 8c141edf8ec0: 19,604 tests, 0 failed ($0.07). Rule 23 delta review (pass 3 of 3) over the non-main delta (the test line and its CHANGELOG sentence): no findings.Re-READY after train 15 (d402d44)
Train 15 refused this PR on a real conflict in
test/fixtures/context-budget.jsonwith #3480 + #3478. Under GStack Minder's rule 19 grant, main v1.91.255.0 was merged in: CHANGELOG additive; context-budget.json by hand (canary keeps its hoisted ceiling, since main's only change there was the adjacent newcaptainrow; captain, now hoisted, drops 12,011 to 6,551 and five loaded-skill ceilings drop by 2 to 22, all to a fresh capture; nothing raised). Generated files and ship goldens regenerated. Focused: context-budget ratchet, size guard, parity, carve, preamble-hoist, gen-skill-docs checks, host-config, section manifest, harness version, touchfiles, ship-document-release-dispatch, skill-size budget, module size (485/0), plus captain, gen-skill-docs, skill-validation and regen tests (862/0). Remote free tier at tree 14cc2a7f31bf: 19,670 tests, 0 failed ($0.06). Rule 23 delta pass over the non-main delta (the context-budget rows and one CHANGELOG sentence): no findings. #3433's design-review parity cap (decision A, measured union under the hoist) waits for #3433 to land on main.Re-READY after train 17 (62ec2e6)
Under GStack Minder's rule 19 grant, main 6c5b2d4 (v1.91.260.0, with #3433 GSTA-92 phase 2 and #3481 review convergence) was merged into d402d44. Conflicts and resolutions (decision A extended):
test/helpers/carve-guards.ts: ship cap 1.7172 with skeleton 60,913, review 1.3136, design-review 1.3758 with skeleton 115,947. Each value is a measured union or skeleton at this head (re-measured by /ship: ship 322,312 and 60,848; review 142,550; design-review 145,613 and 115,947).test/fixtures/skill-md-size-allowlist.json: the hoisted pins plus all 31 pin raises main made (23 in v1.91.260.0 feat(design-review,plan-design-review): craft audit and Persuade plan checks; craft tells as checker-only catalog entries (GSTA-92 phase 2) #3433, 8 in v1.91.256.0 feat(ship): review loop converges (frozen scope, 3-pass cap, NO FINDINGS) and receipts carry across conflict-only merges #3481); every pin is the exact character count the size guard checks.test/fixtures/context-budget.json: seven values raised by exactly main's measured capture growth (recaptured at all four trees by /ship and matched row by row), and the new hoistedubicloudrow at 3,594. Graded against this PR's previous fixture, onlycapy/skill/gstack(3,740 > 3,498) and the new skill strictly need a change; the others keep main's growth as headroom. Every row stays below main's own value exceptgstack(5,346 vs 5,326) andalwaysOnTotal(7,546 vs 7,458), which main never raised.test/helpers/touchfiles-data.ts: the union of both sides (design-craft.ts and preamble-hoist.ts both kept). Main's newubicloudskill regenerates in hoisted form (ubicloud/sections/preamble.md).Checks at tree 5f44b97020be: regen and goldens fresh,
bun run buildleaves the tree unchanged, harness version 6 with no drift, offline version gate reports the head unstamped (VERSION 1.91.260.0 = merge base), focused guards 101/0 (parity, size guard, context-budget ratchet, carve guards, touchfiles, skill-size budget, preamble-hoist). Preflight passed. Remote free tier: 19,775 tests, 0 failed, 690 files, $0.06 (four earlier attempts hit the shared Ubicloud vCPU quota, $0.01 in total; no VM left behind). No paid case reads the merge's own product delta, so thevoice directive tonePASS stands as the paid evidence. The five macOS/Windows lanes clear from CI at this head.Review (rule 23, delta pass over the non-main delta only,
git show --remerge-diff; testing and maintainability specialist plus the Claude adversarial pass):REVIEW_LOOP: passes=1 blocking_found=0 deferred=0 noted=4 fixed=0 scope=frozen. Noted, not fixed (below the floor):gstackandalwaysOnTotalsit 20 and 88 above main's unchanged values, and the earlier bullet's "No ceiling went up" now reads against it.Per GStack Minder, this PR ships at 62ec2e6 as is: the two CHANGELOG wording advisories above (31 raises; gstack and alwaysOnTotal rows above main's unchanged ceilings by the measured growth) are corrected in #3453's CHANGELOG entry when #3434 is re-merged into it, not by a new tree here. The
gstackrow is now also resolved here, because train 19's merge put it back at main's 5,326.Re-READY after train 19 (f45f1ad)
Under a new GStack Minder rule 19 grant, main a50cf58 (v1.91.265.0: #3491, #3483, #3484, #3430 /design-motion, #3493 ETHOS principles) was merged into 62ec2e6. Only CHANGELOG and
test/fixtures/context-budget.jsonconflicted:## [1.91.265.0]down is byte-identical to main.test/fixtures/context-budget.json: the union of both sides. Main's newdesign-motionrow stays at main's 10,409 (per the Captain);eval-plan4,396 → 4,842, this PR's ceiling plus the growth v1.91.265.0 docs: absorb four pstack principles into ETHOS.md, the observable-questions rule and candidate-side eval blinding (GSTA-124) #3493 added (ratchet bill 3,996 → 4,442 tokens);gstackback to main's 5,326 (actual 4,863);alwaysOnTotal7,546 → 7,542 (actual 7,398.5 on the ratchet bill). Every row passes at this head, and every per-skill row is at or below main's value. Main's newdesign-motionskill regenerates in hoisted form (design-motion/sections/preamble.md); its carve guard holds (skeleton 18,080 ≤ 39,900, union 59,960 ≥ 57,500).Checks at tree 2c836abf2643: regen and goldens fresh, harness version 6 with no drift, offline version gate reports the head unstamped (VERSION 1.91.265.0 = merge base), focused guards 20/0 (context-budget ratchet, parity, size guard, carve completeness, preamble-hoist, skill-size budget). Preflight passed. Remote free tier: 19,854 tests, 0 failed, 693 files, $0.09 (two earlier attempts were refused at create time by the shared Ubicloud vCPU quota, $0; no VM left behind). The only paid case on the merge's product delta,
design-motion-showcase(marathon), is deferred to broad coverage, so thevoice directive tonePASS stands. The five macOS/Windows lanes clear from CI at this head.Review (rule 23, pass 2 of 3, the non-main delta only; Claude adversarial pass, specialists skipped for a delta under 50 lines): NO FINDINGS from the reviewer;
REVIEW_LOOP: passes=2 blocking_found=0 deferred=0 noted=7 fixed=0 scope=frozen(this pass: noted 3). Noted, not fixed (below the floor):alwaysOnTotalis the rawlib/context-billbuildBill figure (it counts test fixtures and the connect-chrome alias); the ratchet grades 7,398.5. The value passes; 7,542 is still 84 above main's 7,458, so the earlier "All stay below main's pre-hoist values" still does not coveralwaysOnTotal.design-motionkeeps main's unhoisted 10,409 against a hoisted actual of 4,295 (about 6,100 tokens of headroom), by the Captain's choice.eval-plan's raise uses raw-token growth (+446) where train 17's rows used capture growth (+491 here); both are measured growth, and the row passes with 400 tokens of headroom.GStack Minder approved eval-plan at 4,842 under the same reading as the other rows (ceiling plus main's measured growth, below main's 9,811). Flaky rerun (rule 16): dispatched Free Tests run 40444043418 attempt 1 failed in test/ubicloud-runner.test.ts (main's #3482 timing cases: watchdog exit 0 vs 124, then a quota-wait 2s vs 3s message on serial retry); the PR-triggered Free Tests run at the same head passed, the file passes locally, this PR does not touch scripts/ubicloud; attempt 2 passed. Logged in the crew flaky ledger.
Re-READY after train 20 (234227f)
ATTRIBUTED #3434 under docs/captain.md § Merge trains, with the bisect replaced by this evidence: train 20's gate (run 40454750574) failed auq-format-gate on this PR, and train-bisect broke on
${PREFIX_SHA:0:12}undersh -e(run 40456150409).Cause: the fixture never let the model read the hoisted preamble. The planted SKILL.md points at
~/.claude/skills/gstack/<skill>/sections/preamble.md, and the capture prompts forbade reading anything outside the planted file, so the question format the shared preamble defines never loaded. Repair (5aa7765, fixtures only): the AUQ capture helpers and bothsetupPlanDirs map~/.claude/skills/gstack/<skill>/sections/to the planted copy'ssections/, copy that directory, and allow reads of the files in it. The parallel-capture contract test now expects the mapped path (c8ed25f):test/auq-parallel.test.tsstill checks the planted bytes exactly, after the same mapping; it had failed 31 cases at 191a44c (CI free-suite shard 3 and the remote free tier).Mode menu (e1e2ce1): plan-ceo-review-format-mode now uses
approachQuestionInstruction, the D2 wording the approach menu already used, because its two reds on the fixture repair were refusals at the capture Write (reasoning_extraction).test/plan-format-approach-prompt.test.tscovers both menus.Evidence limit: the probes read SKILL.md with Read, not through Claude Code's Skill tool; 2 of 2 runs loaded
sections/preamble.mdbefore the first question once nothing forbade it.Screen miss: screen run 40269992077 (tree 9dc1dfb) had already failed auq-format-gate with the same message and the READY went out over it; train 20's $29.67 gate is the cost (CREW.md "a red screen is not READY").
Hosts: external renders point at
sections/preamble.mdrelative to their own installed skill dir; test/host-conformance.test.ts, 23/0.Native-runner captures record billing unavailable.
Under a GStack Minder rule 19 grant, main f27a760 (v1.91.270.0: #3494, #3496, #3487) was merged (501353e). CHANGELOG and
test/helpers/carve-guards.tsconflicted:mergeChangelogs; one[Unreleased]above[1.91.270.0], and everything from## [1.91.270.0]down is byte-identical to main.test/fixtures/context-budget.jsonis recaptured whole withbun test/helpers/capture-context-budget.ts; all 66 rows reproduce on HEAD. The non-hoisted rows that rose match main's own recapture at f27a760 exactly (alwaysOnTotal7,769,capy/skill/gstack4,115,diagram4,286,gstack5,350,gstack-upgrade6,297). The hoisted rows that rose (eval-plan4,887,landing-report3,269,plan-tune9,215,retro13,484,ship15,913) are this branch's own recapture with main's growth included; main measures them unhoisted (for exampleeval-plan10,581).design-motionfalls to 4,724 now that it is hoisted.Under a second GStack Minder rule 19 grant, main a1934a1 (v1.91.273.0: #3509, #3500, #3501) was merged (21b8155). The CHANGELOG went through
mergeChangelogs.test/helpers/carve-guards.tsand the size allowlist took main's lines, and then every cap the hoist touches was re-measured on the merged tree:test/fixtures/context-budget.jsonis recaptured, and all 66 rows reproduce on this head. The non-hoisted rows that rise (benchmark8,043,capy/skill/gstack4,120) equal main's own recapture at a1934a1. The hoistedcontext-restore(6,287) andinvestigate(7,795) sit below main's 11,980 and 13,489.96c82c4 recorded those values in a CHANGELOG bullet, but it also truncated CHANGELOG.md by writing back a partial read. CI's version gate caught it, and fa5fe86 restores the file. Everything from
## [1.91.273.0]down is byte-identical to main a1934a1. 6578bcb ends two sentences that main's new prose gate flagged (gstack-prose-lint CHANGELOG.md --unreleased --gate: ok).Under a third GStack Minder rule 19 grant, main d87eec9 (v1.91.274.0, #3507) was merged (234227f). The CHANGELOG went through
mergeChangelogs, and everything from## [1.91.274.0]down is byte-identical to main.test/helpers/carve-guards.tstook main's design-review and design-html lines, and a cap was re-pinned only where it failed or where a hoisted skill had taken an unhoisted line:test/fixtures/context-budget.jsonis recaptured, and all 66 rows reproduce on this head. Only design-html moved, 8,926 → 8,896. Main's own recapture of that unhoisted row is 14,589.Checks at tree 9197f225175b: harness version 6 with no drift. The offline version gate reports the head unstamped (VERSION 1.91.274.0 = merge base, CHANGELOG change confined to
[Unreleased]). The size guard passes 2/0, and the prose gate on[Unreleased]is ok. Reused from the parent at this head: regen fresh, typecheck clean, and 29 focused guard files 1,221/0 (includingtest/auq-parallel.test.tsandtest/host-conformance.test.ts). Preflight passed. Remote free tier: 19,965 tests, 0 failed, 704 files, $0.12, and no VM left behind. The first attempt never got a VM. It waited 10 minutes for Ubicloud capacity, was refused withPREGATE_REMOTE_INFRA, and its VM was destroyed before it ran. At 6578bcb (tree 4872a8ea0dd4) it ran 19,976 tests, 0 failed, 704 files, $0.11. At c8ed25f (tree fbf7fec0c8d7) it ran 19,917 tests, 0 failed, 701 files, $0.07. The run before that, at 191a44c (tree a840d45a342e), reported 1 failed file,test/auq-parallel.test.tswith 31 cases, which reproduced serially ($0.06). The five macOS/Windows lanes clear from CI at this head.eval:select (profile pr, full-fallback): shared runtime inputs restore every gate case and judge, and none is a PR pass. The changed helpers' paid consumers in the table above passed. The other consumers of
setupSkillDir(the carve-section-loading cases, plan-ceo/ship/office-hours section-loading, office-hours-design-draft) are deferred to broad coverage. Their read detection matchessections/<file>.mdby basename, so the mapped path does not change what they grade. No further paid case was run. After the v1.91.273.0 merge, eval:select also defers main's two new carve cases, carve-section-loading-benchmark and carve-section-loading-context-restore, to broad coverage. It names no new PR pass.Review (rule 23, pass 3 of 3, the non-main delta f45f1ad..191a44c; GPT adversarial pass, specialists skipped for a delta of 5 fixture files plus merge resolutions): no blocking findings.
REVIEW_LOOP: passes=3 blocking_found=0 deferred=0 noted=8 fixed=0 scope=frozen(this pass: noted 1). The review did not catch thetest/auq-parallel.test.tsfailure; CI and the remote free tier did, and c8ed25f fixes it. That commit came after the pass-3 record and was read by hand. It is an expectation-only change that keeps the exact byte check. It is not counted as a fourth pass. The delta since c8ed25f is two main merges, test fixtures and helpers (carve-guards.ts, the allowlist,context-budget.json) and the CHANGELOG. The pass cap is spent, so no fourth pass ran or was logged.gstack-ship-receipt carry --dry-runfrom the f45f1ad receipt says the product diff is unchanged ("would carry from f45f1ad to 234227f (3 base merges + test-only change + release metadata)"). The new caps, allowlist pins, recapture and CHANGELOG byte-identity were each re-measured by hand, as listed above. Noted, not fixed (below the floor):eval-plan4,887. That holds for the non-hoisted rows only; main's recapture giveseval-plan10,581. The values themselves reproduce; wording only.Issues
Refs #3425
Refs #3433
Refs #3434
Refs #3451
Refs #3472