Repository navigation
v1.94.29.0 R56: everything in gstack earns its place (principle, evidence gates, inventory) - #3837
Merged
Merged
Conversation
prereg --check now refuses a loaded comparison whose ## No-gstack arm section lacks masking:, install: or audit:, or says none for one of them (PREREG_NO_GSTACK_ARM). On Capy, run agents inherit the instruction to load the user-drive gstack skill, so the arm hides both Capy drives, checks that no gstack is installed, and audits drive reads; an arm that read a gstack skill file is contaminated and gets zero credit (TESTING_INTERNALS.md #no-gstack-arm). The prereg template, /eval-plan, troubleshooting and the two passing prereg fixtures carry the new lines. eval-plan's paragraph was dieted to hold its context-budget ceiling. Evidence: safety: keeps the no-gstack control uncontaminated (Captain order, FOUN-32 via GGST-6) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…VIDENCE docs/inventory/evidence.json holds one row per inventory key: the skill folders discoverTemplates() finds (root skill as gstack, symlinked folders excluded) plus seven shared parts, 71 keys. lib/inventory.ts validates each row by its status's own fields, with named codes (LEDGER_*), and joins the ledger with lib/context-bill.ts buildBill. bin/gstack-inventory writes docs/inventory.md, --check validates (exit 1 with codes), --json prints. Paid dispatches per run for /ship and /autoplan are quoted from their own docs and found by line, so a moved citation fails --check (INVENTORY_SOURCE_MISSING). Every current row is grandfathered and seeded from GEVA-2's table (org-gstack-gbrain/research/gstack-part-evidence.json): in-flight fair tests and proposed registry rows as plans, earlier unfair results under prior with why (stub, unmasked, ceiling, ...), five safety rows. gstack-skill-bill prints the skill's EVIDENCE line. Evidence: plan none (owner ruling R56, GST-914); the ledger itself is the measurement record Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Garry, 2026-11-02: "Also work this principle of only include it if it is beneficial and good into gstack". ETHOS.md principle 9 states the rule, the fair test (claude_code preset, frontier models per R54, masked no-gstack arm), the R48 Why-null cycle, the safety exception, and measure biggest first, linking docs/inventory.md. factory.md item 8 cites it, trimmed elsewhere to 3,298 of 3,300 bytes with no rule dropped. /write-skill's edit ladder step 2 is the evidence rule and section 4 adds the Evidence: commit line. /autoplan's CEO phase lists each part a plan adds to gstack with its evidence or owned, dated plan; two redundant lines went to keep autoplan under its size pins (pins lowered by 17). CONTRIBUTING.md "Adding a new skill" and docs/skills.md point at the ledger. Skill-Size: 3273→3666 bytes (write-skill/SKILL.md.tmpl) Skill-Size: 12738→12721 bytes (autoplan/sections/ceo-phase.md) Evidence: plan none (owner ruling R56, GST-914); write-skill and autoplan rows stay grandfathered untested Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…-bill EVIDENCE Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Ledger rows carry aliases (the name GEVA-2's table uses where it differs from the key, such as upgrade for gstack-upgrade), and docs/inventory.md lists them beside the installed-name rule. Unmapped rows are labeled as evidence-table rows, and the ranking notes that a skill's on-demand files can also be a shared row's. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
docs/inventory.md is generated by bin/gstack-inventory at the previous commit: 71 parts ranked by load per run, 0 helps, 0 no-effect, 0 hurts, 16 untested with a plan (4 fair tests under way, 12 proposed), 50 untested with plan none, 5 safety. Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
|
😎 This pull request was merged. |
…(GST-814) 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>
…t's row is checked (review fix) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…6-11, R56-12) 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>
…nto a gstack install, and the ledger checks aliases, r48 and prior.why (review fixes) Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
…easured 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>
… install guard Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
Conflict: test/fixtures/skill-md-size-allowlist.json took main's file; the 12 autoplan host pins are lowered to the guard's measured sizes (main minus 17 each, none raised). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
3 tasks done
Conflict: test/fixtures/skill-md-size-allowlist.json took main's file; the 12 autoplan host pins are lowered to the guard's measured sizes (main minus 17 each, none raised). Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
4 tasks done
This was referenced Nov 3, 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.
Garry, 2026-11-02: "Also work this principle of only include it if it is beneficial and good into gstack" (R56, GST-914). Plan approved at autoplan-20261102030915 gate_rev 2 ("Do it", relayed by GStack Minder).
What changes
claude_codepreset, frontier models (R54) and a masked no-gstack arm. A fair null gets one R48 "Why null?" cycle, then keep, fix or cut. Safety guards keep their stated reason. factory.md item 8 points at it, still inside its 3,300-byte budget.Evidence:commit trailer. /autoplan's CEO phase lists each part a plan adds and raises a finding when one has neither.docs/inventory/evidence.jsonholds one entry per part: 64 skills plus 7 shared parts. Each entry has a status (helps, no effect, hurts, untested or safety), and the required fields depend on the status. Every part present today is grandfathered. A new part can't pass onplan: none.bin/gstack-inventoryvalidates the ledger (--check) and writesdocs/inventory.md, a dated ranking by load per run with a "measure next" order.gstack-skill-billprints each skill's EVIDENCE line.## No-gstack armnow requiresmasking:,install:andaudit:lines, andprereg --checkrefuses a section without them. docs/TESTING_INTERNALS.md#no-gstack-arm says that any arm that read a gstack skill file is contaminated and gets zero credit.First inventory
Seeded from gstack-evals' fair-test table (GEVA-2, org-gstack-gbrain/research/gstack-part-evidence.json). No gstack part has a fair "helps" verdict yet. The ledger at this head reads 0 helps, 1 no effect (
/investigate), 17 untested with a plan, 48 untested with none and 5 safety. The/investigaterow is a CUT-CANDIDATE from HARN-1 (GST-814), and itsnotesays: "Only Opus 5.5 (claude-opus-5-5) was measured; GPT-6.1 Sol was not, so this verdict covers one of the two frontier models ETHOS.md 9 names." Every earlier result ran on the SDK stub, on a small model or with an unmasked arm, and these are recorded aspriorwith the reason. Nothing is cut in this PR. Cut candidates go to GStack Minder as R48 verdicts.Limits
No ceiling was raised. This PR doesn't touch lib/review-depth.ts, lib/review-risk.ts or .github/workflows, and it changes no eval case's or judge's model.
Review notes
Review ran at full depth (gate class, with money, security and data risk classes) in two passes, with Red Team, the targeted reader and the adversarial pass. The cross-family outside voice ran as a Capy GPT-6.1 Sol reader because Codex was unavailable.
Review-loop exception: accepted by GStack Minder (Captain, GSTA-36) 2026-11-02 via Captain ruling on #3837 relayed by the lane parent (rule 23): option A: one fix batch for review items 1-6, each with a test, plus one scoped fix-check; item 7 filed as GST-939; anything new at the floor ships as an open note (fix-check at a583778, 0 findings)
Fixed in this PR, each with a test that failed first:
grandfathered: trueand skip the plan rules. Now only the 71 seeded keys intest/fixtures/inventory-grandfathered.jsonmay be grandfathered (d0b3028).tasks.jsonor a prereg.md), so "proposed, not yet preregistered" no longer passes (6af40f4).gstack-inventorywith no--rootused to write into the installed gstack copy. It now defaults to the git repo you run it in, refuses a root with no ledger, refuses to write into a gstack install (resolved the way gstack resolves its install), and exits 2 on--rootfollowed by a flag or nothing (6af40f4).aliasesis validated,r48is checked on every status, andprior.whyrefuses prototype keys such astoString(6af40f4).gstack-skill-billprintsEVIDENCE: unreadable: …for a corrupt ledger instead of telling you to add a row (6af40f4).One finding is not fixed here. The PREREG fixtures promise a contamination audit that the model-panel and arm-benchmark runners don't implement yet, so it is filed as follow-up GST-939.
Below the blocking floor, listed and not fixed:
docs/inventory/evidence.jsonor the DISPATCHES source files totest/inventory.test.ts, so a broken ledger row first fails in CI free-tests.results --checkre-runsprereg --check, so a preregistration frozen before the masking rule now failsresults --checkunless edited after the run. Every committed prereg in the repo was updated, so no test fails./investigaterow says so in its note./eval-plan'sinstall:guidance names Claude paths only, not Codex.measure_bydates are never reported as overdue.keyOfhas no production caller./write-skilltrim dropped the words "resolves exact names only".evidence.json, skill-bill now says "unreadable: … ENOENT" where it used to say "missing: … add one".autoplan/sections/ceo-phase.md.tmpldropped "All 11 review sections: run fully, auto-decide each issue, log every decision." and the sentence explaining whatamend-inputexports (85e3eda). Both review passes covered that commit.docs/inventory.mdis a dated snapshot. After the base merge a few of its load numbers drifted (autoplan on-demand 18,701 to 18,756, captain 7,856 to 7,900, ubicloud 3,593 to 3,588) and a /ship source line moved fromdocs/review-policy.md:319to:339. Rerunningbin/gstack-inventoryrefreshes it; this head doesn't.Base merge
Main dd9834d was merged under GStack Minder's rule 19 grant (e56d331). The one conflict,
test/fixtures/skill-md-size-allowlist.json, took main's file and lowered the 12*/gstack-autoplanpins by 17 each to their measured sizes; no pin was raised. 38cf876 registerstest/fixtures/inventory-grandfathered.jsonwith its only reader,test/inventory.test.ts. One fix-check reviewer read both changes at this head and found nothing; the docs were re-audited after the merge.Train gate
ETHOS.md selects the paid office-hours and plan-ceo-review behavioral cases. None ran on this PR; they run in the train gate.
Documentation
Status: current — the branch's R56 docs read correctly next to main dd9834d at head 38cf876; no edits.
Scope: the committed diff from dd9834d (32 files, no staged, unstaged or untracked content), covering ETHOS principle 9, CONTRIBUTING step 6, docs/TESTING_INTERNALS.md no-gstack masking, docs/troubleshooting.md PREREG_NO_GSTACK_ARM, docs/skills.md Evidence, capy/skill/gstack/factory.md and the evidence ledger.
Coverage debt: bin/gstack-inventory and docs/inventory/evidence.json have reference and how-to coverage, but README.md, AGENTS.md and docs/PROJECT_STRUCTURE.md do not list them. Diagram drift: none.