Skip to content

Wire the skill-integrity loop hooks into the develop/research protocol skills - #1064

Merged
aarontrowbridge merged 2 commits into
mainfrom
skill-integrity-hooks-1053
Sep 13, 2026
Merged

aarontrowbridge merged 2 commits into
mainfrom
skill-integrity-hooks-1053

Conversation

@aarontrowbridge

Copy link
Copy Markdown
Member

Closes #1053

What

Spec step 5 of the skill-integrity campaign (design: amicode/skills-integrity/design-20260913-recursive-skill-improvement.md): recursive skill improvement wired into the protocol cards as sentences placed where the phase text already lives — detection autonomous, application still PR-gated.

Per card

Card Hook Added lines
implement-issue front-line findings rule (skill-reality conflicts are findings, not workarounds) in the Invariants standing contract next to the tdd rules; optional skill_findings: field in the step-7 structured return +3 body (+2 yaml)
develop integrate boundary: skills lint (node packages/extension/scripts/lint-skills.mjs) when available on the branch (currently PR #1046) + API-surface diff-check (hits → findings in amicode/skills-integrity/); record boundary: the skill delta row +2
autodev one summary-level line each in the integrate (loop step 6) and record (loop step 7) phases, pointing at the develop walk's full text +6 (incl. F14 text)
research hypothesizer reads the findings ledger as a prior (step 2); analyzer gains the skill-friction check — mismatches → findings, repeated un-skilled idioms → proposal-type findings (step 6); record gains the skill delta row (step 7) +7
director-core the Record clause gains two sentences: the ledger update includes the campaign's skill delta; briefs may carry the front-line findings rule +3

Findings ledger cross-referenced as amicode/skills-integrity/ (personal vault, campaign-scoped) everywhere.

The one sanctioned extra (finding F14)

autodev/SKILL.md shipped the dangling internal marker "Switching modes mid-session: PENDING-D5." — replaced with the honest text: "Switching modes mid-session is not yet specified; the mode cards' posture-honesty sections govern until it is." (pre-authorized by the director).

Consequence: workflow_skills_public.test.ts pins the handoff section's pending status by that exact marker string, so its assertion follows the skill's new wording (asserts "not yet specified" while D5 is unlanded; same parameterized flip when slice 5 lands — never a deleted test). This test-only change is the one file beyond the five targets; without it the sanctioned replacement turns the suite red.

revision bumped 1 → 2 on the four revisioned cards (the D2 content-change convention; research carries no revision field — frontmatter untouched).

Verification

  • Full suite (pnpm --filter amicode test): 3333 passed. Failure set is byte-identical to the pristine origin/main baseline measured in this fresh worktree before any edit — 10 environmental tests / 6 files: the 5 named in the brief (amicode_service_runner ×1, chat_bridge ×2, skill_drift_lint ×2 — Strumento fixture availability) plus 5 worktree-shape failures (cli_gate ×1, main_source_resolver ×2, rebuild_coordinator ×2 — git/binary-path assumptions). No new failures; the intra-set count fluctuates 9–10 with a known flaky member.
  • Boot smoke: [smoke] PASS (after vendoring the pinned binary into the fresh worktree).
  • All guard surfaces over these cards green: workflow_skills_public (incl. the updated D5 pin), package_skills, packaging, skill_salvage_851, naming_records, mode_cards/mode_registry parity, drift-lint structural lane.
  • Content lens verified: no blocklisted proprietary strings, no internal path shapes, no markdown links added (structural lint unaffected).

Notes for review

  • The develop lint hook is phrased "when available on the branch" because lint-skills.mjs is still PR skills-integrity lint: retired namespaces, dead skill paths, frontmatter drift #1046 — it does not exist on main yet, per the issue's explicit instruction.
  • research/SKILL.md also carries a PENDING-D5 marker (its own handoff bullet) — left untouched; the F14 authorization covered autodev/SKILL.md only, and no test pins research's copy.
  • No behavioral change beyond the added contract lines; modes/ cards untouched (O4: post-ADR-0005, per the design note's sequencing).

Draft — not ready for review until the director checks placement voice-over-card.

…search protocol cards (#1053)

Surgical sentence-level additions, each landing where the phase text
already lives (design: skills-integrity/design-20260913):

- implement-issue: front-line findings rule in the implementer's
  standing contract (skill-reality conflicts are findings, not
  workarounds) + the optional skill_findings field in the step-7 return
- develop: integrate boundary gains the skills lint (when available on
  the branch, PR #1046) + API-surface diff-check; record boundary gains
  the skill delta row
- autodev: one summary-level line each in the integrate and record
  phases, pointing at the same hooks
- research: hypothesizer reads the findings ledger as a prior;
  analyzer gains the skill-friction check (mismatches -> findings,
  repeated un-skilled idioms -> proposal findings); record gains the
  skill delta row
- director-core: the Record clause gains the skill delta + the
  briefs-may-carry-findings-rule sentences

Also replaces autodev's dangling 'PENDING-D5' marker with the honest
'not yet specified' text (finding F14, pre-authorized) and updates the
D5-parameterized pin in workflow_skills_public.test.ts to follow the
skill's new wording (same flip semantics, never a deleted test).

Revision bumped to 2 on the four revisioned cards (content change).

Closes #1053
@coderabbitai

coderabbitai Bot commented Sep 13, 2026

Copy link
Copy Markdown

Important

  • 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 6c4ee9df-564f-44c3-8c07-c4fa07f7f42f


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

…honest text (F14 follow-up, #1053)

Same sanctioned F14 cleanup as autodev, now covering research's own
handoff copy: 'Switching modes mid-session: PENDING-D5.' becomes
'Switching modes mid-session is not yet specified; the mode cards'
posture-honesty sections govern until it is.' No test pins research's
copy (workflow_skills_public.test.ts pins only autodev's handoff
section), so no companion test change is needed.

Part of #1053 (PR #1064 follow-up commit)
@aarontrowbridge
aarontrowbridge marked this pull request as ready for review September 13, 2026 21:02
@aarontrowbridge
aarontrowbridge merged commit 6430292 into main Sep 13, 2026
10 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Wire the Skill Integrity Loop hooks into the develop/research protocol skills (recursive skill improvement)

1 participant