'The single code-quality lane — a 5-lens Opus panel (bug hunter, adversary, failure engineer, promise keeper, test warden), minimum 3 rounds for cross-round consensus, scoped to code a user reaches in production. Audits the TEST SUITE as hard as the code: test-quality checks; mutation probes convict hollow tests and trimming counts as much as fixing. Fixes commit directly to `main` (trunk-direct); deferred findings become an epic of beads. Triggers: ''hygiene'', ''clean up the codebase'', ''iterative review'', ''tidy the code'', ''weekly hygiene run''. NOT for a single-module or single-domain deep dive (the audit checklists now live in `ac-review/references/`), the skill/agent registry (skill-builder's registry-audit workflow), or pipeline/board housekeeping (use ac-tidy).'
Resources
2Install
npx skillscat add craigvandotcom/agent-compounds/ac-hygiene Install via the SkillsCat registry.
You are the conductor. A panel of reviewers hunts independently, each through a different
lens. You synthesize, fix, and iterate. Codebase-wide — not tied to any feature branch or diff.
Scaling tiers per ac-pipeline/references/risk-classification.md.
Capability-starved runs: ac-pipeline/references/degraded-mode.md.
The weekly quality pass for a repo (PANEL=full, 5 lenses), or a quick between-session sweep
(PANEL=light, 3 lenses). Scoped to the product surface — references/reviewers.md § THE BAR.
I/O Contract
| Input | Full codebase, recent commits, or specific directory (user-selected scope) |
| Output | Fixed issues committed, health assessment report |
| Artifacts | Round findings in $ARTIFACTS_DIR/round-{N}-{role}.md, consensus registry |
| Verification | Quality gate (test, lint, type-check, build) all passing |
The run
- Initialize — scope, trunk-direct confirm, consensus registry, baseline gate, the
conductor-run lenses (coverage audit, knip, stamp-audit, friction cluster-walk), run ledger.
Full procedure:references/run-init.md. - Review loop (phases 1-4) — spawn the panel, synthesize, auto-apply, converge. Floor
MIN_ROUNDS=3, ceilingMAX_ROUNDS=5. Full procedure:references/run-loop.md. - Finalize (phase 5) — conductor triage, decisions, exhaustive quality gate, close
ceremony, in-session bead refine, report, cleanup. Full procedure:references/run-finalize.md.
Contracts this spine carries
The 5-lens panel (PANEL=full): bug hunter, adversary, failure engineer, promise keeper,
test warden — all Opus, spawned in one message, each writing$ARTIFACTS_DIR/round-{N}-{role}.md. PANEL=light = bug hunter + adversary + test warden.
Prompts: references/reviewers.md; § THE BAR scopes every lens to the product surface.
Auto-apply a fix on any of: Critical/High severity · same-round consensus (2+ agents) ·
cross-round consensus (recurs from a prior round's registry). Everything else defers to the
registry, except DESIGN_DECISION items (UX-affecting or approach-transforming), which defer
to the human regardless of severity. Canon: ac-pipeline/references/review-consensus.md.
Trunk-direct. There is no hygiene branch and no worktree. Fixes commit straight to main
as pathspec commits under full H7 discipline (ac-implement Phase 0) — never git add -A,
never git add ., never git commit -a, never git stash. Commit = push. The panel IS this
run's pre-push review, so there is no separate review step. Canon:ac-pipeline/references/commit-discipline.md.
Deferred-findings routing (Exhaust Rule). Nothing actionable leaves as prose. Out-of-scope
confirmed issues → a bead; worth-chasing uncertainties → -t investigation; genuine
taste/product forks in an autonomous run → -t decision --labels human-gate with a pre-staged
memo. A bead is something you'd schedule; nits stay in the report. Canon:beads-standards/reference/bead-conventions.md. When deferred findings batch into a
per-run epic, every child gets the parent-child (containment) edge, read back
(br show on both ends).
Deletion mandate. A stale/superseded/duplicate finding is REMOVAL-as-disposition, equal to
fixing: a verbatim duplicate deletes outright, a still-needed extract moves to references/,
unique content with no surviving twin routes through MAINTENANCE.md before git-delete. Canon:skill-builder/references/promotion-ladder.md.
Round floor. MIN_ROUNDS=3 is ABSOLUTE — cross-round consensus needs two later rounds to
recur. Never finalize before round 3; ceiling MAX_ROUNDS=5. Details: references/run-loop.md.
Run ledger. One task per major section; add a "Round N" task as each round begins.
Sub-skills invoked by the Ship and Refine tasks (ac-publish, ac-polish) run their OWN
ledgers — this ledger tracks top-level sections only. Contract:ac-pipeline/references/run-ledger.md.
Stamp-audit + coverage lenses are weekly-only conductor passes, not panel lenses:references/run-init.md.
When to use
Weekly quality pass per repo (PANEL=full) or a quick between-session sweep (PANEL=light).
For feature-specific review before merge, use /ac-review. Overrides ("light", "headless",
scope/round overrides), troubleshooting, and the standing-review-of-main duty:references/run-overrides.md.
Hygiene: the recurring codebase quality pass. For session closure: /ac-land.