tailrocks/tailrocks-code-quality-skills

Applies the instruction policy when work touches AGENTS.md, client entries, instruction rules, or rule placement. Adds one rule only when the active task already authorizes that exact change. Audit and repair belong to the manual-only sibling skills.

Audits all repository instruction files read-only for placement, duplicate content, deletion evidence, topology, load cost, and enforceable-rule routing. Use this skill only when the user explicitly requests it. Never edits or repairs.

Applies one explicitly approved instruction-topology repair from a current audit: creates or repairs client entries, relocates or deletes approved rules, and verifies exact parity. Use this skill only when the user explicitly requests it. Adds no new policy.

Establishes or tightens one explicitly approved shrink-only code-health ratchet for architecture, lint, dependency, flake, defect, documentation, or verification debt. Use this skill only when the user explicitly requests it. Audit and measurement-only reporting belong to tailrocks-code-health-audit.

Audits one code-health debt class read-only: inventories gates and exceptions, measures the baseline, evaluates shrink-only enforcement and verification placement, and emits fixed-ID evidence. Use this skill only when the user explicitly requests it. Never installs or edits.

Audits one repository read-only through bounded lanes and returns one evidence-ranked report with verified defects, direction options, and rejected candidates. Use this skill only when the user explicitly requests it. Never plans, implements, reconciles, or changes source.

Executes one approved standalone plans/ file in an isolated disposable worktree, re-runs its criteria, independently reviews the diff, and returns one verdict. Use this skill only when the user explicitly requests it. Never merges. Pushes only with separate explicit authorization.

Converts one selected verified finding or described change into one standalone executor-ready plan under plans/ and its index row. Use this skill only when the user explicitly requests it. Never implements, seeds roadmap work, commits, or pushes.

Reconciles the standalone plans/ backlog against current repository truth, optionally re-verifying every row, and updates only plans/README.md. Use this skill only when the user explicitly requests it. Never edits plans, source, or roadmap items.

Audits one repository read-only for security defects with bounded threat analysis, secret-safe evidence, and adversarial verification. Use this skill only when the user explicitly requests it. Never fixes, exploits, publishes secrets, or changes source.

Applies one explicitly approved root-cause correction while preserving every unapproved behavior and proving instance and defect-class prevention. Use this skill only when the user explicitly requests it. Requires a current diagnosis contract. Never diagnoses, chooses, or approves the design.

Diagnoses one proven defect, reported friction, or failed guarantee read-only, derives the bounded causal class, and returns practical correction options. Use this skill only when the user explicitly requests it. Never edits, contains harm, or approves its own design.

Applies one explicitly approved set of measured code removals within a bound diff while preserving every observable behavior. Use this skill only when the user explicitly requests it. Requires a proven oracle. Never discovers removals or changes behavior.

Audits one pull request, branch, or diff read-only for measured code removals whose observable behavior can be preserved. Use this skill only when the user explicitly requests it. Returns findings only. Never edits, tests in, or applies them.