agentleFS
junhjangjunhjang/claude-harness/.claude/skills/audit-harness-coverage/SKILL.md

audit-harness-coverage

Skill0 starsChanged 6 months agoMITView on GitHub

Audit harness coverage — cross-references SPECs declared under docs/harness/components/ against deployed primitives in .claude/skills/, .claude/agents/, .claude/hooks/, and scripts/cron/. Run when asked "is harness healthy", "is the harness fit for use", "audit harness coverage", "is the harness still doing what it claims", "does every SPEC have an impl", "does every deployment have a SPEC", or as step 3 of the harness-fitness-check composite. Deterministic ($0, no LLM calls), read-only against the harness, writes AUDIT-typed report to .claude/audits/.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 2 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Deterministic-only: no LLM calls. For semantic SPEC review, dispatch /review-doc <path-to-SPEC> separately.

    line 22
  2. Very specific:

    After the script returns, report counts to operator: <n> SPECs / <m> skills / <a> agents / <h> hooks scanned, <B> blockers, <W> warnings, …

    line 39
  3. Distinctive:

    Read-only against the harness: Never auto-fix a SPEC, skill, agent, or hook.

    line 54
  4. Distinctive:

    Deterministic-only: No LLM calls under any flag.

    line 55

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/audit-harness-coverage

Note: Ships with a foundation stub at scripts/audit_harness_coverage.py covering settings-wiring + skill/agent/hook/SPEC discoverability + referenced-but-missing reverse check. The full SPEC defines six checks (3.1 cron parity, 3.2 state stores, 3.3 skills, 3.4 agents, 3.5 hooks, 3.6 referenced-but-missing); the stub implements 3.3/3.4/3.5 + 3.5b settings-wiring + 3.6. Extend in your consuming repo. Contract: skill-audit-harness-coverage/SPEC.md.

Mechanizes the question "is the harness still doing what it claims" — previously answered by manual planning audits. This skill makes it a deterministic check.

When invoked

  1. Run the implementation:

    python3 scripts/audit_harness_coverage.py --source manual
    • --source manual makes the script exit 1 on any BLOCKER (CI-gate behavior). Cron mode (--source cron) always exits 0 (metric-only).
    • The script writes to .claude/audits/<YYYY-MM-DD>-harness-coverage.md (suffixing -N if a same-day audit already exists).
    • Deterministic-only: no LLM calls. For semantic SPEC review, dispatch /review-doc <path-to-SPEC> separately.
  2. The SPEC at docs/harness/components/skill-audit-harness-coverage/SPEC.md §3 declares six checks. The foundation stub implements 3.3 / 3.4 / 3.5 + 3.5b (settings-wiring) + 3.6 only:

    • 3.3 Skill discoverability — every .claude/skills/<name>/SKILL.md parses, has name+description, dir matches frontmatter name.
    • 3.4 Agent discoverability — every .claude/agents/<name>.md parses, has name+description+model.
    • 3.5 Hook discoverability — every .claude/hooks/<name>.sh has a shebang (BLOCKER if missing), is executable (BLOCKER if not chmod +x), and carries a >=40-char header comment (WARNING if shorter).
    • 3.5b Settings wiring — every hook command in settings.json resolves to a real file on disk.
    • 3.6 Referenced-but-missing — backtick-quoted references in docs to harness components that no longer exist on disk (catches deletion / rename drift).
    • 3.1 Cron implementation parity and 3.2 State-store presence are described in the SPEC but not implemented in the foundation stub. Consuming repos extend the stub as their cron components and state stores land.
  3. Severity categorization:

    • BLOCKERS — production-claimed SPECs without impl, malformed frontmatter, dir/name mismatches, non-executable hooks.
    • WARNINGS — state-store paths declared but absent (where SPEC isn't draft); short hook headers.
    • INFO — status: draft SPECs without impl; absent state stores when SPEC declares draft.
  4. After the script returns, report counts to operator: <n> SPECs / <m> skills / <a> agents / <h> hooks scanned, <B> blockers, <W> warnings, <I> infos.
  5. For each BLOCKER, propose a CHANGE-REQUEST. Operator confirms before opening any GitHub Issue — never auto-create tickets.

Composite usage

This skill is one step of a Harness fitness check composite. When the operator types "is the harness healthy" / "verify harness", the LLM should fire (in order):

  1. Hook smoke tests (bash .claude/hooks/test_*.sh)
  2. /audit-docs (deterministic D1–D11)
  3. /audit-harness-coverage (this skill — cross-cutting structural)
  4. Synthesize the three layers' findings.

Rules

  • Read-only against the harness. Never auto-fix a SPEC, skill, agent, or hook. Findings report; operator edits.
  • Deterministic-only. No LLM calls under any flag.
  • No semantic SPEC review — that is /review-doc <path>. This skill answers structural questions only ("is this SPEC's impl wired up?"), not "is this SPEC well-written."

Forbidden

  • Modifying any file under docs/harness/components/, .claude/skills/, .claude/agents/, .claude/hooks/, or scripts/cron/.
  • Auto-creating tickets without operator confirmation.
  • Treating state-store absence as a BLOCKER when the corresponding SPEC has status: draft or "planned" markers.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.