Skills Audit

Periodic review of the project-local skill set under .agents/skills/. The goal is to catch drift before the catalog becomes confusing — too many skills, overlapping triggers, descriptions that no longer match the body, references to skills that were renamed/deleted.

Recommended cadence: weekly, or after any week where >1 skill was added/renamed.

Procedure

1 — Inventory

Build a fresh census of all SKILL.md files. Do NOT trust any prior cached list.

find -L .agents/skills -name SKILL.md | wc -l                      # total count, including symlinked skills
find -L .agents/skills -name SKILL.md -exec wc -l {} \; | sort -rn # by body length, including symlinked skills

Group by domain in a mental table (DB / state / UI / agent / testing / workflow / docs / etc.). Note new arrivals since last audit (git log --since="1 week ago" -- .agents/skills/).

2 — Pull frontmatter for all skills

# Extract name + description for each SKILL.md
for f in .agents/skills/*/SKILL.md; do
  echo "=== $(basename $(dirname $f)) ==="
  awk '/^---$/{c++; next} c==1' "$f" | head -20
done

Read the description block of every skill. The body can stay unread unless step 4 flags it.

3 — Detect overlap / redundancy

For each pair within the same domain, ask:

• Same description? → likely duplicate (one is probably a stale rename leftover, or a global-vs-local collision).
• Trigger keywords substantially overlap? → either merge, OR tighten one description so the model can choose unambiguously.
• One skill's body says "see also: foo"? → confirm foo still exists, AND confirm the cross-reference is still meaningful (the referenced skill may have absorbed the referrer's concerns).
• Skill duplicates content from AGENTS.md? → fold into AGENTS.md or slim the skill to just the delta.

Common false positives (do NOT merge):

• db-migrations vs drizzle — distinct workflows (migration files vs schema authoring).
• agent-runtime-hooks vs agent-tracing vs agent-signal — different surfaces of the agent system.
• testing vs agent-testing — different test types.

4 — Description format consistency

Apply the standard template:

{Topic + key conventions or scope}. Use when {scenarios — verbs + nouns}. Triggers on {`code-symbols`, 'natural phrases', '中文'}.

Skills with disable-model-invocation: true (user-invoked only, slash commands) don't need Triggers on — they're never auto-routed.

Flag descriptions that:

• ❌ Have NO Use when clause (model can't decide when to load it).
• ❌ Have NO Triggers on clause (and aren't disable-model-invocation).
• ❌ Use weird formats (numbered lists (1)(2)(3), Triggers: colon instead of Triggers on, MUST use when ... as opening word).
• ❌ Are dramatically terse for a 200+ line body, or dramatically verbose for a 60-line body.
• ❌ Reference deleted/renamed skills.

5 — Stale-skill check

For narrow domain skills (e.g. response-compliance, one-off CLI workflows):

# Confirm the referenced code surface still exists
rg -l "response-compliance|openresponses" packages/ src/              # adjust per skill
git log --since="3 months ago" -- .agents/skills/ < skill > /SKILL.md # is it being maintained?

If the underlying surface is gone and the skill hasn't been edited in 3+ months → flag for archival.

6 — Cross-reference integrity

Any skill body mentioning another skill by name:

# Scan all skill bodies for skill-name references
rg -o '`[a-z][a-z0-9-]+`' .agents/skills/*/SKILL.md | grep -v ':\s*$' | sort -u

For each name extracted, confirm .agents/skills/<name>/SKILL.md exists. Broken references happen after renames — fix them in the same audit pass.

7 — Output report

Produce a markdown summary back to the user with the same structure as the original audit (this skill was created during one):

## 📊 Inventory

{count, domain breakdown}

## 🎯 Recommendations

### 🔴 High confidence

- {action} — {reason}

### 🟡 Medium confidence

- {action} — {reason needs verification}

### 🟢 Low confidence / no-op

- {item considered but skipping because ...}

## 📋 Suggested order

{table of actions with risk + LOC estimate}

End by asking the user which actions to apply — do NOT auto-apply unless the user passed --apply and even then confirm destructive deletes individually.

Output rules

• Be specific. "Skill X overlaps with Y" is useless without naming the overlapping triggers.
• Cite line numbers when flagging description / body issues.
• Don't recommend merges unless the call sites would actually load the merged skill in the same context.
• Don't recommend deletes for skills that haven't been touched recently — "unused" can mean "stable", not "dead".

What NOT to do

• ❌ Don't rename skill directories without checking for cross-references AND user memory entries that name the old slug.
• ❌ Don't normalize a description by removing trigger keywords just to fit the template — the keywords are the routing signal.
• ❌ Don't fold a heavy 200+ line skill into another just because they share a domain — large skills get loaded selectively and merging makes everything load.
• ❌ Don't propose .agents/skills/INDEX.md or <domain>-<skill> prefix renames unless the user explicitly asks — costs > benefits for cosmetic reorgs.

Related history

• First audit: chore/skills-audit branch (2026-05-25) — deleted source-command-dedupe, renamed data-fetching → data-fetching-architecture, normalized 9 descriptions, created this skill.