audit-plugin

Audit — Plugin Self-Audit

Audit the hero-skills plugin for quality, consistency, and maintainability. This skill is specific to the hero-skills repo itself — it reviews the skills that make up the plugin.

Arguments

• $ARGUMENTS:
  • (none) — Audit and report findings
  • --fix — Audit and auto-fix what can be fixed (formatting, ordering, etc.)

Instructions

Step 1: Inventory All Skills

ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
ls -1 "$ROOT/skills/"

Read every skills/*/SKILL.md file. For each, extract:

• Name (from frontmatter)
• Description (from frontmatter)
• Line count, word count
• Heading structure (h1, h2, h3 hierarchy)
• Whether it references HERO.md fields
• Arguments it supports

Step 2: Run Audit Checks

2a: Structure Consistency

All skills should follow a consistent structure. Check for:

• Frontmatter fields: Every skill must have name and description. Flag missing or inconsistent fields.
• Heading hierarchy: Skills should follow a predictable pattern. Flag skills with wildly different structures.
• Argument documentation: If a skill has argument-hint, it should document arguments in the body.
• Step numbering: Steps should be sequential, no gaps or duplicates.
• Sub-step numbering within a step: If a step has sub-steps like 2a, 2b, 2c, verify they're sequential with no gaps or duplicates.

Report template:

STRUCTURE CONSISTENCY
─────────────────────
[OK] commit: standard structure (frontmatter, args, instructions, principles)
[!!] plan: missing argument documentation for --dry-run
[!!] init: step sub-numbering gap (2a, 2b, 2d — missing 2c)

2b: Size & Complexity

Flag skills that are too large or too small:

• Over 500 lines: Should split into supplementary files
• Over 5000 words: Consuming too much context window
• Under 20 lines (body only): Probably too thin to be useful
• Deep nesting (h4+ headings beyond investigation sub-steps): May need restructuring

Report template:

SIZE & COMPLEXITY
─────────────────
[OK] commit: 180 lines, 1200 words
[!!] init: 830 lines, 6200 words — consider splitting investigation steps into references/
[OK] respond: 200 lines, 1500 words

2c: DRY Violations

Look for instructions that are repeated across multiple skills. Common patterns:

• "Read HERO.md" boilerplate — should each skill repeat how to read it, or should there be a shared pattern?
• "Check for git repo" — appears in many skills
• Similar investigation bash blocks
• Repeated formatting patterns for output (the [OK]/[??]/[--] format)

Flag when the same substantive instruction (not just similar phrasing) appears in 3+ skills.

Report template:

DRY VIOLATIONS
──────────────
[!!] "Read HERO.md and parse sections" — repeated in 8 skills
     Suggestion: This is expected — each skill needs to independently read HERO.md.
     No action needed (skills run independently, not as a pipeline).

[!!] "Present findings in [OK]/[??]/[--] format" — in init, setup
     Suggestion: Consistent by design. No action needed.

[??] Investigation bash blocks in init are 40+ lines each
     Suggestion: Consider moving to references/ if init exceeds 500 lines

Important: Not all repetition is bad. Skills run independently — they can't share runtime state. Only flag repetition that could be eliminated via supplementary files or shared references.

2d: HERO.md Field Coverage

Cross-reference every field in the HERO.md template (as defined in init) against which skills consume it:

HERO.MD FIELD COVERAGE
──────────────────────
Field                          | Produced by  | Consumed by
───────────────────────────────|──────────────|────────────────────
Coding Agent → primary         | init    | setup
Repository → hosting           | init    | push
Repository → branch-template   | init    | plan
Projects → lint-command         | init    | test
Projects → dependency-file     | init    | scan, test
...

[!!] Coding Agent → self-review: produced by init, consumed by NO skill
     → Is this field actually used? Remove or wire up.

[OK] All fields consumed by at least one skill

2e: Description Quality

Check every skill's frontmatter description for:

• Trigger clarity: Does it say when to use the skill? ("Use when...", "Use for...", "Use before...")
• Length: Should be 50-200 chars. Too short = unclear triggers. Too long = wastes context.
• Specificity: Vague descriptions like "helps with code" are useless for Claude's skill matching.

Report template:

DESCRIPTION QUALITY
───────────────────
[OK] commit: "Create a smart git commit..." (85 chars, clear trigger)
[!!] audit: description is 190 chars — consider trimming
[??] create-skill: no trigger phrase — add "Use when..." or "Use for..."

2f: Alphabetical & Organizational Checks

• Are skills listed alphabetically when referenced in tables (e.g., init-hero's "What each skill needs" table)?
• Are HERO.md sections in a logical order?
• Are frontmatter fields in a consistent order across skills?

Step 3: Report Summary

Plugin Audit
═════════════════════════
Skills audited: 17
Total lines: 4,200 | Total words: 28,000

MUST FIX
────────
critical issues: broken references, missing fields, structural errors

SHOULD FIX
──────────
quality issues: size, DRY violations, descriptions

INFO
────
OBSERVATIONS_WITH_NO_ACTION_NEEDED

Overall health: Good / Needs Attention / Critical

Step 4: Auto-Fix (if --fix)

If --fix is passed, automatically fix:

• Alphabetical ordering in tables
• Frontmatter field ordering (name, description, argument-hint, disable-model-invocation)
• Step renumbering gaps
• Trailing whitespace, inconsistent newlines

Never auto-fix: Content changes, description rewrites, structural reorganization — these need human review.

Key Principles

• This skill is for the hero-skills repo only. It audits the plugin, not user projects.
• DRY is not always better. Skills run independently — some repetition is by design.
• Field coverage matters. Every HERO.md field should be produced by init-hero and consumed by at least one skill.
• Size awareness. Skills consume context window. Large skills slow down every invocation.
• Be specific. File, line, what's wrong, how to fix.