Audit skills and memory files for docs-management delegation compliance.
Audits skills and memory files for docs-management delegation compliance.
/plugin marketplace add melodic-software/claude-code-plugins/plugin install claude-code-observability@melodic-softwareAudit skills and memory files for docs-management delegation compliance.
Before auditing, initialize the environment:
.claude/temp/) exists.The ecosystem-health skill provides authoritative validation guidance (auto-loaded when this command runs).
| Argument | Description |
|---|---|
| (none) | Smart mode: audit modified, never-audited, or stale (>90 days) files |
--force | Audit ALL files regardless of status |
--plugin-only | Only audit local plugin skills |
--project-only | Only audit project files (.claude/skills/, .claude/memory/) |
skill-name | Audit specific skill(s) by name |
Search for files that may contain Claude Code references:
Skills:
plugins/*/skills/*/SKILL.md (local plugins).claude/skills/*/SKILL.md (project skills)Memory Files:
.claude/memory/*.md (project memory)plugins/*/skills/*/references/*.md (skill references)Exclusions (auto-exempt):
hooks.json files (configuration specification)hooks/*.py, hooks/*.sh (hook implementations)canonical/ directories (scraped documentation cache)plugins/tac/lessons/ (educational content)Build a list of discovered files with their scope (plugin, project) and full path.
Check for --force, --plugin-only, --project-only flags or specific skill names.
Build the audit queue based on discovered files and arguments:
Display audit mode (SMART or FORCE), targets discovered, and list each with scope and last modified date.
## Audit Plan
**Mode**: SMART
**Files discovered**: 15
1. [plugin:claude-ecosystem] skills/hook-management/SKILL.md
2. [plugin:claude-ecosystem] skills/skill-development/SKILL.md
3. [project] .claude/memory/workflows.md
...
**Will audit**: 8 files (7 skipped - recently audited)
For each file, spawn the docs-delegation-auditor subagent with the following context:
Run subagents in parallel batches of 3-5.
Subagents write findings to .claude/temp/ as both JSON (for recovery/aggregation) and markdown (for human review). The main conversation thread collects results and updates audit logs using its Write/Edit tools.
CRITICAL: After ALL audits complete, sync results to the audit log at .claude/audit/docs-delegation.md.
Report total files audited, results by scope, and details table. List any compliance issues with remediation steps.
## Audit Complete
### Summary
- **Total audited**: 8 files
- **COMPLIANT**: 5 (62%)
- **PARTIALLY COMPLIANT**: 2 (25%)
- **NON-COMPLIANT**: 1 (13%)
### Results
| Scope | File | Classification | Score |
| --- | --- | --- | --- |
| plugin | hook-management/SKILL.md | ✅ COMPLIANT | 95/100 |
| plugin | skill-development/SKILL.md | ✅ COMPLIANT | 100/100 |
| project | workflows.md | ⚠️ PARTIAL | 78/100 |
| plugin | lesson-014-analysis.md | ❌ NON-COMPLIANT | 65/100 |
### Issues Found
1. **lesson-014-analysis.md** (Score: 65)
- HIGH: Hardcoded `PreToolUse`, `PostToolUse` without delegation note
- Missing MANDATORY section
- **Action**: Add verification note pointing to docs-management
Files are scored out of 100 points with deductions for hardcoded patterns:
| Deduction | Amount | Condition |
|---|---|---|
| HIGH risk pattern | -15 each | Hardcoded without delegation note |
| MEDIUM risk pattern | -8 each | Hardcoded without delegation note |
| Missing MANDATORY section | -20 | File has hardcoded data but no delegation section |
| Missing verification notes | -15 | Hardcoded data without verification blockquote |
Thresholds:
These indicate hardcoded Claude Code information that should delegate to docs-management:
| Pattern | Example |
|---|---|
| Hook event types | PreToolUse, PostToolUse, SessionStart |
| YAML frontmatter fields (in prose) | allowed-tools, permissionMode, color |
| Settings keys | CLAUDE_HOOK_*, apiChoice, sandboxMode |
| Pattern | Example |
|---|---|
| GitHub issue references | #10437, #15326 |
| Model names with versions | Opus 4.5, claude-opus-4-5 |
| Token limits | 25K tokens, context window |
| Pattern | Description |
|---|---|
| Verification note | > **Documentation Verification:** blockquote |
| MANDATORY section | ## MANDATORY: or ## 🚨.*MANDATORY |
| Query pattern table | Table with docs-management query patterns |
| Skill reference | invoke the X-management skill |
| Pattern | Reason |
|---|---|
hooks.json event names | Configuration specification |
allowed-tools in YAML frontmatter | Self-specification |
tools in agent YAML frontmatter | Agent definition |
| Educational content | Instructional, not authoritative |
canonical/ directories | Scraped docs cache |
All audit results are written to .claude/audit/docs-delegation.md.
Use /audit-log docs-delegation to view current audit status.
User: /audit-docs-delegation
Claude: Discovering audit targets...
## Audit Plan
**Mode**: SMART
**Files discovered**: 25
1. [plugin:claude-ecosystem] skills/hook-management/SKILL.md
2. [plugin:claude-ecosystem] skills/skill-development/SKILL.md
3. [plugin:tac] analysis/lesson-014-analysis.md
...
**Will audit**: 12 files (13 skipped - recently audited)
[Spawns docs-delegation-auditor subagents]
## Audit Complete
| Scope | File | Classification | Score |
| --- | --- | --- | --- |
| plugin | hook-management/SKILL.md | ✅ COMPLIANT | 95/100 |
User: /audit-docs-delegation --force
Claude: Auditing all files (force mode)...
User: /audit-docs-delegation hook-management
Claude: Auditing hook-management skill...
User: /audit-docs-delegation --plugin-only
Claude: Auditing plugin skills only...