system-audit

Comprehensive infrastructure audit. Run monthly or after major system changes.

Checks

Check 1: Agent Health

• Read every file in .claude/agents/*.md
• Verify each has valid frontmatter (---)
• Verify no TBD/TODO markers
• Check that referenced tools exist
• Check agent-memory directories exist for agents that need them

Check 2: Command Health

• Read every file in .claude/commands/*.md
• Verify each has valid frontmatter
• Check that allowed-tools are reasonable
• Verify no broken cross-references to other commands

Check 3: Hook Health

• Verify .claude/settings.json is valid JSON
• Check every hook script referenced in settings.json exists
• Verify all hook scripts are executable (chmod +x)
• Run a dry test: each hook should exit 0 with empty input

Check 4: Memory Tier Health

• memory.md: Is it under 100 lines? Is "Now" current?
• knowledge-base.md: Is it under 200 lines? Do all entries have [Source:]?
• knowledge-nominations.md: Are there stale nominations (>30 days)?
• agent-memory/: Do directories match existing agents?
• Daily Notes: Are recent notes present?

Check 5: Log Health

• audit-trail.md: Is it under 5000 lines?
• incident-log.md: Are there unresolved CRITICAL/HIGH events?
• failure-log.md: Are there recurring patterns?
• verdicts.jsonl: What's the block rate? Any task-type clustering?

Check 6: Permission & Config Coherence

• .claude/settings.json hooks match actual hook files
• No orphaned hook scripts (exist but not referenced)
• No missing hook scripts (referenced but don't exist)

Check 7: Cross-File Coherence

• CLAUDE.md references match actual file locations
• Command-index.md matches actual commands
• No circular dependencies between commands

Check 8: Backup & Storage

• .claude/backups/: Is auto-pruning working? (no dirs >7 days old)
• Large files in project root that shouldn't be there?

Check 9: Via Negativa Sweep

• Are there files/agents/commands that are never used?
• Could any hook be removed without loss?
• Is there duplicated logic between agents?
• Propose removals — simpler is better

Grading

Grade	Criteria
A	All checks pass, no issues
B	Minor issues only (cosmetic, non-blocking)
C	Some issues need attention (missing provenance, stale nominations)
D	Structural issues (broken hooks, invalid JSON, missing agents)
F	Critical failures (security issues, data loss risk)

Output

Write results to daily note under:

## System Audit — MMDDYY

**Grade:** [A-F]
**Checks:** [passed]/9
**Issues:** [bullets with severity]
**Actions:** [corrective tasks, if any]

Create corrective tasks on Task Board for any D or F grade issues.