| name | agent-audit |
| description | Audit agent health — SOUL.md structure, skills, workspace freshness, satisfaction scores. Reports findings and files issues. |
| version | 1.0.0 |
| author | captain |
| tags | ["fleet","audit","health","satisfaction"] |
| intent | Observable [I13] |
agent-audit
Audit one or all agents for health and completeness.
Trigger
/agent-audit [agent-id] or /agent-audit all
What to check per agent
- SOUL.md structure — Must have
## Identity, ## Purpose, ## Intents headers. Missing headers = role clarity gap.
- Skills directory — Must exist and be populated. Each subdirectory must contain a
SKILL.md.
- TOOLS.md — Must list real skills that match the skills directory. Phantom skills (listed but no SKILL.md) are a flag.
- Workspace freshness — Call
workspace_freshness MCP tool. Flag files older than 30 days.
- Satisfaction score — Call
satisfaction_scores MCP tool. Flag agents below fleet average.
- IDENTITY.md — Must exist and be non-empty.
- BOOTSTRAP.md — Should NOT exist. If present, onboarding is incomplete.
Process
- If
all, get agent list from agents_list MCP tool. Otherwise audit the named agent.
- For each agent, run checks 1-7 above.
- Produce a per-agent report with findings and recommended actions.
- For each real problem found, file an issue via
issue_log MCP tool.
- Search Chartroom via
chart_search for any existing known issues about the agent.
Authority
Report + file issues only. Do NOT modify agent files directly. Delegate fixes to the appropriate specialist or escalate to Reactor.
MCP Tools Used
satisfaction_scores — Fleet satisfaction dataworkspace_freshness — File ages per agentissue_log — File issues for problems foundchart_search — Check for existing known issuesagents_list — Get full agent list (for all mode)
Output Format
Per agent:
### [agent-id] — Score: XX/100
- SOUL.md: OK / Missing [headers]
- Skills: N skills, M phantom
- Freshness: OK / [N files stale]
- Satisfaction: XX/100 (fleet avg: YY)
- Issues filed: [list]
- Recommended: [actions]
