agent-audit

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

1. SOUL.md structure — Must have ## Identity, ## Purpose, ## Intents headers. Missing headers = role clarity gap.
2. Skills directory — Must exist and be populated. Each subdirectory must contain a SKILL.md.
3. TOOLS.md — Must list real skills that match the skills directory. Phantom skills (listed but no SKILL.md) are a flag.
4. Workspace freshness — Call workspace_freshness MCP tool. Flag files older than 30 days.
5. Satisfaction score — Call satisfaction_scores MCP tool. Flag agents below fleet average.
6. IDENTITY.md — Must exist and be non-empty.
7. BOOTSTRAP.md — Should NOT exist. If present, onboarding is incomplete.

Process

1. If all, get agent list from agents_list MCP tool. Otherwise audit the named agent.
2. For each agent, run checks 1-7 above.
3. Produce a per-agent report with findings and recommended actions.
4. For each real problem found, file an issue via issue_log MCP tool.
5. 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 data
• workspace_freshness — File ages per agent
• issue_log — File issues for problems found
• chart_search — Check for existing known issues
• agents_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]