audit

Validate the internal consistency of the Olympus plugin: agent permissions, cross-references, artifact contracts, gate thresholds, and clarity rules. Hephaestus and Athena operate as teammates for cross-phase context sharing.

<Execution_Policy>

• This skill uses FULL TEAMMATE mode. Hephaestus and Athena are teammates.
• Each Step MUST call the specified MCP tool. Do NOT skip MCP calls.
• Do NOT perform validation directly. Hephaestus handles mechanical, Athena handles semantic.
• Leader handles ONLY: team management, report synthesis.
• IMPORTANT: Do NOT skip ToolSearch at Step 0.
• PROACTIVE SPAWN RULE (§6.3): Every Agent() call MUST include IMMEDIATE TASK in prompt. NEVER use "Wait for messages — do not act until prompted."
• MANDATORY CONSULTATION (§7): Athena MUST read hephaestus results before reporting. Athena's report without reference to audit-mechanical.json is incomplete.
• RESPONSE RULE: If a teammate does not report within reasonable time:
  1. SendMessage(to: "{agent}", "Report your findings now. Include consultation results. Keep under 5000 chars.")
  2. Retry up to 3 times.
  3. NEVER do the agent's work directly — this violates §0.
• RESULT CAPTURE RULE: Read-only agents deliver results via SendMessage(to: "team-lead"). Orchestrator writes artifacts from these results. Write-capable agents write files directly.
• SEQUENTIAL SPAWN: hephaestus first → athena after hephaestus completes. Wait for prerequisite agent results before spawning dependent agents. </Execution_Policy>

<Team_Structure> team_name: "audit-${CLAUDE_SESSION_ID}"

Teammates:

Agent	Role	Comm Targets
hephaestus	Mechanical validation (YAML, files, structure)	leader
athena	Semantic validation (permissions, contracts, gates)	hephaestus (evidence), leader
</Team_Structure>

Step 0: Load MCP Tools (REQUIRED FIRST)

Call ToolSearch("+olympus pipeline") to load MCP tools.

---

Step 1: Initialize

1. TeamCreate(team_name: "audit-${CLAUDE_SESSION_ID}")
2. olympus_start_pipeline(skill: "audit", pipeline_id: ...)
3. Create artifact directory: .olympus/audit-{YYYYMMDD}-{short-uuid}/

---

Step 2: Hephaestus Mechanical Validation

heph_result = Agent(name: "hephaestus", team_name: ${TEAM},
      subagent_type: "olympus:hephaestus",
      prompt: "You are Hephaestus in team ${TEAM}, mechanical validator. Artifact directory: ${ARTIFACT_DIR}/
        LEADER_NAME: team-lead
        IMMEDIATE TASK: Validate Olympus plugin structural integrity:
        1-1. YAML Frontmatter: validate agents/*.md against agent-schema.json
        1-2. File existence: verify cross-references between agents and skills
        1-3. Shared doc references: verify docs/shared/ references exist
        1-4. artifact-contracts.json: verify writer/reader agents exist
        1-5. Hook scripts: verify hooks.json scripts exist, are executable, pass bash -n
        Write results to ${ARTIFACT_DIR}/audit-mechanical.json directly (you have Write access).
        When done: SendMessage(to: 'team-lead', summary: 'hephaestus 검증 완료', '{PASS/FAIL summary}')")
olympus_register_agent_spawn(pipeline_id, "hephaestus")

→ hephaestus writes audit-mechanical.json directly; leader reads after SendMessage notification
olympus_record_execution(pipeline_id, "audit", "hephaestus", ...)

Decision:
  All PASS → proceed to Step 3
  Any FAIL → note failures, proceed to Step 3 (audit reports findings, does not block)
  ENV_UNAVAILABLE → proceed to Step 3 with caveat (audit-mechanical.json notes environment unavailable)

---

Step 3: Athena Semantic Validation

athena_result = Agent(name: "athena", team_name: ${TEAM},
      subagent_type: "olympus:athena",
      prompt: "You are Athena in team ${TEAM}, semantic validator. Artifact directory: ${ARTIFACT_DIR}/
        LEADER_NAME: team-lead
        DO NOT write files — you are read-only.
        IMMEDIATE TASK: Perform semantic validation of Olympus plugin.
        Step 1 — Read ${ARTIFACT_DIR}/audit-mechanical.json (hephaestus results).
        Step 2 — Read agents/*.md, skills/*.md, docs/shared/*.
        Step 3 — Validate:
        2-1. Permission-Role Consistency: disallowedTools vs prompt content
        2-2. Artifact Contract Completeness: skill files vs contracts
        2-3. Gate Consistency: gate-thresholds.json vs SKILL.md values
        2-4. Clarity Scan: banned phrases from docs/shared/clarity-enforcement.md
        2-5. Delegation Pattern: Write/Edit disabled agents send results via SendMessage(to: "team-lead")
        2-6. Pipeline State Schema: state structures match pipeline-states.json
        2-7. LEADER_NAME Consistency: every agent spawn prompt in SKILL.md files contains
             "LEADER_NAME: team-lead" (literal). Missing LEADER_NAME is a configuration violation.
        2-8. Proactive Spawn Compliance: every SKILL.md that spawns agents contains
             PROACTIVE SPAWN RULE and IMMEDIATE TASK in the Execution_Policy. Missing either is a violation.
        2-9. Spawn Registration Coverage: every olympus-typed spawn (subagent_type matching "olympus:")
             in SKILL.md has a following olympus_register_agent_spawn call. Missing registration
             is a violation. Exception: general-purpose agents (no olympus subagent_type) are exempt.
        2-10. Harness Pattern Coverage: every SKILL.md that spawns agents contains
              SEQUENTIAL SPAWN, RESULT CAPTURE RULE, MANDATORY CONSULTATION, and RESPONSE RULE.
              Missing any of these four patterns is a violation.
        MANDATORY CONSULTATION: Before outputting final results, you MUST:
          SendMessage(to: 'hephaestus') with at least one cross-check question.
          Wait for response. Include consultation exchange in your output.
        When done: SendMessage(to: 'team-lead', summary: 'athena 검증 완료', '{audit-semantic.json content}')")
olympus_register_agent_spawn(pipeline_id, "athena")

→ Write audit-semantic.json from athena SendMessage
olympus_record_execution(pipeline_id, "audit", "athena", ...)
olympus_log_collaboration(pipeline_id, "athena", "hephaestus", "의미 검증: athena↔hephaestus 크로스검증")

DEADLOCK FALLBACK: If athena does not send a cross-check question to hephaestus within 3 minutes:
  → SendMessage(to: "athena", "Cross-check consultation timeout. Finalize audit-semantic.json using audit-mechanical.json findings only. Note 'hephaestus consultation pending'.")
  → Leader flags incomplete cross-verification in audit-report.md.

---

Step 4: Audit Report

Leader synthesizes both results:

# Olympus Audit Report

## Summary
- Mechanical: {PASS/FAIL}
- Semantic: {CLEAN/WARNING/VIOLATION}
- **Overall: {CLEAN/WARNING/VIOLATION}**

## Violations (immediate fix required)
| # | Category | Target | Issue | Location |

## Warnings (manual review recommended)
| # | Category | Target | Issue | Location |

## Coverage
- Agents: {n}/{total}, Skills: {n}/{total}
- Schemas validated: agent-schema.json, pipeline-states.json, gate-thresholds.json

Save to ${ARTIFACT_DIR}/audit-report.md

Verdict: CLEAN / WARNING / VIOLATION

---

Step 5: Teardown

Shutdown all teammates → TeamDelete

<Tool_Usage> MCP Tools:

• olympus_start_pipeline: Step 1 (MUST)
• olympus_register_agent_spawn: after each spawn (MUST)
• olympus_record_execution: after each agent (SHOULD)
• olympus_log_collaboration: Step 3 athena↔hephaestus consultation (SHOULD)

Team Tools:

• TeamCreate: Step 1
• Agent (name + team_name): spawn hephaestus, athena
• SendMessage: sequential (athena reads hephaestus results)
• TeamDelete: Step 5 </Tool_Usage>

<Artifact_Contracts>

File	Step	Writer	Readers
audit-mechanical.json	2	hephaestus (direct)	athena
audit-semantic.json	3	Leader (from athena)	Leader
audit-report.md	4	Leader	User
</Artifact_Contracts>