---
description: "Comprehensive harness audit: boundary integrity, claim-vs-reality, governance, context poisoning, cross-platform compatibility. Triggers: 'audit harness', 'check boundaries', 'verify architecture', 'audit skills', 'cross-platform audit'."
agent: hf-l0-orchestrator
tools:
  - delegate-task
  - delegation-status
  - configure-primitive
  - hivemind-steer
subtask: true
---

<objective>
Orchestrate a multi-phase audit of the HiveMind V3 harness project. You are the coordinator — you MUST NOT perform audit work yourself. Dispatch parallel subagents for each audit phase, collect their reports, and synthesize findings.
</objective>

<injected_state>
!bash
echo "=== FILE TREE ==="
find src/ .opencode/ dist/ -type f 2>/dev/null | head -100
echo "=== GIT STATUS ==="
git status --short 2>/dev/null | head -30
echo "=== MODULE SIZES ==="
wc -l src/lib/*.ts src/plugin.ts src/index.ts 2>/dev/null
echo "=== SKILLS ==="
ls .opencode/skills/ 2>/dev/null
echo "=== COMMANDS ==="
ls .opencode/commands/ 2>/dev/null
echo "=== AGENTS ==="
ls .opencode/agents/ 2>/dev/null
</injected_state>

<delegation_plan>
You will dispatch 5 parallel subagents (Phases 1-4 are independent; Phase 5 is the synthesis gate).

PHASE 1 — Source & Distribution Inventory (parallel)
PHASE 2 — Claim vs Reality Validation (parallel)
PHASE 3 — Governance & Workflow Coherence (parallel)
PHASE 4 — Cross-Platform & Environmental Audit (parallel)
PHASE 5 — Synthesis & Mandate Gate (sequential, after 1-4 complete)

Dispatch Phases 1-4 in PARALLEL. Wait for all 4 reports. Then run Phase 5 to synthesize.
</delegation_plan>

<instructions>
1. Load the `hm-opencode-project-audit` skill — it contains the full dispatch envelopes for each phase.
2. Read the injected state above for current project snapshot.
3. Dispatch 4 parallel subagents using the Task tool. Each subagent gets:
   - A constructed prompt (full task text, NOT a file reference)
   - Clear scope boundaries
   - Output format requirements
4. Collect all 4 reports.
5. Run Phase 5 synthesis: merge findings, check mandate gates, produce final audit-report.md.
6. STOP on any critical finding. Report facts only. Do NOT fix anything.
</instructions>

<anti_patterns>
- DO NOT scan files yourself — delegate to subagents
- DO NOT read every file in the project — subagents handle discovery
- DO NOT produce the final report until all 4 phase reports are collected
- DO NOT fix issues found — this audit reports facts only
</anti_patterns>