astro44/bookend-reconcile

bookend-reconcile - Agent Report Reconciliation

Merges structured outputs from bookend agents into a single assessment. Revises the initial heuristic scores with evidence-backed data and produces the final DecompositionPlan.

Input Schema

{
  "reports_dir": "/path/to/reports/bookend",
  "original_readiness": {},
  "original_complexity": {},
  "bookends_dir": "/path/to/go-autonom8"
}

Instructions

1. Collect Available Reports

Read all JSON files from reports_dir:

• cartography.json
• dependency-graph.json
• contracts.json
• quality-audit.json
• risk-assessment.json

Not all will exist — only agents that were spawned produce reports.

2. Revise Readiness

Starting from original_readiness.score:

• If cartography created artifacts: add their weight to score
• If quality-audit shows coverage data: adjust type_coverage and test_proxy
• Recalculate documentation_ok against threshold

3. Revise Complexity

For each dimension with agent evidence, replace heuristic score:

• dependency_depth <- dependency-graph-agent.complexity_contribution
• integration_density <- contract-scanner-agent.complexity_contribution
• historical_churn <- risk-assessment-agent.complexity_contribution
• type_coverage <- quality-audit-agent.complexity_contribution
• test_proxy <- quality-audit-agent.complexity_contribution

Recompute total, reclassify tier.

4. Aggregate Risk Factors

Collect all risk_factors from agent reports:

• Deduplicate
• Rank: critical > high > medium > low
• Cap at 10 most significant

5. Produce Decomposition Plan

Map revised tier to agent configuration. Generate architect_guidance from top risk factors. Generate challenger_questions from shared contracts and circular deps.

Output Format

{
  "skill": "bookend-reconcile",
  "status": "success",
  "reconciled_readiness": {
    "original_score": 85,
    "revised_score": 78,
    "documentation_ok": true,
    "agents_contributed": ["quality-audit-agent", "cartography-agent"]
  },
  "reconciled_complexity": {
    "original_score": 64,
    "revised_score": 72,
    "original_tier": "complex",
    "revised_tier": "complex",
    "dimension_revisions": []
  },
  "decomposition_plan": {
    "tier": "complex",
    "suggested_agents": 3,
    "requires_cartography": false,
    "risk_summary": "...",
    "architect_guidance": [],
    "challenger_questions": []
  }
}

Decision Logic

All agent reports missing?
    → Return original scores unchanged, flag "no_agent_evidence"

Only cartography ran?
    → Revise readiness only, keep complexity unchanged

Multiple agents ran?
    → Full reconciliation with dimension replacement

Token Efficiency

• Reads JSON reports only, no filesystem scanning
• Sub-second execution
• Deterministic merge logic