# Sub-Agent Orchestration Use this file only when the platform supports sub-agents and the user or platform policy allows delegation. ## Objective Sub-agents can improve speed and coverage, but only if their work is bounded and synchronized through shared contract artifacts. ## Canonical Shared Inputs Before delegating, create: - one canonical `intake.yaml` - one canonical `document-map.md` - one canonical `priority-profile.yaml` All sub-agents must work from those artifacts, not from separate ad hoc summaries. ## Recommended Agent Split ### Intake Agent Responsibility: - normalize intake - identify missing docs - draft `intake.yaml` ### Playbook Agent Responsibility: - normalize playbook, templates, fallback language, and prior deals - draft `playbook.normalized.yaml` ### Contract Map Agent Responsibility: - identify clause families - list exhibits and incorporated documents - check definitions and cross-references - update `document-map.md` ### Review Agent Responsibility: - perform clause-by-clause comparison - draft issue candidates in `issues.csv` ### Redline Agent Responsibility: - draft preferred and fallback edits in `redlines.md` ### QA Agent Responsibility: - run failure-mode checks - check consistency across issues, redlines, and approvals - draft `qa-report.md` ## Delegation Rules - Delegate only bounded, file-scoped tasks. - Keep one owner for each artifact to avoid merge conflicts. - Do not let sub-agents invent independent priorities or approval rules. - The lead reviewer must integrate and resolve contradictions. - Final legal judgment stays with the lead reviewer. ## When Not To Use Sub-Agents Do not use sub-agents when: - the platform does not support them - the user has not allowed them - the contract is too small to justify orchestration - the document is so sensitive that the user wants a single-agent review only ## Maximum Useful Parallelism In most contract reviews, 3 to 5 bounded sub-agents is enough. More than that usually creates integration overhead unless the contract is very large.