subagent-dispatch

Protocol for delegating work to subagents: picking the model per role, writing a dispatch prompt that carries the task instead of the session history, handing artifacts over as files, keeping a durable ledger that survives context compaction, handling the four report statuses, and capping the fix loop before it turns into churn. USE FOR: dispatch a subagent, delegate to a subagent, subagent prompt, which model for a subagent, subagent model selection, subagent context, subagent report, task brief, review package, subagent ledger, resume after compaction, parallel subagents, fix loop, re-review, blind re-performance, independent recomputation, the agent said it was done, verify a subagent's work, runSubagent, controller context. DO NOT USE FOR: the review rubric itself (use code-review-and-quality), authoring a Custom agent definition or a Skill (use skill-creator), measuring whether a Customization works (use agent-evals), monitoring a long-running process (use long-running-job-monitor).

Compatibility

Requires a harness that can dispatch subagents and select their model explicitly, such as the VS Code Copilot chat agent surface or the GitHub Copilot CLI. Ledger and artifact handover assume a writable working tree.