Task Scope Validator Decision Tree and Dispatch Policy

This diagram documents the pre-dispatch scope check used by the Task Scope Validator agent.

• Source diagram: task-scope-validator-decision-tree-and-dispatch-policy.excalidraw
• Primary source terminology: agents/basecoat-60-workflow-task-scope-validator.agent.md and skills/task-decomposition/automation-fitness-matrix.md

What it covers

1. Ordered checks for single issue, deterministic path, measurable success, and time bound.
2. Classification and confidence bands map to JSON output fields: checks.* drive classification; top-level confidence, approved_for_dispatch, and recommended_agent are the dispatch outputs.
3. automatable + confidence >= 0.8 → approved_for_dispatch: true, recommended_agent: task | general-purpose
4. gather-findings-only + confidence 0.6–0.79 → approved_for_dispatch: true, recommended_agent: explore (explore-only dispatch; task and general-purpose are not valid targets)
5. defer or confidence < 0.6 → approved_for_dispatch: false, recommended_agent: escalate
6. Dispatch mapping from classification to agent choice: task or general-purpose (automatable), explore (gather-findings-only), or refinement/escalation (defer).
7. Remediation path when scope is too broad, success criteria are vague, or timing is unbounded.

Intended operator usage

1. Run the validator before dispatching a sub-agent.
2. Route narrow, measurable work to task or general-purpose, evidence gathering to explore, and ambiguous work back for refinement.
3. Treat defer or confidence < 0.6 as a prompt rewrite signal unless an explicit human override exists.