What problem does it solve? Capable coding agents often load too much context, inherit stale skill selections, or claim outcomes they never verified. This Skill defines the entry contract for material work: decide per task whether to engage the Agentit workflow, select only the expertise needed just in time, and report results strictly against observed evidence. ## Core Features & Use Cases - Dispatch and context lifecycle: Makes a fresh DISPATCH_DECISION: bare | agentit per session, keeping trivial work direct while routing material work through the Agentit core without inheriting stale selections. - Material execution protocol: Coordinates task-router execution modes, JIT skill body delivery, schema-3 worker delegation, and review-branch publishing with real tool permissions. - Evidence-bound verification: Distinguishes command-bound Loop/Graph evidence (exit status, output digests, timestamps) from self-reported observations, and forbids claims like "done", "secure", or "independent" beyond what was observed. - Use Case: An agent asked to refactor a payment module uses this contract to select only relevant domain skills, delegate bounded context to a worker, run a command-bound verification loop, and open a review branch while honestly reporting that no independent model review occurred. ## Quick Start Ask your agent to install Agentit for this environment and use it automatically for all material work going forward.
