What problem does it solve? Running small, repetitive governance actions (atomic instincts) inside the main Opus conversation window pollutes its token budget and context. This Skill routes those instinct dispatches to an independent cheap-model agent (Gemini-backed by default, Haiku optional) behind a frozen JSON contract, keeping the main window clean. ## Core Features & Use Cases - Contract-first dispatch: A frozen JSON schema (instinct-dispatch.schema.json) defines the InstinctDispatchInput/Output contract that all 10 L2 atomic instincts must follow before any implementation. - Tri-mode dispatch engine: Route dispatches via the PHYCOOL_DISPATCH_ENGINE environment variable to gemini (default), haiku (claude-haiku-4-5 subprocess), or mock (zero-token testing). - State tracking and evaluation gate: Every dispatch writes a state_diff to .context-db/ecc-state/ and appends to dispatch_log.json; the ecc-haiku-eval.cjs script computes success rate, latency, and token cost against the D4 quality gate. - Use Case: Before implementing an instinct dispatch story, read the contract schema and IInstinctDispatcher module, then call Invoke-InstinctDispatch with an instinct name and payload to get a schema-valid result plus state diff without consuming main-window tokens. ## Quick Start Ask the AI to dispatch the skill_invocation instinct with a trigger context and rule reference using the ecc-haiku-dispatch contract, then verify the returned status and state_diff.
