What problem does it solve? Multi-agent development teams suffer from false consensus, where agents approve work without executed evidence and duplicate shallow reviews masquerade as independent coverage. This reference bundle gives every devkit skill a single canonical source for typed evidence output profiles and a task-to-skill routing index so agents emit consistent, verifiable results. ## Core Features & Use Cases - Canonical Evidence Profiles: Defines the required fields for planning-artifact, production-handoff, review-result, qa-result, close-ack, and na-result outputs, marking each field as bus-validated or skill-policy. - Skill Routing Index: Maps tasks (new code, refactor, CI fix, QA strategy, reviews, release close) to the correct devkit skill, with negative triggers, capacity guidance, and dual review-mode rules. - Use Case: A lead agent closing a release milestone reads the routing index to assign review lenses, then each reviewer emits a review-result whose bus-validated fields (risk_class, release_blocker, tests_executed, evidence) pass the agenttalk gate validator. ## Quick Start Read the routing reference to pick the right devkit skill for your task, then follow the evidence profile for that skill when emitting your result.
