What problem does it solve? Turning a recurring multi-step process into a background agent workflow is error-prone: graphs fail on ambiguous edges, missing entry nodes, wrong timeouts, and quota deaths mid-run. This Skill guides an agent to translate a user goal into ordered specs and a validated, persisted Canopy loop graph that survives real execution failures. ## Core Features & Use Cases - Spec authoring contract: Structures every spec as ROLE / WHAT / HOW so a colder, cheaper model can execute it without the author's context. - Reusable graph patterns: Provides eight field-tested patterns (feature delivery, bugfix, gated implement, resilience branch, cross-platform check) with routing rules learned from real broken runs. - Pre-run validation and recovery: Ships a nine-rule fatal-shape checklist and a recovery matrix covering zombie runs, failed loops, and quota-based autorun scheduling. - Use Case: A user asks to orchestrate a developer/reviewer/verifier pipeline; the Skill splits the work into specs, picks the gated-implement pattern, persists the loop with loop_create/loop_add_node/loop_add_edge, validates it, and summarizes before running. ## Quick Start Ask the agent to plan a Canopy loop that implements a feature with build-and-test verification, review, and automatic retry on failure.
