What problem does it solve? Coordinating one substantial outcome across multiple agents, issues, branches, PRs, or execution runtimes often leads to lost state, duplicate dispatches, stale retries, and unverified completion claims. This Skill keeps one authoritative work graph with explicit gates, recoverable dispatches, and evidence-based acceptance so delegated work stays supervised and reconciled. ## Core Features & Use Cases - Minimal Work Graph Model: Define WorkNodes, dependency edges, Gates, and Dispatches with explicit states, evidence contracts, and dispositions instead of a second task database. - Safe Recovery and Retry: Apply read-before-retry recovery that inspects live runtime state, dispositions stale attempts, and preserves one authoritative dispatch per node. - Repository Lane Mapping: Map issues to WorkNodes, branches and PRs to Dispatches, and CI or review events to evidence, with collision review before parallel dispatch. - Use Case: Hand a set of current tasks spanning several branches and agents to the supervisor, which frames the graph, serializes shared-file work, dispatches bounded execution attempts, and reconciles completion against named evidence. ## Quick Start Ask the assistant to plan, route, and supervise your current set of cross-agent tasks as a work graph with explicit dependencies, gates, and evidence requirements.
