← ClaudeAtlas

orchestrated-deliverylisted

Plan, dispatch, monitor, reconcile, and close multi-agent or multi-chat repository work. Use when a brain or lead chat owns a larger outcome, when work must be split across subagents or separate Codex tasks, when a task graph has parallel lanes, or when child handoffs must be integrated into one verified repository result. Do not use for a small self-contained change that one agent can finish directly.
dills122/ai-central · ★ 0 · AI & Automation · score 72
Install: claude install-skill dills122/ai-central
# Orchestrated Delivery Keep the lead task responsible for requirements, integration, and final delivery. Move bounded work to child tasks without letting chat history become the source of truth. ## Classify Every Assignment Choose one delivery unit before dispatch: - **Internal subagent:** bounded work inside the lead task's branch and worktree. The lead owns conflicts, commits, verification, and the final PR. - **Independent task:** a user-visible outcome with its own branch or worktree and complete delivery handoff. - **Research task:** read-only or experimental work that returns evidence and a recommendation. Convert it before retaining implementation changes. Delegate only when the user, applicable repository instructions, or runtime policy authorizes it. Prefer parallel agents for independent exploration, review, tests, and research. Serialize work that shares files, contracts, migrations, or mutable state. ## Build The Execution Index Read the repository's canonical specs, plans, ADRs, and applicable `AGENTS.md`. Write or update one repo artifact that records: - objective and completion boundary; - work items with stable IDs; - dependencies and parallel lanes; - owner or delivery-unit type; - acceptance and verification commands; - current status, blockers, and integration destination. Do not duplicate product truth into the execution index. Link to the canonical requirement or decision. ## Dispatch A Complete Packet Give every child: ```markdown Work item: