Multi-agent delivery discipline: your AI architects and reviews while external executors implement, under adversarial review-to-GO and real-runtime verification.
Works standalone or alongside superpowers.
Skills 13 in the Claude listing
Adversarial review to go Use when a spec, a plan, or an implemented diff needs an external adversarial reviewer — locks the reviewer to the altitude of what is being reviewed, converges diff findings to zero (GO), re-reviews Checkpoint verification Use when accepting an executor's task or about to claim work done — refuses green tests as proof and requires inspecting the call-site and driving the real runtime path to catch missing wire-ups. Concept briefing Use as the first step on any new request, before design begins — locks a user-confirmed system profile (scale today and expected, users, trade-off priorities) and tiers the request to route it to the Convention commit gate Use at every checkpoint review before committing — enforces centralized enums, no magic string/number literals, constant-based collection/model names, and the project's commit-message style. Executor context Use when dispatching work to executors or subagents and the same project conventions keep getting retyped into every prompt — maintains one fixed context file the coordinator writes and updates, so ha Lessons ledger Use before starting work in an area to retrieve only the lessons that apply to it, and right after a wrong diagnosis, a defect the tests missed, or an environment trap to record one — keeps hard-won l Multi lane coordination Use when the working agreement sets coordination mode to multi-lane, or when design finds two or more services or large independent areas that could each carry their own coordinator — in one repo or a Orchestrating executors Use when a plan is ready and implementation will be delegated to external coding agents or subagents — covers knowing your workforce and what each has proven, choosing between a subagent and an extern Planning for delegation Use after a plan is drafted and before any task is handed to an executor — settles how detailed this project's plans are (asking once and recording it in the working agreement), keeps spec and plan at Pointer handoff Use at the start of any session on an ongoing project to recover where the work stands, and at the end of a session (or on any significant discovery) to record it — maintains a single short pointer fi Project working agreement Use before executing or dispatching work on a project, and whenever the user changes how the work should run — keeps one project file of working rules (stop after each task or continue, parallel or se Using crewmarshal Use when starting delegated multi-agent delivery on a project — maps the full workflow arc and points to the right crewmarshal skill and superpowers skill at each stage. conductor Create, run, monitor, manage, and review Conductor workflows — including agentic workflows (LLM + MCP + tools). 