Skill Claude CodeCodex
Overnight autonomous improvement loop for an agent harness - run challenges, judge with a different model, apply bounded meta-improvement, report on a heartbeat. USE WHEN running overnight loops, benchmarking the harness, or building a self-improvement cycle.
Skill Claude CodeCodex
Convene the board ON THE SYSTEM ITSELF - inventory the harness's own recent failures and weaknesses, attack them from eight adversarial angles, and leave with a prioritized determinism backlog and a scaling plan. USE WHEN the same class of mistake keeps recurring, before scaling the system up, quarterly as a health…
Skill Claude CodeCodex
Multi-agent strategic deliberation - spawn a board of specialist officers as parallel subagents to research, debate, and converge on a decision. No code is written during a boardroom session. USE WHEN facing a major decision, evaluating a big feature or external spec, or when the right path is not obvious and needs…
Skill Claude CodeCodex
Git branch management, PR workflow, and the verification discipline that makes autonomous merging safe. USE WHEN creating branches, opening PRs, merging, cleaning up worktrees, or coordinating parallel agents on one repo.
Skill Claude CodeCodex
Socratic retrospective over a session - find what can become deterministic, adopt at most one change. USE WHEN a session ends, when the same mistake happened twice, or nightly over the day's transcripts.
Skill Claude CodeCodex
Plan a mission bigger than one session as a map of DECISION tickets, resolved one at a time until the route is clear - and bind worktrees to tickets so ambition does not become sprawl. USE WHEN work is too big for one session, when you do not know the route to the destination, when the fog is the problem rather than…