overnight

Build or update a Vercel-hosted static wiki for Git-tracked implementation plans, with left-sidebar accordion navigation from plan index to docs, task queues, and individual task pages; include validation, protected hosting, and review gates.

Run implementation work unattended/overnight with a review-driven PR loop. The agent picks one reviewable slice, owns its branch through an agnostic code review (CodeRabbit CLI or a fresh independent reviewer), may integrate agent-owned non-main branches after green gates, never merges to main, and reports a structured closeout. Replaces the retired kanban/dispatcher/polling model. Use when prompting an agent to work autonomously against a task queue or implementation plan.

Agent-owned CodeRabbit review loop for implementation branches. Use when reviewing or fixing a branch before closeout, with or without a hosted pull request.

Selects and applies the repo's engineering quality lens for implementation-plan Goals. Use when planning, reviewing, or starting a task and you need to choose TDD, diagnosis, architecture, DDD, DDIA, Release It, refactoring, or service-layer guidance.

Build or update a complete repo-native implementation plan from product intent, research, PRDs, specs, or conversations. Use when the user asks to create an implementation plan, roadmap, task queue, launch-ready overnight slices, plan index, decision log, or self-contained plan package for autonomous agents.

Interactively install or upgrade the Overnight autonomous orchestration system in a repository. Use when the user asks to set up Overnight, install the skill library, port overnight orchestration, configure AGENTS.md, wire plan preflight, choose plan roots, or make another repo ready for autonomous implementation-plan work.

Convert product conversations, PRDs, plans, or specs into repo-native implementation-plan artifacts and TASK_QUEUE.md rows instead of GitHub/Linear/external issue tracker tickets. Use when asked to create a PRD, break down a PRD/spec/plan into implementation work, create agent-ready tasks, adapt issue workflows to this repo, or maintain PRD-to-task traceability.

Optional layer for running MULTIPLE PRs at once. The per-PR engine (decompose, draft PR, loop on CodeRabbit comments to approved+green, never merge to main, allowed non-main integration) lives in overnight-agent-runbook and needs no orchestrator for serial/stacked work. Use this only to run independent PRs in parallel — it starts one native agent thread/session per PR in the human-selected harness, each with its own isolated working copy, and sequences non-main integrations by dependency.

Reviews implementation-plan work for module depth, service-layer boundaries, DDD seams, data/runtime contracts, and safe refactoring opportunities. Use when a task touches shared architecture, repeated mechanics, domain modeling, persistence, runtime boundaries, or risky existing code.

Use when multiple workflows duplicate the same operational logic, when deciding what belongs in actions vs shared services, or when refactoring repeated operational blocks across domain flows. Use when adding new features that share mechanics with existing ones.

Disciplined diagnosis loop for bugs, failing tests, regressions, or performance problems. Use when something is broken, flaky, slow, throwing, or producing wrong output.

Runs behavior-first test-driven development for implementation-plan feature slices. Use when building behavior, fixing a bug with a known seam, adding integration tests, or when a prompt asks for red-green-refactor.