pipeline

The implementer's own quality + correctness pass over its UNCOMMITTED work, before it commits. Use when an implementer in the worktree flow has finished writing a change and is about to commit it, whenever an `orchestrate` brief says to run a review pass for the slice, and whenever you are asked to review, tighten, simplify, or clean up a change you just wrote and have not committed. Reviews reuse, simplification, efficiency, altitude AND correctness in one pass over the working tree, then applies what it judges right and reports what it rejected. It NEVER dispatches sub-agents and NEVER commits — the agent running the slice is the one that decides what goes in, and the commit step belongs to the flow that called this. Not for reviewing a committed range or someone else's PR (that's the dispatcher reading the diff, plus the gate).

Onboard a repo onto the /pipeline:write-issue → /pipeline:orchestrate pipeline. Use whenever a repo has no `.agents/worktree.json` yet, whenever `setup-worktree.sh` warns "no config", when someone asks to SET UP / ONBOARD / WIRE UP the pipeline (or a gate queue / enqueue system) for a project, or when a dispatcher refuses to dispatch because the project is unconfigured. You GROUND the repo in its real scripts, CI, and AGENTS.md — never guessed commands — then write `.agents/worktree.json` and, when the project wants one, scaffold a durable gate queue INTO THAT REPO. The artifacts belong to the project: this skill carries the knowledge, the repo carries the code, so each project owns and evolves its own queue. You do NOT slice work, write features, or dispatch implementers.