subagent-pipelinelisted

# Subagent Pipeline Use this to execute one issue end to end with independent, cold-start subagents instead of one agent implementing and self-reviewing. ## Why Cold-Start An agent that wrote the code cannot judge its own intent objectively. A fresh reviewer agent that has only the issue and the diff can only judge what the code does, not what the implementer meant. That gap is where bugs hide. Do not let the implementer review its own PR. ## Preconditions - The issue exists and is claimed (`picked by agent` or the project's equivalent label) before any subagent starts. - Check the issue's stated dependencies. If a blocking issue is still open, resolve it first or get explicit sign-off to build against a stub. - Confirm which branch is actually current. If the local checkout is behind the deploy target (for example a solo-maintained `staging` that lags a local feature branch), branch the worktree from the remote target branch, not from the locally checked-out branch. ## Workflow 1. **Implementer subagent** — cold start, full write access, own git worktree per issue (for example `/tmp/<repo>-issue-<n>`), branched from the target base branch. Give it: the issue body, the base-branch commit to branch from, relevant existing conventions and file paths already discovered by prior research, explicit out-of-scope boundaries, and the exact PR target branch. It writes code, writes or updates tests, and commits without AI co-author lines — but does not push or open a PR yet. 2.