The Build pipeline
Once a Story is refined, /we:build takes it from plan to reviewed pull request — largely on its own. This is the one altitude with no Solo and no Council. There is a single autonomous mode, and it runs the whole chain.
What runs, step by step
Section titled “What runs, step by step”/we:build loads the Story’s plan and works through the pipeline:
- DoR check — is the Story actually ready to build?
- Git prep — branch off, clean working tree.
- Develop — implement the plan, inline or as parallel sub-agents when the plan declares parallel groups.
- AC verification — check the result against every acceptance criterion. This is a gate, not a glance.
- Simplify — a cleanup pass before the gates.
- Quality gates — code review, static analysis, and tests, run in parallel. All three must pass.
- Docs — update what the change touched.
- PR — open the pull request.
- CI review — collect findings from CI and Claude review, batch-fix, and only push when everything is addressed.
The safety rails
Section titled “The safety rails”Two mechanisms keep an autonomous run honest.
Checkpoints. The pipeline records its progress. If a run is interrupted — your laptop sleeps, a session ends — it resumes from the last completed step instead of starting over.
Circuit breaker. When the same step fails repeatedly, the pipeline stops and asks instead of thrashing. Three strikes and it hands control back to you. An autonomous pipeline that refuses to give up is just a slow way to make a mess.
The acceptance criteria are the contract. Code does not reach a PR until it has passed the AC gate and every quality gate — review, static analysis, tests, green.
What stays human
Section titled “What stays human”The pipeline is autonomous up to the PR. It is not autonomous past it. Claude opens the pull request; you review it, you merge it, you close the ticket. Delivery is human-only by design — the AI never merges and never closes on your behalf.
/we:build ships in the we plugin.