- 01
/plugin marketplace add Automattic/radical-pipelines - 02
/plugin install radical-pipelines@automattic
Runs a software issue through five reviewed stages — intent, spec, design, build, and documentation — dispatching a separate agent to draft and check each one before moving on. The pipeline only advances once a stage is approved, and it finishes with a pull request ready to open.
WHEN TO USE
USE FOR
- Turn a bug report or feature request into a pull request without writing the spec, design, or code by hand.
- Fix a CI failure or address pull request review comments, then have the fix reviewed automatically.
- Check which pipelines are running, what phase each is in, and what work is still pending.
- Feed a correction into a pipeline partway through and let the affected stages re-converge.
- Write or edit the issue definition that a new pipeline will be built around.
- Decide whether to open, merge, or close out a pipeline's finished pull request.
This is a saved copy of the skill's SKILL.md instructions. It may differ from the upstream file or the version you install.
Radical Pipelines
Overview
You are an orchestrator. A pipeline is a set of artifacts that converge: it is done when every artifact through the target phase exists, is approved, is fresh with respect to its inputs, its tasks are executed, every in-scope challenge and claim is resolved, and every commit after its base outside the pipelines folder is claimed by a task report. Your loop is always the same — take the first item of the frontier and dispatch the agent that resolves it.
Rules
- Humans only talk with you, never with the other agents.
- You never produce artifacts in the autonomous workflow; agents do. You compute state, dispatch, stamp, and merge. Read artifacts to prepare prompts, dispatch, and land work; judge their content only at triage and owner escalation.
- State lives in the working tree.
reference/run/state.mddefines it;scripts/rp.mjscomputes it, and itscheckis the only source of the frontier: when the tree contradicts it or it names something you cannot dispatch, stop and report a defect in the skill. Run policy is the only thing you carry in conversation. - In triage, finish reading and scanning before confirming the run; collect every confirmation question and ask once. Once autonomous work is dispatched, ask nothing until an owner escalation.
- Every agent instance is fresh and sealed: it sees its profile and the prompt you build from its template, nothing else.
Phases
| # | Phase | Folder | Artifacts |
|---|---|---|---|
| 0 | Intent | 0-intent | intent.md, corrections |
| 1 | Spec | 1-spec | spec.md, spec-research.md, reviews |
| 2 | Design doc | 2-design-doc | design-doc.md, design-doc-research.md, reviews |
| 3 | Build | 3-build | build-plan.md with its tasks, build-plan-research.md, plan reviews, task reports, code, build reviews |
| 4 | Document | 4-document | document-plan.md with its tasks, document-plan-research.md, plan reviews, task reports, documentation, document reviews |
Two workflows advance a pipeline: autonomous (reference/run/loop.md) and assisted (reference/run/assisted.md, spec and design doc only). The owner chooses at triage.
Project conventions
Each project supplies its conventions in .rp.md. Load them now: read reference/conventions/load.md.
Entry points
| When the owner wants to... | Read |
|---|---|
| Create or modify an issue | reference/entries/manage-issues.md |
| Inspect pipelines: status, history, what is pending | reference/entries/report.md |
| Work: an issue, PR feedback, a CI failure, a bug, a correction | reference/entries/triage.md |
| Open, merge, or close a pipeline's pull request | reference/run/close-out.md § Closure actions |
No requirements beyond installing this skill were found.
- Listed as free
- Connected services may require an account or charge fees.
- Installation
- 2 commands — the skill arrives with its plugin.
This skill ships inside the Radical Pipelines plugin. Listed package contents include 13 agents. Installed components and skill versions may vary by AI tool.
Install as a Claude Code plugin
Typed inside the agent's own prompt, not in a terminal. This skill uses files kept elsewhere in its plugin, so install the whole plugin — on its own the skill would be missing them.
The repository is licensed under GPL-3.0 — use, modification, and redistribution are allowed under that license’s terms.
View the repository’s license file on GitHub →