Safe phase-based workflow for pull request agents
This is a compact workflow skill for agents working on pull requests. It keeps PR work split into explicit phases so an agent can scout, inspect, patch, prove, and publish without skipping approval gates.
Use it when you want cautious agent-assisted maintenance, bug fixing, upstream PR work, or review handoffs without turning the agent into autopilot.
Skill file | Install | Workflow guide | Output templates | PR body template | Failure patterns | Skill test
| Area | Purpose |
|---|---|
| Scout and gate | Find useful PR candidates and avoid duplicate or owned work. |
| Inspect and patch | Read narrowly, define the smallest safe change, then patch only after GO. |
| Proof and publish | Run proof, prepare PR notes, and publish only the exact approved action. |
Ask your OpenClaw agent to use the skill by name:
Use pr-agent-workflow. Phase 0.
Use pr-agent-workflow. Phase 2 <project or search window>.
Use pr-agent-workflow. Phase 3 <issue or ticket>.
The detailed phase guide lives in docs/WORKFLOW.md.
| Path | Purpose |
|---|---|
| SKILL.md | Main skill entry file. |
| docs/WORKFLOW.md | Full README-style guide with phases, stop conditions, and usage patterns. |
| references/ | Safety rules, phase details, templates, and failure-pattern notes. |
| agents/openai.yaml | Agent metadata/config included with the skill package. |
MIT. See LICENSE.