plan-next

Plan the next issue

One planning iteration: claim → explore → write plan → push → request review. Do not implement anything in this mode.

1. Read configuration

Read .microfactory/config.yaml in the project root. If it does not exist, tell the user to run /microfactory:init-factory and stop.

Load the backend skill matching backend (jira-tasks, github-tasks, or todo-tasks) for all task operations below.

2. Prepare the working tree

• If the working tree has uncommitted changes, report it and stop this iteration — never plan on top of someone's half-done work.
• Pull the latest default branch.

3. Claim an issue

• No argument: claim the next planning-eligible issue using the backend skill's claim protocol (planning mode). If there is no eligible issue, say "No planning work available." and end the iteration — the loop cadence handles waiting.
• With an issue key argument: view that issue, assign it to yourself, and transition it to Planning (best effort) instead of searching.

4. Write the plan

Explore the codebase to understand what the issue touches, then write plans/<KEY>.md with these sections:

• Summary — what the issue asks for and the chosen approach
• Files to Change — concrete files, with what changes in each
• Implementation Steps — discrete, ordered steps
• Testing — tests to add or modify
• Risks / Edge Cases — risks, edge cases, dependencies

The plan must be executable by another agent with no context beyond the repo and the plan itself.

5. Publish and request review

1. Commit only the plan file (Add plan for <KEY>) and push to the default branch.
2. Comment on the issue with a link to the plan file (derive the blob URL from the git remote when it is a GitHub repo; otherwise mention the repo path plans/<KEY>.md).
3. Transition the issue to Awaiting Plan Review.

Steps 2–3 are best effort: if they fail, warn and finish — the pushed plan is the important artifact. A human reviews the plan and transitions the issue to Plan Approved (or edits the plan); implement-next picks it up from there.

Handle one issue per iteration.