seda-pr
Seda PR
Own the PR/MR from its requested entry point to its requested stopping point. Optimize the reviewer's understanding, not a changelog of agent activity.
Scope the operation
- Description: return a draft, or update an existing body when explicitly requested. This grants no commit, push, PR creation, state change or babysitting authority.
- Publication: commit scoped work, push and create or update the item. Stop after verified publication unless babysitting was also requested.
- Stewardship / babysitting: investigate CI, conflicts and feedback, make justified in-scope corrections, verify, commit, push normally and reply/resolve with evidence. A watch/status-only request remains read-only, including the body.
Use the explicit target; otherwise reuse the current branch's unambiguous open item. For publication with no item, use the task branch or create a feature branch from the established integration branch or specified stack parent. Ask only for genuinely ambiguous targets or authority. Supplied descriptions/diffs need no live PR.
Preserve unrelated work, human content and the repository template. Approval, merge/close/reopen, draft-state changes, retargeting, history rewrite, force-push, hook bypass, reviewer/assignee changes and unrelated edits need separate authority.
Use gh or glab, falling back to the provider API; discover uncertain mechanics from current help/docs. Delegate bounded investigations only when separate context or expertise earns the overhead; require findings with decisive evidence.
Describe and publish
Read reviewer brief before composing or materially refreshing a body. Reuse the task, actual diff, applicable domain meaning and current proof; gather surrounding context only where needed to explain behavior or ownership.