/work-on-issue
Work through these three steps in order. Do not skip ahead to writing code.
1. Pick an issue from the brain
If I gave you an issue number as an argument, fetch it with get_issue.
Otherwise, look at the brain for this project (start_session if you haven't called it yet this session, list_issues otherwise) and pick a candidate:
- An issue already
in_progresswins — resume it rather than starting something new. Read its comments for where it left off. - Otherwise prefer what the status doc's "What's next" names, then milestone issues over loose ones.
- If several are equally sensible, show me the top 2–3 with one line each on why, and let me choose.
2. Ground it in the code
Before proposing anything, check the issue's claims against the actual repository:
- Read the issue body and comments in full (
get_issue), plus any decisions or docs it references (searchthe brain if it names concepts you can't place). - Explore the code the issue touches. Establish what already exists, what is partially implemented, and what the issue assumes that has since changed — issue bodies go stale.
- Note anything the issue's spec gets wrong or leaves undecided.
3. Report and ask before starting
Give me a short summary: the issue, what's already in place, what remains, your proposed approach, and anything stale or undecided in the spec that needs my call.
Then ask whether to start. Do not edit code, mark the issue in_progress, or make any other change until I say yes. Once I confirm, set the issue to in_progress and begin.