Issue scout
Point AM at a repo and it audits the code and files GitHub issues — no pull request, no code change. It's the inverse of issue solving: instead of fixing existing issues, it discovers gaps worth tracking and opens issues for them.
Enable it
In Repo settings → Features (Pro plan or higher), toggle Issue scout on. The repo's owner must have a GitHub token with issue write access — that's how AM files the issues.
It runs on the repo's normal schedule and respects the spawn interval, so it takes a scheduled slot like any other run. It only runs when the repo has no standing objective set.
How it works
- A scheduled cycle picks the repo and dispatches an
issue_scoutrun. - The worker prepares a read-only worktree and spawns a single agent with the repo's first team role (harness + model).
- The agent audits the code (docs gaps, untracked TODO/FIXMEs, obvious bugs,
missing tests, hardening gaps), lists existing issues to avoid duplicates,
and files up to 5 new issues via
gh issue create. - The run completes with no branch and no PR — the issues are the output. The dashboard shows the run with an issue count (e.g. 3 issue(s) created) instead of a PR link.
What it does not do
- No code changes. The agent never edits files, commits, pushes, or opens a PR. The worktree is read-only and discarded.
- No runaway. A single run files at most 5 issues, and the agent is told to skip anything that duplicates an existing open issue.
Gotchas
- Needs a GitHub token with issue write on the repo.
- Issues are filed on the canonical repo immediately — there's no review step before they appear. Start on a repo where that's welcome.
- It shares the schedule/cooldown with normal runs, so a cycle spent scouting is a cycle not spent on a code run.