Issue solving

Point AM at your GitHub issues and it fixes them: it discovers open issues, picks the highest‑value one, implements a fix on its own branch, runs the Verification Gate, and opens a PR that closes the issue.

Enable it

In Repo settings → Issue solving (Pro plan or higher):

Issue solving — toggle on.
Solutions per issue — how many independent attempts to spawn per issue (Pro = 1; Max = up to 3; Enterprise up to 5). Each solution uses a different approach (minimal → refactored → test‑driven) on its own branch.

A verify command is strongly recommended — it's what makes a generated fix trustworthy.

How it works

Sync — AM pulls open issues from GitHub into a candidate queue, scored by signal (bug label, help‑wanted, comment activity). Pull requests are skipped; issues closed on GitHub are retired from the queue automatically.
Select — the scheduler (or you, manually) picks the top‑scored open candidate. A standing repo objective takes priority over issue solving.
Solve — an agent team implements the fix with the issue context in focus, commits, and the gate must pass.
PR — a pull request opens with Fixes #N in the body, so merging it auto‑closes the issue. The cap (solutions_target) stops AM after the configured number of attempts — no runaway.

Manual control

POST /api/repos/{id}/issues/sync — refresh the candidate queue.
GET /api/repos/{id}/issues — list candidates.
POST /api/issues/{id}/run — solve one now.
POST /api/issues/{id}/reject — skip a candidate.

(Or drive it from the CLI / an agent — see the Agent cheatsheet.)

Issue filter

By default every open issue is a candidate, except those carrying a skip label (am-skip, am-hold, do-not-treat, wontfix, blocked, on-hold). Open Repo settings → Automation → Issue candidates → Issue filter to narrow the queue further (mirrors the PR watcher policy):

Include labels — keep only issues carrying at least one (e.g. bug, good first issue).
Exclude labels — drop issues carrying any (on top of the skip labels).
Authors — keep only issues opened by these logins.
Min comments — require at least N comments (a rough "has discussion" signal).
Max age (days) — keep only issues created within the last N days.

Filters are applied at sync time, work the same for GitHub and GitLab, and can only tighten the queue. An issue that no longer passes is marked skipped; it returns automatically once it matches again (or the filter is relaxed). The filter is stored on the repo as the issue_solve_policy JSON field — settable via PATCH /api/repos/{id} for automation.

Gotchas