Reading a run
Open any run to see exactly what happened — the agents' activity, the gate result, and the resulting PR.
Run statuses
| Status | Meaning |
|---|---|
| spawning / running | In progress |
| success | Completed and published a PR |
| success_no_changes | Agents did real work but found nothing to commit (why?) |
| no_op | Agents finished in ~0s with no activity — a likely silent provider/harness failure, not a genuine no‑change run (why?) |
| verify_failed | Gate stayed red; a draft PR was opened |
| pushed | Commits pushed but PR not opened (recover it) |
| failed / interrupted | The run errored or was cut off |
Runs are also tagged by kind — code (the default), issue-scout (audits
the repo and files GitHub issues, no PR), or issue-solve (fixes a tracked
issue). Non‑code kinds show a chip in the run list so a scout run that filed an
issue isn't mistaken for a wasted success.
What you'll see
- Activity / dev output — a summary of the agents' tool calls and a tally
(e.g.
Bash×39 Write×5 Read×1), so you can tell a real run from a no‑op. - BUS — the live peer‑message conversation between agents (dev announcing done, qa reviewing).
- Verification — the gate status, iterations, and (on failure) the output tail.
- PR — the title, body, and link. Titles are derived from the agent's commit message, not the raw focus.
- Cost — token usage and an approximate dollar amount, when the worker has cost capture enabled → Run costs.
Quick reads
- Run is
no_op(finished in seconds, no activity)? Almost always a provider key/balance issue → My run made no changes. A repo that keeps producing no‑progress runs is automatically backed off by the scheduler. verify_failed? The gate is red → Gate failures.pushedbut no PR? → Recovering a PR.