You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 85cf325
Browse filesBrowse the repository at this point in the historyBrowse files
feat(server): agents read failing CI logs and re-run checks through MCP tools
Agents read red CI with gh run view --log-failed or the job logs API (about 500
times in 45 days) and re-ran it with gh run rerun, on the user's own quota and
outside T3's forge sign-in.
ci_logs returns the end of the PR's failing check logs, or of one named check;
ci_rerun re-runs them. Both work from the PR's checks, which every forge reports
by name, so agents never look up run ids.
- PullRequestProvider gains optional getCheckLogs and rerunChecks.
- GitHub: an Actions check run is a job, so its link gives the job's log, and a
run's failed jobs re-run together (GitHub refuses a second job re-run while the
run re-runs).
- GitCafe: logs of checks the GitHub mirror ran come through GitCafe's check-log
endpoint; GitCafe Actions runs re-run, and mirrored checks say to re-run them
on GitHub.
- PullRequestService picks the failing checks (failure or cancelled), or the
named one, and lists the checks when the name matches none.
The tool list grows by another 756 bytes. New mcp-tool-use cases pass on
Sonnet 5.5 (ci_logs 3/4, ci_rerun 4/4, control 4/4; 0/4, 0/4, 4/4 on main).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
0 commit comments