The Failed run path #5

Parent

#2

What to build

The Failed run path, from the spec's "Failure path" and "Cleanup" decisions. Any failure after the worktree exists goes through it:

• In the worktree, commit everything including uncommitted work as the failure commit: thirdshift: failed run (<reason>), a blank line, then <ISO timestamp>, host <hostname>. Uncommitted work at the time of failure is included in this commit. An unfinished merge is aborted first.
• If the Issue branch has no changes against the Base branch, skip the commit and push; otherwise push.
• If an open PR exists for the Issue branch, convert it to a draft.
• Print the most recent session log path to stderr and exit non-zero. Cleanup still runs.

Failure triggers covered here: claude exits non-zero, no PR found, PR targets the wrong base, and SIGINT/SIGTERM (Ctrl-C). Also: a PR the agent left as a draft is marked ready with gh pr ready and the run still succeeds.

Acceptance criteria

• claude exiting non-zero → failure commit with reason claude exited <code>, containing the agent's uncommitted work, pushed to origin
• No PR after the session → failure with reason no PR found; PR against the wrong base → failure naming both bases
• No changes against the Base branch → no failure commit and no push (no Issue branch on origin)
• An open PR is converted to a draft on failure
• A draft PR left by the agent is marked ready, and the run exits 0
• SIGINT while the session runs → session stopped, reason interrupted, same failure path, cleanup done
• Every failure: non-zero exit, log path on stderr, nothing on stdout unless a PR exists, no worktree, local Issue branch, or temp directory left

Blocked by

• Walking skeleton: a fresh run on the happy path #3