Plan: Return main repo to master after phase execution#

Summary#

After PhaseRouterService.executePhase() completes, the main repo checkout (/home/agent/repos/<repoName>) can be left on a feature branch when the phase ran without an isolated worktree. Add post-phase cleanup in the finally block to git checkout master when execution used the main repo directly.

Files#

Steps#

1. S1: Add resetToMaster() to WorktreeService — Add a public method resetToMaster(repo: string) that resolves REPOS_DIR, computes repoPath, and runs git checkout master. Wrap in try/catch: on failure, log a warning and return (never throw).
2. S2: Call resetToMaster() in PhaseRouterService.executePhase() finally block — After the existing worktree cleanup (line ~437), add a new block: if !worktreePath (phase ran in main repo), call this.worktree.resetToMaster(task.repo). Wrap in try/catch with warning log, same pattern as the worktree cleanup.
3. S3: Add unit tests — In phase-router.service.spec.ts: test that resetToMaster is called after a non-worktree phase completes, and that a thrown error is caught and logged. In worktree.service.spec.ts: test resetToMaster calls git checkout master with the correct cwd, and test that failures are swallowed with a warning log.

Verification#

• npm run test passes with new tests covering the reset-to-master path
• npm run lint passes with zero warnings
• Manual inspection: resetToMaster is only called when !worktreePath (main repo execution), never for isolated worktree phases

Risks#

• If the main repo has uncommitted changes, git checkout master will fail — mitigated by catching the error and logging a warning (acceptance criterion)