Plugin: detect and reject stale worktree paths on resume

[huntharo]

Summary

Adds an early guard in requestConversationBinding() (src/controller.ts) that checks whether a worktree-shaped workspace path still exists on disk before binding the conversation. Returns a clear error immediately instead of letting Codex fail later with a cryptic shell or git error.

Why this matters

From #27: a thread can be resumed even if its recorded cwd no longer exists. The failure only shows up later when Codex tries to execute shell commands in the missing directory — at which point the error is confusing and hard to connect back to the missing worktree.

Early detection turns a cryptic downstream failure into an immediate, actionable message:

Cannot resume: workspace path no longer exists on disk.
`/path/to/worktrees/bold-bartik/repo`

The worktree may have been removed. Check your local paths or start a new session.

Changes

• src/controller.ts: added existsSync import from node:fs. Added guard in requestConversationBinding() (right after the !requestBinding check) that returns { status: "error" } if the workspace path matches the worktree pattern and doesn't exist on disk.
• src/controller.test.ts: added a test covering resume with a nonexistent worktree path — asserts the error reply is returned before any binding proceeds.

The check is scoped to worktree-shaped paths (/worktrees/branch/repo) using the existing isWorktreePath() helper, avoiding false positives on configurable workspace roots that may not be local paths.

Testing

All 122 tests pass (pnpm typecheck + pnpm test).

Fixes #27

This contribution was developed with AI assistance (Claude Code).

[mvanhorn]

Summary

The early guard catches the stale worktree case before Codex hits a cryptic shell error downstream. Scoping to isWorktreePath() avoids false positives on non-worktree workspace roots. Error message is clear and actionable.

Recommendation

Looks good to merge.

---

Review assisted by Claude Code