Harness Session

Use session files for handoff state only. A session is not a log.

The scripts resolve one shared harness, serialize registry mutations with a common file lock, and write files atomically so agents in different worktrees cannot overwrite one another.

Before Work

1. Resolve the shared path with the harness-init resolver or the session script.
2. Read <harness>/index.md.
3. Read <harness>/sessions/registry.md.
4. Read an active session only when its worktree, branch, scope, or user request matches.
5. Create a new session when no active session matches.

If the session runs in a Git worktree whose development environment is not ready or isolated, use worktree-setup before implementation work. Keep setup behavior in canonical project automation and documentation, not in the session note.

Use: