dev-orchestrator tmux restart guidance is incompatible with Copilot CLI command-safety policy

[rysweet]

Summary

The dev-orchestrator instructions recommend using tmux for recipe execution and implicitly encourage restarting or replacing a recipe-runner session. In practice, the obvious cleanup path is tmux kill-session -t recipe-runner, but Copilot CLI's shell safety policy blocks commands containing that verb unless they target numeric PIDs.

Reproduction

1. Follow the dev-orchestrator tmux-based recipe-runner instructions.
2. Need to restart the tmux session because a prior recipe-runner session already exists or the launch command needs to be retried.
3. Try tmux kill-session -t recipe-runner.
4. Observe that the shell policy rejects the command because it does not target a numeric PID.

Actual behavior

The documented tmux workflow lacks a shell-policy-safe cleanup/restart path.

Expected behavior

The skill/docs should provide a safe restart pattern that works inside Copilot CLI, for example:

• always use a unique tmux session name, or
• document how to find the tmux server PID or pane PID and terminate with numeric kill <PID>, or
• avoid instructions that imply the session-subcommand form is acceptable in this environment.

Why this matters

This makes recovery from a failed or stale recipe-runner session unnecessarily fragile and blocks normal retry behavior during long-running orchestrated tasks.