When spawning an agent with a name that was recently killed, agent-shell sometimes fails.
Current workaround: retry after 0.5s delay with sit-for (see meta-agent-shell-start-named-agent).
Root cause likely in agent-shell async cleanup. May need to file upstream issue or find proper fix.
Maintain an index file mapping buffer names to transcript paths in ~/.agent-shell/. When a session starts, write {buffer-name -> session-id/path} to ~/.agent-shell/index.json. This would let agent-search and other tools show which buffer each result came from.
Hook into agent-shell session creation, write mappings, add lookup function.
Simple CLI that creates/appends to an agent’s personal notes file. Each agent gets their own file to avoid conflicts: .tasks/agent_<description>.org
Usage: note-for-later “refactor auth” “Found a bug in X, should investigate Y” Creates/appends to .tasks/agent_refactor_auth.org with timestamped entry.
Separate files per agent = no coordination needed, no merge conflicts.