AI agent skill · by @greenfield-inc
Pane Manage And Message Agents
Use when the user asks to start, kick off, spawn, check on, nudge, follow up with, or babysit another coding agent (Claude, Codex, or Cursor) in Pane, or to create a Pane for that work.
Archived 2026-10-23 · Source updated 2026-10-12. Check the original repository for the latest version.
SKILL.md instructions
# Manage and message Pane agents You coordinate. Another agent in a Pane does the work. If you are already Pane Chat / a Session orchestrator, follow `pane-orchestrator` and `runpane`. ## Trivial work stays here Lookups, one-line nudges, and status checks belong in this turn: list panes, read doctor, send "CI failed on X". When they asked for an agent, that agent owns implementation, debugging, commits, and the PR. "It's not done until a PR exists" stays with that agent until they take the task back. ## Use Pane Prefer Pane MCP (`agents_start`, `agents_status`, `agents_send`, `repos_list`, `panes_list`, `docs_search`). If those tools aren't in this session, use `runpane` the same way (`runpane agent-context --json` first). `yes: true` / `--yes` only when they asked for the change (start a pane, send a message, archive). Discover ids from `repos_list` / `panes_list` before guessing. ## Start, check, follow up - **Start:** `agents_start` with repo, a short pane name, the agent they named (`codex` / `claude` / `cursor`), and the full task. The worker has none of this chat: put goal, constraints, paths, and "done means" in the prompt. Create in the background. - **Check:** `agents_status`. Report what the screen shows. Check once per ask; babysit when they asked you to keep watching. - **Follow up:** `agents_send` for a message. Menus and trust prompts: `panels_input` for keys, and ask the user before accepting a permission prompt. If start fails, say so and wait. ## Hand them the link as text Print the `pane://open?...` URL on its own line, exactly as returned. ## When they say keep going Nudge the same pane with `agents_send`.