herdr

First verify the agent runs inside a Herdr-managed pane by checking that HERDR_ENV equals 1. Then use the herdr CLI on PATH to list workspaces, tabs, panes, and agents, and issue control commands that return JSON.

Split a sibling pane with herdr pane split, read the new pane ID from .result.pane.pane_id, then run herdr agent start with a unique name, a supported kind, and the pane ID. The command returns once Herdr detects the agent is ready for input.

Pane commands control bare terminals such as shells, tests, and servers, including input and output. Agent commands control a recognized coding agent occupying a pane and expose lifecycle states like idle, working, blocked, done, and unknown.

No. The skill requires running inside a Herdr-managed pane with HERDR_ENV=1. If the check fails, the agent must state it is not inside Herdr and stop rather than probing or controlling the focused session from outside.

A prompt sent to an agent not in the working state must produce an observable lifecycle change within five seconds. If none occurs, Herdr returns agent_prompt_stalled instead of waiting indefinitely, so inspect agent get and agent read before retrying.

Agents running on the alternate screen do not write exited lines into the host scrollback buffer, so increasing --lines cannot recover that content. As a fallback, ask the agent to write its full reply to a temp file and read the file directly.