Verified Agent Skill record

herdr · Herdr

Herdr is a terminal multiplexer that organizes terminals into workspaces, tabs, and panes, recognizes coding agents running inside panes, and exposes session control through the `herdr` CLI. This skill controls Herdr only when the user explicitly mentions it or asks to inspect and control panes, tabs, workspaces, commands, or another agent.

Tasks

Split the current pane into a sibling shell using `herdr pane split --current --direction right --cwd "$PWD" --no-focus`, then start a named coding agent in the returned pane with `herdr agent start <name> --kind <kind> --pane <returned-pane-id>`
Run an ordinary shell command in a separate pane using `herdr pane run <pane-id> "<command>"` and wait for expected output with `herdr pane wait-output <pane-id> --match "<text>" --timeout 120000`

Inputs

HERDR_ENV environment variable set to 1
HERDR_WORKSPACE_ID, HERDR_TAB_ID, and HERDR_PANE_ID caller context variables
A pane ID, agent name, or workspace ID from prior JSON responses
Native agent arguments passed after `--`

Outputs

JSON responses containing workspace, tab, pane, and agent identifiers and state
Agent lifecycle states: idle, working, blocked, done, unknown
Pane output via read sources: visible, recent, recent-unwrapped, detection
CLI server errors as JSON on stderr with exit status 1 and syntax errors with exit status 2

Limitations and checks

Cannot inspect or control the focused Herdr session from outside Herdr
Does not create, split, or move layout for `agent start`; an existing available shell pane is required
CLI reads do not mark a tab as seen, so `idle` and `done` visibility depends on UI focus or a focus command
Rows leaving an agent's alternate screen do not enter host scrollback, so increasing `--lines` cannot recover them
Verify the environment with `test "${HERDR_ENV:-}" = 1` before issuing any control command; if it fails, state you are not running inside Herdr and stop
Confirm any pane targeted for `agent start` is an available shell at its interactive prompt with no foreground command, editor, or agent running
Parse all IDs from JSON responses rather than deriving them from sidebar order or examples, and use `--current`, an explicit pane ID, or a unique agent name instead of relying on another client's focused pane
Do not run bare `herdr` for discovery or probe mutating nested commands by omitting arguments, since commands like `herdr workspace create` will execute with defaults