← ClaudeAtlas

herdrlisted

Inspect or control Herdr. In a Herdr-managed pane (HERDR_ENV=1), use the local CLI; from ChatGPT Web, Mobile, or Desktop, use the herdr_control MCP Bridge and never fall back to the CLI.
yuanhaorannnnnn/agent-skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill yuanhaorannnnnn/agent-skills
# herdr — agent skill ## runtime routing choose exactly one control path before doing anything: 1. if `HERDR_ENV=1`, this is a Herdr-managed pane. use the `herdr` CLI instructions below. 2. otherwise, if the `mcp__herdr_control__*` tools are available, this is an external ChatGPT client. use only the MCP Bridge instructions below. 3. otherwise, report that the Herdr Bridge is unavailable and stop. when `HERDR_ENV` is not `1`, never inspect or control Herdr through the local `herdr` CLI, shell commands, socket files, or an HTTP fallback. ChatGPT Web, Mobile, and Desktop must always use the MCP Bridge. ### external ChatGPT clients: MCP Bridge use only the capabilities actually exposed in the current tool list: - `mcp__herdr_control__list_agents` — list current agents and status. - `mcp__herdr_control__list_panes` — list current panes and obtain fresh pane ids. - `mcp__herdr_control__read_pane` — read an existing pane. - `mcp__herdr_control__prompt_agent` — prompt an existing agent when the user authorized that write action. - `mcp__herdr_control__pane_send_text`, `pane_send_keys`, `pane_run` — type, key, or run a user-authorized command in an existing pane. - `mcp__herdr_control__pane_wait_output`, `pane_split`, `pane_close` — wait for output or manage a pane; closing requires explicit authorization for the exact pane. pane ids are ephemeral. call `list_agents` or `list_panes` immediately before every pane operation; never reuse an id from an earlier turn. for writes, v