rt agent
rt › agent
Hand a prompt to a coding agent, claude or codex (herdr pane or headless), and keep the receipt
Usage
rt agent <verb> <id> [flags]
Arguments & flags
| Flag / Arg | Type | Default | Description |
|---|---|---|---|
<verb> | text | The agent action to run | |
<id> | text | For resume/show: the handoff id or session uuid | |
--repo | text | Repo path (default: the current repo) | |
--prompt | text | Initial prompt (required for headless) | |
--prompt-file | text | Read the prompt from a file (mutually exclusive with --prompt) | |
--surface | text | Where the agent runs (default herdr) | |
--provider | text | Which CLI to launch (override agent.provider, default claude) | |
--model | text | Override agent.<provider>.model | |
--effort | text | Override agent.<provider>.effort | |
--yolo | boolean | Bypass permission prompts for this launch (override agent.<provider>.yolo; omitted falls back to the setting, not false) | |
--no-yolo | boolean | Keep permission prompts for this launch even when agent.<provider>.yolo is true (omitted falls back to the setting, not false) | |
--account | text | cswap account, claude only (override agent.claude.account) | |
--label | text | Caller's display label; used as the herdr tab name | |
--caller | text | Identifies what invoked this handoff | |
--workspace | text | herdr workspace label (default: the repo label) | |
--tab | text | herdr tab label (default: the label or handoff id) | |
--extra-args | text | Opaque extra provider arguments appended to the launch (override agent.<provider>.extraArgs) | |
--bg | boolean | false | Launch onto the daemon-owned background herdr server instead of the visible one (herdr surface only; requires the rt daemon) |
--json | boolean | false | Emit the record as JSON |
See code: commands/agent.ts › agent
Examples & notes
rt agent launches a coding agent session and keeps the receipt.
Providers
--provider selects which CLI to launch. The default is claude (Claude Code). codex launches OpenAI Codex CLI instead. Each provider has its own settings namespace (agent.claude.*, agent.codex.*) for model, effort, yolo, account, and extra args.
Permission bypass
--yolo bypasses permission prompts for the launched session. --no-yolo keeps them even when the setting enables yolo. Both override the provider-scoped setting agent.<provider>.yolo. When neither flag is passed, the setting decides.
Background mode
--bg launches the agent onto the daemon-owned background herdr server instead of the visible one. This is useful for agents that don't need a visible terminal.