| name | agent-step |
| description | Execute a single agent step with custom system prompt and wait for completion. |
| metadata | {"CARNELIAN":{"emoji":"🤖","requires":{"env":["GATEWAY_URL","GATEWAY_TOKEN"]},"primaryEnv":"GATEWAY_URL"},"carnelian":{"runtime":"node","version":"1.0.0","sandbox":{"network":"localhost","resourceLimits":{"maxMemoryMB":64,"maxCpuPercent":10,"timeoutSecs":60},"env":{"GATEWAY_URL":"${GATEWAY_URL}","GATEWAY_TOKEN":"${GATEWAY_TOKEN}"}},"capabilities":["net.ws"]}} |
agent-step
Execute a single agent step with custom system prompt and wait for completion.
Ported from CARNELIAN agent-step-tool.ts.
Input
{
sessionKey: string;
message: string;
extraSystemPrompt: string;
timeoutMs?: number;
channel?: string;
lane?: string;
}
Output
{
ok: true;
reply: string;
}
Or on failure:
{
ok: false;
status: string;
}
Notes
- CARNELIAN dependency: The CARNELIAN gateway must be running
- Default gateway URL:
ws://127.0.0.1:18789
- Calls
agent method, then polls agent.wait for completion
- Reads
chat.history to extract the latest assistant reply
- Filters out tool messages from the response
- Maximum timeout: 60 seconds
- No npm packages needed in the wrapper
- Uses WebSocket JSON-RPC for communication