| name | babysitter-breakpoint |
| description | Use the breakpoint API to communicate with users during babysitter runs (post breakpoints, poll for release, fetch feedback, and read context files). Trigger whenever a babysitter workflow needs approval, input, or status updates via breakpoints. |
babysitter-breakpoint
Use this skill to handle all user communication via the breakpoints CLI. Do
not prompt the user directly. Always post a breakpoint and poll until it is
released, then fetch feedback and continue.
Workflow
- Create a breakpoint with the required payload and
payload.context.files.
- Poll
breakpoints breakpoint wait <id> --interval 3 until released.
- Apply feedback from the printed details and continue.
Payload structure
Always include a context.files array for referenced files:
{
"context": {
"runId": "run-...",
"files": [
{ "path": ".a5c/runs/<runId>/artifacts/process.md", "format": "markdown" },
{ "path": ".a5c/runs/<runId>/inputs.json", "format": "code", "language": "json" },
{ "path": ".a5c/runs/<runId>/code/main.js", "format": "code", "language": "javascript" }
]
}
}
See references/breakpoint-api.md for curl examples and endpoints.