Goal
Make MonkeyProof production-ready as the remote coding-agent worker substrate for Herdr/A2A-controlled runs.
Context
MonkeyProof already exposes REST + WebSocket session control for CLI coding agents:
POST /sessions
GET /sessions
GET /sessions/:id
POST /sessions/:id/input
DELETE /sessions/:id
WS /sessions/:id/ws
For the Herdr controller architecture, MonkeyProof should be the remote execution substrate, while Herdr stays the local cockpit and A2A provides the standard task protocol.
Work
- Verify Codex presets are current and launch the intended Codex binary/flags on the remote host.
- Add or document a standard remote-worker preset for Codex low/no-thinking style implementation work where selectable.
- Harden session lifecycle semantics:
- list running/exited/killed sessions;
- kill runaway sessions;
- preserve recent output and transcript access where applicable;
- avoid unbounded session/output growth.
- Document the standard request bodies for:
- one-shot Codex worker;
- interactive Codex worker;
- follow-up input;
- read/kill session.
- Confirm auth/token loading through environment or approved secret source; do not rely on dev defaults for production.
- Add focused tests or smoke scripts for create/list/read/input/kill flows.
- Document host/deployment assumptions for CT 120 and any supported remote hosts.
Acceptance Criteria
- README or docs include a Codex remote-worker recipe that does not require ad hoc curl discovery.
- A one-shot Codex session can be created and read through the API.
- An interactive session can receive follow-up input.
- A session can be killed and later inspected.
- Production service config does not depend on development token defaults.
- Validation evidence is captured in the PR.
Goal
Make MonkeyProof production-ready as the remote coding-agent worker substrate for Herdr/A2A-controlled runs.
Context
MonkeyProof already exposes REST + WebSocket session control for CLI coding agents:
POST /sessionsGET /sessionsGET /sessions/:idPOST /sessions/:id/inputDELETE /sessions/:idWS /sessions/:id/wsFor the Herdr controller architecture, MonkeyProof should be the remote execution substrate, while Herdr stays the local cockpit and A2A provides the standard task protocol.
Work
Acceptance Criteria