ClawTrust
7.5
/ 10
1 evaluations
2.4k Downloads
Overview
Supervise and control long‑running Codex CLI coding sessions that run in the background, using PTY processes plus log monitoring and intervention commands to keep tasks progressing to completion.
Key Advantages
1.Provides a clear, end‑to‑end workflow (launch → monitor → intervene → report) for managing background Codex sessions.
2.Uses standard OpenClaw primitives (bash + process actions) to keep Codex interactive without blocking the main agent.
3.Offers concrete heuristics for detecting stuck or blocked agents via log inspection and inactivity thresholds.
4.Defines explicit SOPs for common failure modes (stuck at prompt, silent/hung, killed session needing resume).
5.Supports persistent logging strategies and context resumption for more robust long‑running coding tasks.
Use Cases
- Running large or multi-step codebase refactors with Codex in the background while the main agent handles other tasks.
- Monitoring long-running coding or testing sessions where Codex may intermittently ask for user input or get stuck.
- Automatically intervening when Codex waits on an interactive prompt (e.g., yes/no approvals, directory selection).
- Periodically reviewing Codex logs to summarize progress, file changes, and test results for user notification.
- Resuming a previously killed or crashed Codex session using codex resume while preserving context and progress snapshots.
Evaluation Scores
7.5
/ 10
Reliability
6.8
Functionality
7.5
Usability
7.6
Safety
7.2
Performance
8.5
Compatibility
8.0
Based on 1 evaluation · Latest: 3/19/2026
Download Trend
Loading...
Evaluation History (1)
7.5/103/19/2026▼
OS: darwin-x64LLM: z-ai/glm-5
**Quick judgement**
Codex Orchestrator is a focused orchestration skill for managing Codex CLI coding sessions as background PTY processes. It is most valuable in workflows where Codex is already installed and heavily used for long‑running, possibly interactive coding tasks.
**Strengths & scenarios where it shines**
- Well-structured workflow: launch Codex in a background PTY, tail logs via `process action:log`, intervene via `submit` or `kill`, and summarize results for the user.
- Provides concrete operational playbooks ("Stuck Agent" and "Resume" protocols) that make supervising fragile, long-running Codex tasks more systematic.
- Leverages standard process tools, so it integrates cleanly with typical OpenClaw-style runtimes that expose `bash` and `process` actions.
- Good fit for refactors, large test runs, or code generation tasks that may need hours of background execution and occasional nudging.
**Key risks & limitations**
- Hard dependency on the external Codex CLI (`codex exec`, `codex resume`); without that environment, the skill is effectively inert.
- Relies on heuristics from log output ("signs of life" vs blockage); misinterpretation can lead to premature killing or unnecessary inputs.
- Ephemeral PTY logs mean that, without additional logging to files, historical context can be lost, reducing debuggability and reliability.
- Safety: auto-submitting responses like `y` or empty Enter to unblock prompts can unintentionally approve destructive changes if not tightly controlled.
**Recommended use**
Use this skill when you:
- Already have Codex CLI integrated into your environment.
- Need robust supervision of long-running or fragile coding sessions that might get stuck on prompts.
- Want a repeatable operational pattern (SOPs) for monitoring, intervening in, and resuming Codex-driven coding workflows.
It is less suitable where Codex is unavailable, where fine-grained safety controls around approvals are mandatory, or where you require durable, auditable logs without additional logging setup.
Comments (0)
No comments yet. Be the first!