skill-debug-codex-requests

Open-source Codex skill for capturing and inspecting the exact provider requests that Codex sends through an OpenAI-compatible proxy.

It is useful when you need to debug:

• request size and payload shape
• injected instructions and system_prompt
• exposed tools and provider overrides
• profile differences from ~/.codex/config.toml
• TTFT, warm generation speed, and near-context behavior

What Is Included

• SKILL.md with the operational workflow
• scripts/codex_proxy.py for request capture and forwarding
• scripts/inspect_proxy_log.py for log inspection
• scripts/run_codex_benchmark.py for multi-phase throughput benchmarking
• references/ with field semantics and benchmark guidance
• agents/openai.yaml for UI metadata

Install

Clone or copy this repository into your Codex skills directory so the folder name stays debug-codex-requests.

One simple layout is:

mkdir -p ~/.codex/skills
git clone git@github.com:ivanopcode/skill-debug-codex-requests.git ~/.codex/skills/debug-codex-requests

Main Flows

1. Capture a fresh Codex request through the bundled proxy.
2. Inspect sanitized system_prompt, instructions, input, tools, and request sizes.
3. Run the multi-phase benchmark for cold TTFT, warm raw speed, and near-context behavior.

License

MIT

Author

Ivan Oparin