ARX version
0.1.27
Operating system
macOS (Apple Silicon)
AI tool you were using
OpenAI Codex
Steps to reproduce
Running a multi-prompt session Codex (e.g. a planning session with asking the agent to ask one question at a time and replying to them one by one).
Expected behavior
- Arx reports the timeline in a correct order with stops and timing.
- Arx measures the time of the whole converstation, not just until the first reply.
Actual behavior
- Arx reports the timeline in a seemingly wrong order (the stop is the last entry although the timestamp is from earlier).
- Arx reports at the stop entry only the time of the first response, it seems to ignore the rest of the converstation.
- Arx otherwise correcrly reports the tool calls and the prompts and responses, but the timeline and so the scorecard is possibly wrong.
- It is somewhat confusing that the timeline showing an interrupted session. I could not produce a non-interrupted session, even with a
codex exec which just runs a prompt and then exits, it has the same behavior. Not sure what is the expected here.
Relevant logs or output
Before submitting
ARX version
0.1.27
Operating system
macOS (Apple Silicon)
AI tool you were using
OpenAI Codex
Steps to reproduce
Running a multi-prompt session Codex (e.g. a planning session with asking the agent to ask one question at a time and replying to them one by one).
Expected behavior
Actual behavior
codex execwhich just runs a prompt and then exits, it has the same behavior. Not sure what is the expected here.Relevant logs or output
Before submitting