Summary
When using OpenClaw with the Codex runtime, Opik traces are being created for LLM activity, but Codex/runtime developer tool calls do not appear to be recorded as Opik tool spans.
Example: during an active Telegram/OpenClaw session, multiple shell/tool operations were executed through the Codex runtime, but the Opik project showed only LLM spans after the gateway restart. Historical OpenClaw-visible tools are present, but recent Codex runtime tool calls are absent.
Observed behavior
- Opik plugin loads successfully.
- Gateway logs show exports to the expected Opik project.
- LLM spans appear in Opik.
tool spans for Codex runtime/developer tools such as shell/exec calls do not appear.
- The latest tool spans in Opik were older OpenClaw-visible tools, while newer Codex tool activity was not represented.
Current hypothesis
The plugin appears to instrument OpenClaw typed hooks such as before_tool_call / after_tool_call. Codex runtime/developer tools may be bypassing those OpenClaw hook events, so Opik only sees OpenClaw-visible tool calls and not Codex runtime tool calls.
Question
Should Codex runtime/developer tool data be funneled into Opik separately, or is the intended direction that these tool calls will eventually be surfaced through OpenClaw's normal hook path when using the Codex runtime?
In other words: is this expected current behavior / unsupported surface area, or should Codex runtime tool calls be expected to show up as normal Opik tool spans?
Summary
When using OpenClaw with the Codex runtime, Opik traces are being created for LLM activity, but Codex/runtime developer tool calls do not appear to be recorded as Opik
toolspans.Example: during an active Telegram/OpenClaw session, multiple shell/tool operations were executed through the Codex runtime, but the Opik project showed only LLM spans after the gateway restart. Historical OpenClaw-visible tools are present, but recent Codex runtime tool calls are absent.
Observed behavior
toolspans for Codex runtime/developer tools such as shell/exec calls do not appear.Current hypothesis
The plugin appears to instrument OpenClaw typed hooks such as
before_tool_call/after_tool_call. Codex runtime/developer tools may be bypassing those OpenClaw hook events, so Opik only sees OpenClaw-visible tool calls and not Codex runtime tool calls.Question
Should Codex runtime/developer tool data be funneled into Opik separately, or is the intended direction that these tool calls will eventually be surfaced through OpenClaw's normal hook path when using the Codex runtime?
In other words: is this expected current behavior / unsupported surface area, or should Codex runtime tool calls be expected to show up as normal Opik tool spans?