Eval OTel Skill

Use This Skill For

• Proving we can record and retrieve prompt input/output during eval runs.
• Debugging missing trace/session/run correlation in rows.jsonl.
• Running a repeatable telemetry validation loop across codex, claude, and louie.

Success Criteria (Do Not Skip)

• For each runtime under test:
  • rows.jsonl has non-empty case_prompt.
  • rows.jsonl has non-empty response_text for successful runs.
  • rows.jsonl includes trace_id and raw_ref.
  • raw_ref file exists and is readable.
• OTel/log retrieval is demonstrated for the same run IDs/trace IDs when available.
• Any runtime not proven is explicitly marked as not proven (with error evidence).