fix: [#90] Add OpenTelemetry tracing support as optional backend

[nandanosql]

Fixes #90

Problem

The current trajectory recording writes JSON files to disk. While useful for debugging, this approach cannot integrate with standard observability toolchains (Jaeger, Zipkin, Grafana Tempo, Datadog, etc.) that rely on distributed tracing.

Solution

Added OpenTelemetry (OTel) tracing as an optional parallel backend that runs alongside the existing JSON recorder. This is non-breaking — the JSON recorder continues to work exactly as before.

New module: trae_agent/utils/otel_recorder.py

• setup_otel_tracing() — initializes OTLP gRPC exporter from OTEL_EXPORTER_OTLP_ENDPOINT env var
• OTelTrajectoryRecorder — emits spans for:
  • agent.run (root span) with task, provider, model attributes
  • llm.call with token counts, model, finish reason
  • agent.step with step number, state, tool names, errors
• Graceful no-op when opentelemetry packages are not installed

Integration in Agent

• __init__ detects OTEL_EXPORTER_OTLP_ENDPOINT and initializes OTel if available
• run() wraps execution with start_recording / finalize_recording

Dependencies

Added optional [otel] extra:

uv sync --extra otel

Usage

OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317 trae-cli run "Fix the bug"

Spans will appear in any OTLP-compatible backend. The JSON trajectory file is still written alongside.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}