create-tracing-module

name	create-tracing-module
description	Creates the tracing.py module with enable_tracing(), health check, and framework-specific autolog configuration.
argument-hint	<agent_path> [framework]
disable-model-invocation	true

Create the Tracing Module (tracing.py)

Usage: /create-tracing-module <agent_path> Example: /create-tracing-module agents/autogen/chat_agent

You are creating the tracing.py file for a new agent template, following this repo's established patterns.

Input

The agent path is: $ARGUMENTS

You also need the framework name (e.g., autogen, langgraph, crewai) and the autolog support report (which classifies the framework as Level A, B, or C). The framework may be included as a second word in $ARGUMENTS. If not provided, determine it from the agent's pyproject.toml dependencies.

If the autolog report is not available in the current conversation context, invoke the check-autolog-support skill (i.e., /agentic-starter-kits-skills:check-autolog-support) with the framework name to produce the report before continuing.

Context

Read tracing.md at the repo root for the full tracing architecture, design principles, and how each existing agent integrates with MLflow.

Reference Files

Read the reference tracing.py that matches the coverage level. These are the source of truth — follow their patterns exactly:

• Level A — autolog variant (full autolog): agents/langgraph/react_agent/src/react_agent/tracing.py
• Level A — OTel variant (framework emits OTel spans): agents/google/adk/src/adk_agent/tracing.py
• Level B (partial autolog): agents/crewai/websearch_agent/src/crewai_web_search/tracing.py
• Level C (no framework autolog): agents/vanilla_python/openai_responses_agent/src/openai_responses_agent/tracing.py

What to Create

Create the file at: <agent_path>/src/<package_name>/tracing.py

The <package_name> is the Python package name — find it by looking at the existing src/ directory or pyproject.toml [tool.setuptools.packages.find].

How to Build It

Shared code (all levels)

Copy these exactly from the reference file — they are identical across all agents:

• Logger setup (module-level logging.getLogger("tracing") with handler)
• check_mlflow_health() function (retry loop with time budget)
• enable_tracing() skeleton (load_dotenv, MLFLOW_TRACKING_URI check, health check with graceful degradation, set_tracking_uri, set_experiment, enable_async_logging)

Framework-specific code

The only parts that differ between levels are inside enable_tracing() and whether wrap_func_with_mlflow_trace() exists:

Level A (autolog variant) — Read the LangGraph reference. Adapt by:

• Replacing the import mlflow.langchain / mlflow.langchain.autolog() with the correct module for the new framework (from the autolog report)
• No wrap_func_with_mlflow_trace() needed

Level A (OTel variant) — Read the Google ADK reference. Use this when the framework has no mlflow.<framework>.autolog() but natively emits OpenTelemetry spans. Adapt by:

• Setting up TracerProvider with OTLPSpanExporter pointing at {tracking_uri}/v1/traces
• Passing the experiment ID via the x-mlflow-experiment-id header (get it from mlflow.set_experiment())
• The MLflow server requires a SQL backend (--backend-store-uri sqlite:///mlflow.db)
• Extra package: opentelemetry-exporter-otlp-proto-http
• No wrap_func_with_mlflow_trace() needed

Level B — Read the CrewAI reference. Adapt by:

• Replacing the framework autolog module (e.g., mlflow.crewai → mlflow.<new_framework>)
• Adjusting the provider autolog routing if the new framework's LLM provider path differs. If the framework uses a single LLM path, simplify by hardcoding the provider autolog instead of the provider_autolog_map + LLM_PROVIDER routing
• Copying wrap_func_with_mlflow_trace() as-is (with name parameter)

Level C — Read the Vanilla Python reference. Adapt by:

• Replacing the provider autolog module if the framework uses a different LLM SDK (from the autolog report)
• Copying wrap_func_with_mlflow_trace() as-is

Final Checklist

After creating the file, verify:

• Logger setup is at module level (not inside a function)
• check_mlflow_health() is identical to the reference
• enable_tracing() has graceful degradation (returns on unreachable server, doesn't crash)
• MLflow imports are lazy (inside enable_tracing() / wrap_func_with_mlflow_trace(), not at module top)
• wrap_func_with_mlflow_trace() returns the original function unchanged when MLFLOW_TRACKING_URI is not set
• The correct autolog module is used for the framework
• enable_async_logging() is called before autolog

Self-Update

Before finishing, check whether this skill file needs updating. If any of the following are true, propose the specific changes to the user and only update this file if they approve:

• You deviated from these instructions because they were inaccurate, outdated, or insufficient — and your deviation worked.
• You encountered a new pattern not covered here (e.g., a framework that uses OpenTelemetry instead of autolog, or a new provider path).
• File paths, function names, or API patterns referenced here have changed.
• A step didn't work as described and needed a different approach.

If nothing needed changing, move on.