wire-into-lifespan

name	wire-into-lifespan
description	Wires enable_tracing() into the FastAPI lifespan and adds manual trace wrapping imports to main.py.
argument-hint	<agent_path>
disable-model-invocation	true

Wire Tracing into the FastAPI Lifespan

Usage: /wire-into-lifespan <agent_path> Example: /wire-into-lifespan agents/autogen/chat_agent

You are connecting the tracing.py module to the agent's FastAPI app so that tracing is initialized at startup.

Input

The agent path is: $ARGUMENTS

You also need the package name and coverage level (A, B, or C). If not provided, determine them by reading the agent's pyproject.toml and src/ directory.

Steps

1. Read the agent's main.py

Read <agent_path>/main.py to understand:

• The existing lifespan() function
• How the agent is initialized (what global variables, closures, etc.)
• Whether there's a streaming path that creates the agent differently from non-streaming

2. Add the tracing import

Add the import at the top of main.py, near the other package imports:

Level A (full autolog):

from <package_name>.tracing import enable_tracing

Level B or C (needs manual wrapping):

from <package_name>.tracing import enable_tracing, wrap_func_with_mlflow_trace

3. Add enable_tracing() to the lifespan

Add enable_tracing() as the first call inside the lifespan() async context manager, before any agent initialization:

@asynccontextmanager
async def lifespan(app: FastAPI):
    enable_tracing()       # <-- ADD THIS LINE

    # ... existing agent initialization code ...
    yield
    # ... existing cleanup code ...

This placement ensures tracing is configured before any autolog-patched code runs.

4. Wire wrap_func_with_mlflow_trace (Level B and C only)

Skip this step for Level A — autolog handles everything.

For Level B/C, you need to wrap agent and tool functions with trace spans. Where you add the wrapping depends on how the agent initializes its components. Read the agent's code to find:

Tool wrapping

Find where tools are registered or instantiated. Wrap each tool's callable:

Pattern 1 — Tool objects with _run method (CrewAI-style, Level B):

# In crew.py or wherever tools are created
tool._run = wrap_func_with_mlflow_trace(tool._run, span_type="tool", name=tool.name)

Pattern 2 — Tool functions registered by name (Vanilla Python-style, Level C):

# In agent.py or main.py, wherever tools are registered
func = wrap_func_with_mlflow_trace(func, span_type="tool")
agent.register_tool(name, func)

Agent orchestration wrapping

Find the main agent entry point (the function that runs the full agent loop) and wrap it:

agent.query = wrap_func_with_mlflow_trace(agent.query, span_type="agent")
# or
agent.run = wrap_func_with_mlflow_trace(agent.run, span_type="agent")

Streaming path

Check if _handle_stream in main.py creates the agent differently from _handle_chat. If it does (e.g., creates a new agent instance directly instead of using the closure), the wrapping must be duplicated in the streaming path.

Reference: In the Vanilla Python agent, _handle_stream creates AIAgent directly and wraps tools + agent.query inside the run_agent() function. See agents/vanilla_python/openai_responses_agent/main.py, search for def run_agent().

Reference Files

Read these to see the wiring patterns in practice:

• agents/vanilla_python/openai_responses_agent/main.py — Level C wiring (search for import enable_tracing, enable_tracing() in lifespan(), and run_agent() in _handle_stream)
• agents/langgraph/react_agent/main.py — Level A wiring (search for import enable_tracing and enable_tracing() in lifespan())
• agents/crewai/websearch_agent/main.py — Level B wiring (search for import enable_tracing and enable_tracing() in lifespan())
• agents/crewai/websearch_agent/src/crewai_web_search/crew.py — Level B tool wrapping (search for wrap_func_with_mlflow_trace in ai_assistant())

Final Checklist

• enable_tracing() is called before agent initialization in lifespan()
• Import statement is at the top of main.py with other package imports
• For Level B/C: wrap_func_with_mlflow_trace is imported
• For Level B/C: Both streaming and non-streaming paths have wrapping applied
• No MLflow imports are added directly to main.py — all MLflow interaction goes through tracing.py

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.