Run Tracekit on an open-source agent built on a framework and report what broke #795

Running Tracekit on two open-source agents found real bugs (case study). More agents will find more.

What to do

1. Pick an open-source agent built on a framework tracekit.instrument() wires: the OpenAI Agents SDK, LangGraph, the Claude Agent SDK, an MCP client, Browser Use, CrewAI or Pydantic AI. FRAMEWORKS (tracekit/autowire.py L344-357) lists them with the versions each is tested on. Browser Use 1.0 isn't gated yet (Browser Use 1.0: gate its tool calls (the 0.13 seam is gone, so the integration gates nothing on 1.0) #724), so use 0.13 for Browser Use.
2. pip install tracekit-ai, add import tracekit; tracekit.instrument() as its first line, and run it on a real task.
3. Run tracekit last and tracekit view --dev.
4. Comment here, or open a bug, with: the agent and its version, the framework and its version, what you ran, what worked, and anything that broke, was missing from the record, or was confusing.

For an experimental integration you wire by hand (Google ADK, Haystack, LlamaIndex and the others), use #743; for what Tracekit would have held in your coding agent's work, #739.

Done when: a comment here names the agent, the framework and their versions, and what you saw. A short report is useful. No code change needed.

Refreshes #682.