This is the question ClawMetry exists to answer: an agent has been running for forty minutes and made a lot of tool calls — is it working, or is it stuck? Pick a session. One of these three is looping; the other two are fine, and one of those is longer than the broken one. No account, nothing to install.
Run this on your own agents. ClawMetry reads the session files your runtimes already write — no SDK, no code change. The same three sessions ship in the package, so you can land on this exact screen before you have data of your own:
pip install clawmetry && clawmetry --sample
Drop --sample to watch your real agents.
OpenClaw, NVIDIA NemoClaw and Goose are free in the open-source package;
the other 30 runtimes need a plan.
Runtimes expose different data — the compatibility matrix
records exactly what each one can and cannot show.