Mechanical governance score from agent-contracts scan against a clean clone. No vibes, no testimonials — reproducible from the exact commit below.
pip install "agent-contracts @ git+https://github.com/impartshadow/agent-contracts.git" git clone --depth 1 https://github.com/Significant-Gravitas/AutoGPT.git git -C AutoGPT checkout e2711b1748 agent-contracts scan --root AutoGPT --json
Scored commit: e2711b1748bdc3fe702ab4e44c6a11df98458c53. The score is the
output of a deterministic scanner, not a judgment call — a clean clone reproduces it.
This index measures observable governance surface — the guardrails a reliable agent needs. It does not certify that agents built on AutoGPT behave correctly at runtime. The most expensive agent failure — confidently claiming a task is "done" with nothing to back it — has no surface to grep, so no static score (including this one) can see it. Here's the proof, on the #1-ranked framework.
Every framework is re-scanned on a schedule. The weekly delta — what moved, on which commit, and what it means for teams building on it — goes out at echofromshadow.substack.com.