Deprecated projects

4,643 Git commits, public and private reposWeekly, 4-week average, square-root scale
March 2026
October 2026

Andon

Code measurement an agent could call as it worked.

Each verdict carried its evidence, and a prototype recompute tested whether the agent's own report held up.

Use instead, as of

Not covered

A check of a coding agent's recorded measurements against an independent recompute.

Their docs describe no such comparison. In Andon, that check stayed a prototype.

Why I stopped

Established tools already measure code for agents, and Qwen Code ships agent-run review. The independent check was the one part left that was Andon's own, and I chose not to finish it alone. NVIDIA's Open Agent Safety Platform takes a related approach to agent behavior at runtime.

What it took

commits
455
lines of Rust
84,485