Improve Agent Benchmarks

Cheap iterative improvement of a shared agent. Runs the next stage for the chosen agent (contract, cases, iterate). Contract stage drafts a cited, history-validated contract.json with an Opus subagent, shows each check with its source, historical pass rate and real failing examples, and freezes it on approval. Use when the user invokes /improve-agent-benchmarks [agent-name].

Eric-Lingren Updated

File contents

Eric-Lingren/claude-code-toolkit commit 368cd840a4

Frequently asked questions

npx get-skill add eric-lingren/improve-agent-benchmarks