gnibu · Agent Skills

gnibu

Community

bento-hillclimb

Tune anything measurable — retrieval quality, latency, eval scores, a prompt — by disciplined one-variable-at-a-time experiments with a kept/reverted log. Use when: (1) improving a metric, (2) tuning retrieval/RAG or performance, (3) iterating on a prompt against a dataset. An unexpected failure/regression belongs to bento-investigate first; running an existing suite once is not tuning. Triggers: "tune", "improve the score", "optimize a metric", "iterate on a measured baseline".

1h installs

1

GitHub stars

0

Updated

Oct 20