Back to work

01 / TOOLING · EVALUATION

skillfit

Put agent configurations to the test.

Compare Skills, rules, and MCP setups on the same tasks. Keep quality, trigger behavior, and token cost in view.

TypeScript / CLI / Paired evaluation

The problem

Adding a Skill or changing an agent rule can sound useful without improving the work. Comparing configurations needs the same tasks, explicit checks, and a record of what actually happened.

The approach

skillfit compares a baseline and a treatment through paired runs. Reports keep verification results, configuration behavior, token costs, and measurement limits together. It is available as an npm CLI.

Evidence & scope

The repository includes replayable case studies, verifier checks, and evaluation reports. Small pilots and negative results remain visible; a working measurement pipeline does not establish that a configuration is beneficial.