darnit is a framework for harnesses that evaluate software projects against practices and standards (OpenSSF Baseline, reproducibility, gittuf, license compliance), collect the project data those evaluations need, and remediate gaps. Its core principle: do as much as possible deterministically, and hand work to an LLM only where that is more accurate or cheaper than the alternative -- never letting a model alone declare that something is compliant.
This page summarizes the current architecture, what an evaluation of it found, and the direction proposed in two RFCs:
- RFC-0001 September 2026 revision: darnitdevorg/darnit#499
- RFC-0002, darnit as an enablement layer: darnitdevorg/darnit#500
- Findings tracker: darnitdevorg/darnit#491