
Agentic eval
SkillGuides agents through self-critique, reflection loops and LLM-as-judge evaluation
- Open source
- 41.3K GitHub stars
- AI models and LLM tooling


Guides agents through self-critique, reflection loops and LLM-as-judge evaluation

Guides agents to write only the source comments that carry non-obvious information

Guides an agent through drafting, testing and tuning new agent skills with measured evals

Runs an automated review-and-fix loop that improves a Claude Code skill until it passes

Drafts copy-paste-ready /goal completion conditions for long-running agent tasks

Guides reviewers through pull requests to awesome-copilot for fit, value and provenance

Coaches junior developers through Socratic questions instead of handing them answers

Guides an agent from a plain-text problem to an LP, MILP or QP formulation

Guides an agent through gradual, behavior-preserving refactors of existing code

Guides an agent through building LLM eval pipelines with datasets, evaluators and pixie

Sets up a lightweight producer, dev and QA agent team for planning and building projects

Runs an autonomous review-and-fix loop on a branch until its PR review comes back clean
The closest alternatives to Writing lean proofs are Agentic eval, Lean comments, Skill creator, Skill improver and Goal prompt.
Yes. Agentic eval, Lean comments, Skill creator have a free or open-source option.