Claude Code eval skills for AI testing released by developers

youtube.com8 min read

The author and their colleague Shreya Shankar developed eight evaluation skills for coding agents to help developers build better AI evals by examining product failure modes. The two most critical skills are error discovery, which guides agents in parsing data and building custom annotation interfaces, and eval audit, which checks eval setups against common mistakes such as using non-binary scales or unvalidated LLM judges. Open article →