https://peaty.app/blog Methods, guides, and findings for evaluating AI agent skills against real workflows. en https://peaty.app/blog/replay-the-work-not-the-prompt https://peaty.app/blog/replay-the-work-not-the-prompt Fri, 28 Aug 2026 12:00:00 GMT Methods A skill changes the path an agent takes. A fair evaluation preserves the work around it, then changes one thing. https://peaty.app/blog/skills-need-contracts-not-more-prose https://peaty.app/blog/skills-need-contracts-not-more-prose Wed, 26 Aug 2026 12:00:00 GMT Skills Longer instructions often hide the real interface. A useful skill states when it applies, what it owns, and how success can be checked. https://peaty.app/blog/a-pass-rate-is-not-a-verdict https://peaty.app/blog/a-pass-rate-is-not-a-verdict Fri, 21 Aug 2026 12:00:00 GMT Field notes Tests can pass while the workflow gets slower, noisier, or less trustworthy. Review the evidence that sits around the score. https://peaty.app/blog/evaluating-tool-use-skills https://peaty.app/blog/evaluating-tool-use-skills Tue, 18 Aug 2026 12:00:00 GMT Methods Test whether an agent chose the right tool, used it with the right scope, and turned its result into a defensible outcome. https://peaty.app/blog/a-taxonomy-of-skill-failures https://peaty.app/blog/a-taxonomy-of-skill-failures Wed, 12 Aug 2026 12:00:00 GMT Field notes A useful failure label identifies the decision that broke, not just the symptom visible at the end of the run. https://peaty.app/blog/turn-one-failure-into-a-test-case https://peaty.app/blog/turn-one-failure-into-a-test-case Wed, 05 Aug 2026 12:00:00 GMT Methods Preserve the smallest real context that reproduces the failure, then keep it in the suite after the immediate fix ships.