OpenClaw deleted ~400k lines of AI-written tests with little coverage loss using a test-audit skill
steipete · x · 2026-09-24
steipete shares that OpenClaw deleted around 400k lines of its own AI-written tests with little change in code coverage — modern models love writing tests for every tiny change, useful or not.
The companion SKILL.md defines three modes: an authoring gate that blocks new tests unless they answer four questions (what behavior it protects, what regression fails it, etc.), an audit mode sweeping low-value, implementation-coupled, or duplicative tests, and a campaign mode pruning a whole subsystem's test surface — optimizing for confidence, not deletion count.
Related event: OpenClaw uses agent to delete 400K lines of tests(2 posts)→
More from coding & agent
- Meta's Proactive Memory Agent fixes context rot, lifting Claude Sonnet 4.5 from 37.6% to 45.9% — DeepLearningAI · 2026-09-25
- A practical guide to Claude Code 2.0: sub-agents, hooks, and context engineering — dejavucoder · 2026-09-25
- mitsuhiko: The Experience of Models Editing Files via Code Is Still Unacceptable — mitsuhiko · 2026-09-25
- Nimble launches official OpenAI plugin for ChatGPT and Codex — OpenAIDevs · 2026-09-25
- Asupersync: A Cancel-Correct Async Runtime for Rust, Built Heavily with Frontier Models — doodlestein · 2026-09-25
- Building Macross Plus Mecha with Codex: A Dev's Side Project — algo_diver · 2026-09-25