Claude accidentally does recursive skill improvement, spawning sub-agents to A/B its SKILL.md
holdenmatt · x · 2026-11-07
Developer Matt Holden shares an accidental milestone: while writing a SKILL.md for PR style guidelines, Claude disliked its own draft and spawned sub-claudes to A/B test it against itself.
- Each fresh agent was given only the skill plus the diff of an already-merged PR (one UI change #2178, one backend change #2240, one small fix #2124)
- It then compared the skill-generated write-ups against what humans actually wrote, to see if the skill alone reproduces the style and which rules get skipped
Holden jokes that "(very slow) Recursive Skill Improvement (RSI) has been achieved internally" — the model writing instructions for improving itself. The screenshot also doubles as a reusable method for validating agent skills.
Related event: Claude Spontaneously Spawns Sub-agents to A/B Test Its Own SKILL.md(2 posts)→
More from coding & agent
- Conductor shuts down as cloud agent startups fold, devs hunt alternatives — sergeykarayev · 2026-11-07
- Incorporate a DE C-Corp via Stripe CLI + MCP, plus $100k in Cloudflare credits — seanmcdonaldxyz · 2026-11-07
- Dev plans to SFT a 4B model into a specialized coding agent, starting with 100-200 mini SaaS apps — ConfidentDinner6648 · 2026-11-07
- Mohit Bansal's Northwestern lecture: fixing agent trust, world discovery, and long-horizon memory — ManlingLi_ · 2026-11-07
- Hackathon winners build GLaDOS AI escape room that phishes you for permissions — charles_irl · 2026-11-07
- GitHub's agentic engineering guide: commits up 400% YoY, but PR review waits up 441% — GitHub Blog AI/ML · 2026-11-07