Ant's Code2Skill mines 1M+ skill cards from 19,769 GitHub repos, boosting agent benchmarks 11.7%
aigclink · x · 2026-09-22
Ant released Code2Skill, a third path for equipping agents with reusable engineering skills: mining them directly from code instead of relying on model experience or hand-written docs.
- Scale and results: Extracted 1M+ skill records from 19,769 GitHub repos. With skills attached, models improved on 57 of 72 comparison groups (avg +11.7%) and ranked first on all 7 benchmarks against an "experience-based" skill library under identical conditions.
- Blind-reconstruction QC: One AI writes a skill card from code; another AI rewrites the code using only the card. Large divergence triggers review and possible discard. Human sampling found 92% of cards accurate, 80% worth keeping.
- Notable finding: In a small control experiment, skills mined from passing AI-generated code matched those from human code, suggesting AI code is viable raw material.
- Outlook: The authors foresee a "skill library as a service" infrastructure layer where agents query a subscribed skill API on demand.
More from coding & agent
- Photoroom reclaims ~1,000 idle GPU hours a week with SkyPilot across Slurm and Kubernetes — skypilot_org · 2026-11-12
- Google launches Gemini Decisions API: scores 1,000 questions per pass, output tokens free — dean_rie · 2026-11-12
- Zed v1.28 adds background auto-fetch for current and default branch remotes — dean_rie · 2026-11-12
- Indie Britpop AI music video built with Nano Banana Pro, Kling, and free lip-sync via WAN2GP + LTX 2.5 — cocktailpeanut · 2026-11-12
- Cloudflare Talk: Swapping Some Agent LLM Calls for Fast Models Like Jev Gives Big Speedups — michellechen · 2026-11-12
- Engineer Curates 15 AI Agent Design Patterns Across Single-Agent, Multi-Agent and Feedback Tiers — MaryamMiradi · 2026-11-12