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
- Anthropic's masterclass on building AI agents, plus agent observability — _jaydeepkarale · 2026-09-22
- Curated AI agent learning path kicks off with Anthropic's official agent-building masterclass — _jaydeepkarale · 2026-09-22
- Full free course: build and deploy your own AI agent with OpenAI, Langchain and Render — _jaydeepkarale · 2026-09-22
- An AI pentesting agent with credentials and broad access is itself an attack surface, says Terra OffSec CEO — TechNadu · 2026-09-22
- Karpathy open-sources AutoResearch, agents that run the full research loop — mdancho84 · 2026-09-22
- Codex plus Computer Use is 'undefeated' for real agent workflows, argues longtime user — soumitrashukla9 · 2026-09-22