// AI news · 2026-11-13 · GitHub Blog AI & ML
Build once, run on any agent: agent skills guide
GitHub introduces agent skills as reusable, versioned workflows that define repeatable tasks.
Published 2026-11-13 06:54 (Beijing time)Source GitHub Blog AI & MLAuto-summarised, fact-checked against source
01Summary
Skills capture instructions in SKILL.md files but do not grant tools or permissions. They enable consistent execution across agents while maintaining environment-specific access. The approach emphasizes clear task boundaries, separation of project configuration from team standards, and human-in-the-loop checkpoints for critical decisions.
Summarised from the original article by an open-source model, then checked automatically: every number and English name must appear in the source. The original article is authoritative.
02Key points
- Agent skills define reusable, versioned workflows via SKILL.md files.
- Skills must clearly specify tasks, inputs, and outputs; self-validation is not verification.
- Multiple skills coordinate in pipelines like Data Forge, each with defined responsibilities.
- Project config is externalized to a manifest; tools and permissions remain environment-bound.
03Source
Build once, run on any agent: A practical guide to agent skills
GitHub Blog AI & ML · published 2026-11-13 06:54 Beijing time
05Related news
- Same topic · Google launches Gemini 4 Hydrogen for complex question understanding2026-11-13 · The Verge
- Same topic · Anthropic shares safety eval design recommendations2026-11-13 · Anthropic Engineering
- Same day · Muse Can Now Make Calls to U.S. Businesses with Paid Subscription2026-11-13 · TechCrunch
From AIMap AI news: collected from 1,100+ official and media sources. Summaries only restate the original article; follow the link for the full story. All stories of 2026-11-13 · Archive by date · All AI news · Today