GitHub's practical guide: write an agent skill once, run it on any agent
0xkarasy · x · 2026-11-13
GitHub Blog turns a GitHub Universe 2026 session into a practical guide on agent skills: a skill is a versioned directory around a SKILL.md describing when/how to do a repeatable task; tools, credentials and permissions stay with the host. Key practices: one clear job with defined input/output and frontmatter; a separate verifier (Proof Agent) so self-review isn't biased; business context in one file, one way to use the platform, instructions to guide plus checks to verify. The goal is turning tribal knowledge into reviewable, reusable team assets.
More from coding & agent
- llm-drift-bench: open-source tool diffs LLM outputs to catch regressions when switching models — Worldly-Knowledge-90 · 2026-11-13
- Anthropic outage forces mid-deployment model switch, proving vendor lock-in risk — heyneighbor · 2026-11-13
- ONES publishes a 7-field handoff checklist for AI agents, sparking workflow debate — ombre-ai · 2026-11-13
- Dev builds AI newsletter pipeline with Jev, Haiku and Sonnet that clusters news into narratives — runwithdata · 2026-11-13
- Running Multiple AI Coding Agents: Coordination Is Harder Than Generating Code — Time_Positive_9123 · 2026-11-13
- Cloudflare launches Spend Management with hard spending limits for Workers — irvinebroque · 2026-11-13