Addy Osmani's agent-skills hits 100k stars: 24 modules enforcing AI coding discipline
大模型之路 · wechat · 2026-10-15
Addy Osmani (Anthropic engineer, ex-Google Chrome director) open-sourced agent-skills, a library that encodes senior-engineer discipline into 24 structured skill modules that AI coding agents structurally cannot skip.
The core idea: agents are capable but lack professional instincts — they skip anything not visible in the diff (specs, reviews, evidence). The 24 skills map to six phases (DEFINE/PLAN/BUILD/VERIFY/REVIEW/SHIP) with 7-8 slash commands, each containing steps, verification gates, and anti-rationalization tables that specify which excuses don't count as done.
Architecturally, skills are one layer among peers: AGENTS.md as rolling rules, hooks as deterministic enforcement, tools as actions, session logs as memory. Skills use progressive disclosure so only relevant ones load into the system prompt, keeping token overhead minimal — critical for long-running agents. Install via npx skills add and reuse the same SKILL.md across Claude Code, Cursor, Gemini CLI, Codex, Copilot, and Windsurf.
The author stresses skills aren't a silver bullet — hooks, subagents, and sandboxes are still needed where appropriate — and compares with Superpowers and Matt Pocock's skills, positioning agent-skills as the most complete engineering-discipline set. The key framing: force the discipline agents would otherwise talk themselves out of, rather than relying on bigger models.
More from coding & agent
- Claude Code Pushes Urgent Security Update Across All Installations — umeshonai · 2026-10-23
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23