🔥 Hot Repo: 103K Stars — Addy Osmani Wired Senior Engineering Into Claude Code

agent-skills is a 25-skill, 9-command production engineering pack that turns Claude Code, Cursor, Codex, and 70+ AI coding agents into disciplined senior engineers — now with a /build auto mode that ships an entire feature autonomously after one plan approval.

By OMC Editorial on 2026-10-10

One-liner — agent-skills is a 25-skill, 9-command production engineering pack for AI coding agents that encodes the workflows, quality gates, and best practices that senior engineers use, so agents consistently spec, plan, build, verify, review, and ship — without the shortcuts. - Repo: addyosmani/agent-skillshttps://github.com/addyosmani/agent-skills - Stars: ⭐ 103,000 +21K since August 2026 - Language: Markdown / JavaScript - License: MIT --- What It Does agent-skills packages 25 production engineering workflows into Markdown skill files that Claude Code, Cursor, Codex, Copilot, and 70+ other AI agents load as context. Each skill encodes a specific lifecycle stage — from spec-driven-development before a line is written to git-workflow-and-versioning and shipping-and-launch at the end. Nine slash commands /spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify, /ship map to each phase, and skills activate automatically based on what the agent is doing. Why It's Blowing Up The core insight driving the repo's growth is one the author — Addy Osmani, formerly Google's Chrome DevRel lead — made plain in several posts: prompts are not the point. What separates a senior engineer's output from a junior's isn't what they ask; it's the disciplined process they run automatically — writing a spec before touching code, committing atomically, proving correctness with tests before calling anything done. Skills encode that discipline so agents run it consistently, not occasionally. The project has grown from 45K to 103K stars since June 2026, fueled by a simple dynamic: every "my AI agent wrote broken code" complaint is a testimonial. Version 0.6.2 introduced the headline feature — /build auto. Approve the plan once, and the agent generates, tests, and commits every task without human steering between steps. It pauses on failures or risky actions; everything else runs autonomously. For teams running long Claude Code sessions, this is the shift from "

More agent news