Image: GitHub
Why it mattersA team using an AI coding agent gets one place to pull named workflows for specs, tests, review and release. The agent then follows a shared standard rather than the engineer's memory of it.
A pack of prewritten workflows for AI coding agents has become one of the most-starred new repositories on GitHub. The repo is called "agent-skills" and the owner is Addy Osmani, who works on Chrome at Google. The repo reports 116,163 stars on the morning of 12 November, with 12,172 forks. It sits at rank 22 on Trendshift's daily list that same day.
Trendshift tracks which repositories are getting new attention on GitHub. A rank in the top 25 means the project is in the stream of what developers are discovering this week. The repository itself was created on 15 February 2026. The last commit before the ranking was the 0.6.15 release on 31 October, followed by a merge on 1 November.
What the pack is
The repo bundles 25 named skills that an AI coding agent can load. Each skill is a short workflow for one job the agent has to do. Every skill carries its own quality gates. The README groups the 25 skills across the lifecycle of a change: Define, Plan, Build, Verify, Review and Release. The quoted summary is Osmani's: "Skills encode the workflows, quality gates, and best practices that senior engineers use when building software."
The Build group carries the ones a working engineer meets every day: incremental-implementation, test-driven-development, context-engineering, source-driven-development, doubt-driven-development, frontend-ui-engineering and api-and-interface-design. Review carries code-review-and-quality, code-simplification, security-and-hardening and performance-optimization. The release group carries git-workflow-and-versioning, ci-cd-and-automation, observability-and-instrumentation and the release workflow itself.
Nine slash commands sit on top: /spec, /plan, /build, /test, /constraints, /review, /webperf, /code-simplify and the release command. Each maps to one phase of the pack. The licence is MIT. Collaborators named in the README are Federico Bartoli and Joan León.
What works, what to check
The pack works with the five best-known coding agents, and the README names installers for each. A universal one-line add is given for teams who do not want to pick. The repository has 807 commits and 136 open issues as of 12 November. Those two numbers together say the pack is maintained and that it is not fire-and-forget.
A Reveneau team reading this would start with one skill before pulling the whole pack. The Review group is the one most likely to show value on its first run. Code-review-and-quality and security-and-hardening both carry pass-or-fail steps the agent has to meet. Those are exactly the checks a person does slowly. Running them through the agent against a shared standard is quicker than the memory of the engineer in the seat.
Two things to watch. A skill is a prompt plus a workflow and both age. The 0.6.15 bump is this month, so the pack is current. Any team pinning to a version takes on the job of re-reading it when the underlying model changes. And a 25-skill pack is still a defined menu. A Reveneau client with its own eval suite should still run those evals against the agent's output. The skills do not know what passes in that codebase.
Source
Primary source: github.com/addyosmani/agent-skills. Repository metadata: api.github.com/repos/addyosmani/agent-skills.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.