AI Agent Skills Framework — Senior Engineer Workflows in Markdown

Review

What it is:

A 24-skill Markdown ruleset that encodes senior engineering workflows (spec → plan → build → test → review → ship) for AI coding agents. Designed to run in Claude Code, Cursor, Antigravity, Gemini CLI, Windsurf, and other agent platforms. Entry points are 7 slash commands (/spec, /plan, /build, /test, /review, /ship, /code-simplify); /build auto runs plan + implementation in one approved pass with individual commits and pauses on failures.

Who it's for:

Teams using AI agents (Claude Code, Cursor, etc.) who want consistent, disciplined development workflows — spec first, test-driven, incremental slices, quality gates before merge.

What's genuinely good:

• Lifecycle clarity: The six-stage diagram and command mapping make it obvious when to invoke which skill. Not a grab-bag.
• Practical structure: Skills include acceptance criteria, verification gates, and explicit anti-rationalization (what NOT to do). This is teaching, not just cheerleading.
• Multi-platform support: Works across Claude Code, Cursor, Antigravity, Gemini, Windsurf, GitHub Copilot, and plain Markdown agents. Low friction to adopt.
• 24 documented skills: Covers spec → API design → frontend engineering → test strategy → code review → deployment. Breadth is there.
• Autonomous mode with safety: /build auto removes manual stepping between tasks but keeps verification — tests and commits stay per-task, failures still pause the flow.

Honest limitation:

The README claims these are "production-grade" and encode what "senior engineers use," but provides no evidence: no case studies, no metrics, no before-after examples of a team using it, no real output samples. It's entirely aspirational instruction — which is fine, but "production-grade" is marketing language without proof. You must trust Addy Osmani's judgment and test it yourself. The README also doesn't explain which skills are mandatory vs. optional, or how they interact when conflicts arise (e.g., if /review and /doubt-driven-development have different conclusions).

One-line verdict:

A well-organized, multi-platform ruleset for disciplined AI-assisted development with clear entry points and anti-patterns, but unproven in production and relies entirely on vendor credibility.

REPO: addyosmani/agent-skills

---

🔗 Repo: https://github.com/addyosmani/agent-skills

An honest review by the Flowork team — we read the README so you don't have to. We build open-source tooling too; this isn't a sponsored post.