Addy Osmani packages Claude Code agent skills into slash commands that mirror the dev cycle
A post says the kit wraps senior-engineer patterns into slash commands like /spec, /plan, /build, /test, /review, /ship; it claims seven commands but lists six.
By Ryan Merket ·
Why it matters
Agent workflows are getting real. Turning spec-plan-build-test-review-ship into first-class commands pushes code assistants beyond chat into repeatable engineering process, which is what teams need to trust agents in production.
Addy Osmani published a production-oriented set of "Agent Skills for Claude Code" that map the software development cycle to slash commands, according to an X post from Aligned News. The post describes Osmani as a Google Chrome VP and says the skills are designed so AI coding agents adopt senior-engineer patterns by default.
The kit is framed as covering the full path from specification to shipping. While the post claims seven commands, it lists six: /spec, /plan, /build, /test, /review, and /ship. The intent, per the post, is to turn process into a first-class primitive for Claude Code so agents move through work the way seasoned engineers do.
What shipped, per the post
- /spec: drive a concise product or technical spec before any implementation.
- /plan: translate the spec into tasks, propose an architecture, and sequence the work.
- /build: implement code to the plan and spec.
- /test: generate and run tests, probe edge cases.
- /review: self-review or propose a diff with comments before submission.
- /ship: prepare release notes and deployment steps.
The post characterizes these as production-grade behaviors rather than ad hoc prompts. If the execution matches the framing, the skills would standardize how a coding agent scopes, executes, and validates work, with explicit gates between phases.
Why a Chrome leader would push workflow into the agent
Even without a public repo or docs linked in the post, the thesis is clear: process is a feature. Encoding spec-plan-build-test-review-ship as commands pushes the agent to think in deliverables, not just generate code. That aligns with the broader industry turn toward agents and work orchestration. Google has leaned into this direction with agentic products like Gemini Spark, highlighted on Google during I/O 2026, and Osmani is positioned in the Aligned News post as a leader on the Chrome side of that push.
For teams experimenting with AI pair programmers, a thin layer of workflow primitives can be the difference between a helpful autocomplete and a reliable teammate. Packaging senior patterns into defaults reduces variance, helps newcomers adopt proven practices, and gives reviewers consistent artifacts to inspect at each step.
Open questions
The X post did not include links to code or documentation, so key details remain unverified here:
- The post says there are seven commands but enumerates only six. Which phase is missing?
- Where are the skills hosted, and under what license?
- Are they a personal release by Osmani or affiliated with Google or another organization?
- How do they integrate with Claude Code in practice, and are there examples or benchmarks from real teams?
What is clear from the post is the bet: ship a process layer that makes an AI coding agent behave like a senior engineer out of the box. If the artifacts back it up, it is the kind of small, opinionated abstraction that can compound across teams quickly.