Addy Osmani releases 19 Agent Skills

Addy Osmani, a director at Google Cloud AI, released “Agent Skills” as an open-source project on May 3, 2026, framing it as a way to make AI coding agents follow the same checkpoints senior engineers use before software ships. The GitHub repository describes the package as “production-grade engineering skills for AI coding agents” and says the skills encode workflows, quality gates and best practices across the software development lifecycle. The project has since become a reference point in developer posts about “agentic” software workflows because it turns practices such as spec writing, test-first development and review gates into reusable instructions that can be invoked by coding agents. In a companion blog post dated May 3, Osmani wrote that AI coding agents “skip those parts by default” and said the project was his attempt to make those steps “not optional.” (github.com) ### What exactly did Osmani publish? The GitHub repository lists 19 skills and seven slash commands. The skills are organized as reusable workflow modules, while the commands map to stages of development including defining, planning, building, testing, reviewing, simplifying and shipping. The seven commands shown in the repository are `/spec`, `/plan`, `/build`, `/test`, `/review`, `/code-simplify` and `/ship`. (addyosmani.com) The README says each command activates the right skills automatically, with examples such as API design triggering `api-and-interface-design` and UI work triggering `frontend-ui-engineering`. ### Which engineering habits are baked into the system? (github.com) Osmani’s May 3 blog post says the package is meant to restore work that “doesn’t show up in the diff,” including specs, tests, reviews, scope discipline and verification. He wrote that a useful skill is a workflow with steps, checkpoints and an exit criterion, not a long reference document. The repository’s command table ties those habits to explicit principles: “Spec before code” for `/spec`, “Tests are proof” for `/test`, “Clarity over cleverness” for `/code-simplify`, and “Faster is safer” for `/ship`. (github.com) The skills directory includes modules named `spec-driven-development`, `test-driven-development`, `security-and-hardening`, `code-review-and-quality`, `incremental-implementation` and `shipping-and-launch`, among others. (addyosmani.com) ### How does the project handle review and deployment? GitHub release notes for version 0.6.0 say the project now separates personas, skills and slash commands into “three explicit layers.” The same release says `/ship` runs three specialist personas in parallel — code reviewer, security auditor and test engineer — and then synthesizes a go/no-go decision. The 0.6.0 notes also say new integrations were added for Gemini CLI, Kiro and OpenCode. (github.com) For Gemini CLI, the release documents seven commands mirroring the Claude Code set, with `/planning` used instead of `/plan` to avoid a command collision. ### Why are developers treating this as more than a prompt pack? Osmani’s blog post says the distinction is between prose and procedure. (github.com) He wrote that if an agent gets a long essay on testing, it may generate plausible text and skip the testing itself, but if it gets a workflow — such as writing a failing test first, running it, then writing minimal code to pass — the human operator has something concrete to verify. The repository reflects that design choice in the names of the modules. Alongside coding tasks, the skill list includes `planning-and-task-breakdown`, `documentation-and-adrs`, `debugging-and-error-recovery`, `deprecation-and-migration`, `source-driven-development` and `interview-me`, indicating the project is aimed at the full path from idea refinement to launch rather than code generation alone. (addyosmani.com) ### Where can readers see what changes next? The GitHub repository, the May 3 blog post and the releases page are the primary public records for the project. As of the latest visible release notes, version 0.6.0 added orchestration changes, Gemini CLI support and a revised `/ship` workflow, and the repository’s skills directory shows ongoing updates including a recently added `interview-me` skill. (github.com 1) (github.com 2)