Addy Osmani Agent Skills

addyosmani/agent-skills is Addy Osmani's engineering workflow skill pack for AI coding agents. It packages lifecycle slash commands, SKILL.md instruction files, review personas, and reference checklists for agents that need to move from product definition through implementation, verification, review, and launch without losing engineering discipline.

Use this listing for the skill pack itself. Use separate tool, MCP, or platform entries when evaluating a runtime, agent framework, cloud service, or execution environment rather than installable agent instructions.

Knowledge Freshness

The pack is current and active, but it intentionally focuses on engineering process. Exact framework APIs, package commands, cloud behavior, browser interfaces, CI syntax, deployment controls, and security guidance still need fresh source checks during implementation.

For high-stakes work, pair these skills with current documentation, local test results, repository-specific agent instructions, and maintainer review. Treat autonomous build or shipping workflows as approval-gated execution paths, not as a substitute for release control.

Retrieval Sources

This listing is grounded in:

• The upstream addyosmani/agent-skills README.
• The repository plugin.json metadata.
• Representative skill files for using-agent-skills, spec-driven-development, incremental-implementation, test-driven-development, security-and-hardening, and shipping-and-launch.
• The MIT license file and current GitHub repository metadata.

Core Workflow

Claude Code is the recommended install path upstream:

/plugin marketplace add addyosmani/agent-skills
/plugin install agent-skills@addy-agent-skills

If SSH marketplace cloning is not configured, the README documents an HTTPS fallback:

/plugin marketplace add https://github.com/addyosmani/agent-skills.git
/plugin install agent-skills@addy-agent-skills

The same repository also documents setup paths for Cursor, Antigravity CLI, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro, Codex, and generic agents that can load Markdown instructions or skills directories.

Command Surface

The README maps seven slash commands to the software lifecycle:

Phase	Command	Purpose
Define	/spec	Turn an idea into a concrete product or engineering spec.
Plan	/plan	Break the spec into small implementation tasks.
Build	/build	Implement in incremental, testable slices.
Verify	/test	Prove behavior through tests and debugging.
Review	/review	Run quality, maintainability, and readiness review.
Simplify	/code-simplify	Reduce complexity while preserving behavior.
Ship	/ship	Prepare release, rollout, monitoring, and rollback work.

The README also documents /build auto, which creates a plan and implements tasks in a single approved pass. That mode is useful for bounded specs, but it should stay behind explicit plan review, test checks, and stopping conditions.

Capability Scope

The pack includes 24 skills: the using-agent-skills meta-skill plus 23 lifecycle skills. They cover idea refinement, specs, task planning, incremental implementation, TDD, context engineering, source-driven development, doubt-driven review, UI engineering, API design, browser testing, debugging, code review, simplification, security hardening, performance, git workflow, CI/CD, migrations, ADRs, observability, and shipping.

It also includes specialist agent personas:

Persona	Focus
code-reviewer	Staff-level code review across quality and maintainability.
test-engineer	Test strategy, coverage analysis, and proof-oriented QA.
security-auditor	Threat modeling, vulnerability detection, and OWASP review.
web-performance-auditor	Core Web Vitals and browser performance review.

Reference checklists cover testing, security, performance, accessibility, and multi-persona orchestration patterns.

Production Rules

• Start with /spec and /plan for ambiguous or multi-step work.
• Keep /build auto limited to reviewed, bounded specs with testable tasks.
• Require human approval for commits, pushes, dependency changes, CI changes, infrastructure changes, and deploys.
• Verify current docs and installed versions before accepting framework, package, cloud, browser, or security-specific code.
• Treat review personas as focused reviewers, not as authority to bypass the repository's maintainer process.
• Redact secrets and private production data from prompts, browser traces, logs, screenshots, and PR text.

Use Cases

• Add a lifecycle command set to Claude Code for spec, plan, build, test, review, simplify, and ship workflows.
• Give Codex, Cursor, Gemini CLI, OpenCode, or another agent a consistent engineering process through Markdown skills or project instructions.
• Use the source-driven-development and doubt-driven-development skills to force current documentation checks and adversarial review before implementation.
• Run a structured code review with the code-reviewer, security-auditor, test-engineer, or web-performance-auditor persona.
• Standardize engineering rituals across a team without baking all of that context into every prompt.

Troubleshooting

Marketplace install fails over SSH

Use the HTTPS install form documented in the README, or configure GitHub SSH keys before adding the marketplace repository.

The agent starts making broad changes

Pause the run and ask it to return to /spec or /plan. Break the work into small tasks with acceptance criteria, tests, rollback notes, and approval gates.

Framework code looks stale

Invoke or mirror the source-driven workflow. Re-check official documentation, installed package versions, changelogs, and local type errors before accepting implementation details.

/build auto is too aggressive

Use it only after reviewing the plan. If the task is exploratory, risky, security-sensitive, or production-facing, require task-by-task approval instead of one-pass execution.

Source Review

Verified on 2026-06-18:

• The README describes the repository as production-grade engineering skills for AI coding agents.
• The README documents Claude Code marketplace installation, an HTTPS fallback, local development setup, and setup paths for Cursor, Antigravity CLI, Gemini CLI, Windsurf, OpenCode, GitHub Copilot, Kiro, Codex, and generic agents.
• The README lists seven lifecycle commands: /spec, /plan, /build, /test, /review, /code-simplify, and /ship.
• The README says /build auto creates a plan and implements tasks in one approved pass while still pausing on failures or risky steps.
• The README lists 24 skills total, including the using-agent-skills meta-skill and 23 lifecycle skills.
• The repository includes review personas for code review, tests, security, and web performance.
• plugin.json names the plugin agent-skills and describes it as production-grade engineering skills for AI coding agents.
• GitHub reports MIT licensing, a current release, and active repository updates for addyosmani/agent-skills.

Duplicate Review

Checked current content/skills/, content/tools/, content/mcp/, content/agents/, open pull requests, and repository-wide content for addyosmani/agent-skills, Addy Osmani Agent Skills, Addy Agent Skills, agent-skills@addy-agent-skills, /spec, /plan, /build auto, and the repository URL. Adjacent official skill-pack entries exist, but no dedicated Addy Osmani Agent Skills collection entry, exact source URL duplicate, target file, or open duplicate PR was found.

Disclosure

Editorial listing. No paid placement or affiliate link is used. The collection is maintained by Addy Osmani and published from the public GitHub repository.