Everything you need to know about production-grade engineering skills for AI coding agents
Comprehensive answers to common questions about our development lifecycle, skills, integrations, and implementation
Find detailed answers about Agent Skills, how they work, and how to implement them in your development workflow.
Agent Skills are structured workflows that encode the best practices and engineering judgment of senior developers. They're designed to be used by AI coding agents to ensure production-quality code across all phases of development.
Each skill follows a consistent anatomy with clear processes, verification gates, and anti-rationalization tables to prevent shortcuts that compromise quality.
AI coding agents often default to the shortest path, skipping critical practices like specifications, testing, and security reviews. Agent Skills enforce the same discipline that senior engineers bring to production code.
They bake in best practices from Google's engineering culture, concepts from Software Engineering at Google, and proven industry standards to ensure reliable, maintainable code.
Production-grade means:
Skills include concepts like Hyrum's Law in API design, the Beyonce Rule in testing, trunk-based development in git workflows, and feature flags in CI/CD.
There are 24 production-grade skills total: 23 lifecycle skills plus the meta-skill using-agent-skills. Skills are organized across the development lifecycle:
There are 8 slash commands that map to the development lifecycle:
/spec - Define what to build (Spec before code)/plan - Plan how to build it (Small, atomic tasks)/build - Build incrementally (One slice at a time)/test - Prove it works (Tests are proof)/review - Review before merge (Improve code health)/webperf - Audit web performance (Measure before optimize)/code-simplify - Simplify the code (Clarity over cleverness)/ship - Ship to production (Faster is safer)Commands like /build auto can generate and implement tasks autonomously while maintaining verification.
Skills activate based on context and activity:
api-and-interface-designfrontend-ui-engineeringtest-driven-developmentcode-review-and-qualityThe meta-skill using-agent-skills maps incoming work to the right skill workflow based on the task context.
The fastest installation method uses the skills CLI:
For individual skills:
Agent Skills integrates with 70+ AI coding agents including:
Each tool has specific setup instructions in the documentation.
Requirements vary by integration:
All skills are written in plain Markdown and work with any agent that accepts system prompts or instruction files.
The lifecycle follows a structured approach:
Each phase has dedicated skills that enforce best practices and verification gates before proceeding to the next phase.
The Define phase focuses on clarifying requirements:
interview-me: Extracts actual requirements through structured questioningidea-refine: Explores and refines rough conceptsspec-driven-development: Creates detailed PRDs before any codeThis phase ensures you're building the right thing with clear specifications and acceptance criteria.
The Ship phase includes multiple safety mechanisms:
ci-cd-and-automation: Feature flags and quality gate pipelinesshipping-and-launch: Staged rollouts and rollback proceduresobservability-and-instrumentation: Structured logging and tracingdeprecation-and-migration: Code-as-liability mindsetThe principle "Faster is safer" means automated deployments with proper monitoring are safer than manual releases.
Quality is enforced through multiple mechanisms:
Skills like security-and-hardening and performance-optimization include specific quality checks.
The test-driven-development skill enforces:
Tests are treated as proof of functionality, not just documentation.
The security-and-hardening skill includes:
Security is treated as a first-class concern throughout the development lifecycle.
Two methods for Claude Code:
Marketplace install:
Local development:
SSH errors can be resolved by adding SSH keys or using HTTPS URLs.
For GitHub Copilot:
agents/ as Copilot personas.github/copilot-instructions.mdThis creates a hybrid approach where Copilot generates code while following structured workflows.
Yes, skills are designed for broad compatibility:
See the documentation for integration guides for your specific tool.
Contributions are welcome! Follow these guidelines:
docs/skill-anatomy.md for the format specificationCONTRIBUTING.mdAll contributions follow the MIT license and are reviewed by maintainers.
Support resources include:
docs/ directoryJoin our community of 422+ contributors working on improving AI development practices.
Current focus areas include:
Roadmap is driven by community feedback and emerging industry needs. Check GitHub Issues for planned features.