A coding agent that produces a working API endpoint has completed one step. The endpoint still needs a contract, input validation, tests, a review against a spec, monitoring before it goes live, and a rollback path if the rollout fails. Those are not optional extras. They are the difference between a prototype and production software.
What the repository is
Agent Skills is a collection of production-oriented engineering workflows maintained by Addy Osmani, with documented collaborators Federico Bartoli and Joan León. It installs into Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI, and other Agent Skills-compatible hosts. Skills span six phases—define, plan, build, verify, review, and ship—with nine slash commands as entry points.
The README describes the skills as encoding practices “from Google’s engineering culture,” citing Software Engineering at Google and Google’s engineering practices guide. Hyrum’s Law, the Beyonce Rule, and trunk-based development appear as workflow steps. The repository does not claim to be an official Google product.
Four areas across the lifecycle
Specifications and planning. /spec activates spec-driven-development: a product requirements document covering objectives, commands, code structure, testing strategy, and boundaries—before implementation. /plan activates planning-and-task-breakdown: small tasks with acceptance criteria and dependency ordering. The README’s principle: “Spec before code.”
Incremental implementation and tests. /build activates incremental-implementation: thin vertical slices—implement, test, verify, commit—with feature flags and rollback-friendly defaults. /test activates test-driven-development: red-green-refactor enforced. The Beyonce Rule is embedded: if you liked it, you should have put a test on it.
Security and performance review. security-and-hardening covers OWASP Top 10 prevention, authentication patterns, secrets management, and dependency auditing. performance-optimization enforces a measure-first rule: profile before optimizing, with Core Web Vitals targets and bundle analysis. /webperf invokes a performance auditor persona with a documented metric-honesty rule. code-review-and-quality applies a five-axis review before merge, targeting changes of ~100 lines.
Deployment and observability. observability-and-instrumentation covers structured logging, RED metrics (Rate, Errors, Duration), OpenTelemetry tracing, and symptom-based alerting—added as part of the build, not after. shipping-and-launch covers pre-launch checklists, feature flag lifecycle, staged rollouts, and rollback procedures. The stated principle: “Faster is safer.”
The API endpoint, beyond implementation
A developer asks an agent to add a customer-facing API endpoint. The route, handler, and database query are the middle of the work.
Before: api-and-interface-design produces a contract—what the endpoint accepts, what it returns on success, on validation failure, and on a downstream timeout. Hyrum’s Law applies: every observable behavior becomes a dependency. Defining error semantics is the work.
During: input validation at the boundary, tests for the happy path and each documented error case, committed in a reviewable slice.
After: code-review-and-quality checks the change against the spec. observability-and-instrumentation adds a structured log line and an error rate metric before the endpoint receives traffic. shipping-and-launch defines the rollout: feature flag, staged percentage, rollback trigger. The endpoint is not done until the first-hour failure mode has an owner.
Who this is for
Engineers and engineering teams using coding agents to produce changes that will run in production. The repository positions itself against Superpowers and Matt Pocock’s skills as covering more of the lifecycle—verify, review, and ship—rather than going deeper on any single phase.
The limitation
Installing workflow instructions does not create a mature engineering organization. /review produces a structured review; it does not produce an accountable reviewer. observability-and-instrumentation describes what to instrument; it does not confirm anyone receives the alerts. Quality depends on functioning CI, real reviewers, and operational ownership. The skills structure the work. They do not replace the organization around it.
The argument
Developer value here is structure that makes agent-generated work verifiable: a spec to check against, tests that produce evidence, a review with explicit axes, and a rollback path with a defined trigger. Generating more code faster is a different goal, and these skills are not designed for it.
Source: github.com/addyosmani/agent-skills — README. Retrieved 8 October 2026.
#SoftwareEngineering #AgentSkills #ProductionReadiness