addyosmani/agent-skills for Reliable AI Coding

Overview

addyosmani/agent-skills treats AI-assisted coding as a controlled lifecycle rather than a prompt-to-code shortcut. The series follows how clarified intent becomes a bounded specification and dependency-aware plan, how focused and trust-aware context supports a thin tested slice, and how interfaces, retries, quality gates, telemetry, staged rollout, and rollback carry evidence toward operation. Its central claim is conditional: explicit workflows make agent decisions more inspectable and mistakes easier to localize and recover, but skills, evals, and host adapters do not guarantee semantic correctness, host equivalence, production safety, or durable cross-session memory. Adoption should match risk, from focused mechanisms for small problems to fuller lifecycle controls for high-impact changes and new projects.

Learning outcomes

Explicit lifecycle phases make agent decisions and evidence easier to inspect, but they do not guarantee good outcomes.
Focused, trust-aware context reduces invented APIs and stale assumptions without replacing project-specific judgment.
Thin slices, failing reproductions, and atomic history make failures easier to localize, correct, and reverse.
Interfaces, quality gates, telemetry, staged rollout, and rollback extend evidence beyond a green local build.

Keywords

addyosmani/agent-skillsworkflowsspecificationstestingcontextinterfacessecurityobservabilityrolloutrollback

Chapters

addyosmani/agent-skills: Workflow to Release

Why explicit phases matter more than prompt speed

8 min · Why explicit phases matter more than prompt speed

8 min

addyosmani/agent-skills: Five-Layer Model

How workflows and evidence keep responsibilities separate

8 min · How workflows and evidence keep responsibilities separate

8 min

addyosmani/agent-skills: Intent to Plan

Clarify scope before writing a dependency-aware slice

10 min · Clarify scope before writing a dependency-aware slice

10 min

addyosmani/agent-skills: Context for Correctness

Select, place, and distrust information at the task boundary

9 min · Select, place, and distrust information at the task boundary

9 min

addyosmani/agent-skills: Readiness Gate

Five conditions before a tested slice

9 min · Five conditions before a tested slice

9 min

addyosmani/agent-skills: Reversible Slices

RED, GREEN, and focused evidence before expansion

10 min · RED, GREEN, and focused evidence before expansion

10 min

addyosmani/agent-skills: Trust Boundaries

Interfaces, retries, and untrusted inputs need stateful controls

8 min · Interfaces, retries, and untrusted inputs need stateful controls

8 min

addyosmani/agent-skills: Quality Gates

Keep green checks from hiding weak or unfinished work

10 min · Keep green checks from hiding weak or unfinished work

10 min

addyosmani/agent-skills: Safe Operation

From CI and baselines to telemetry, rollout, and rollback

10 min · From CI and baselines to telemetry, rollout, and rollback

10 min

addyosmani/agent-skills: Portability and Evals

Shared workflow core, native hosts, bounded evidence

9 min · Shared workflow core, native hosts, bounded evidence

9 min

addyosmani/agent-skills: Guarantees and Limits

Adopt controls that match risk, evidence, and maintenance cost

12 min · Adopt controls that match risk, evidence, and maintenance cost

12 min

Preview every chapter · Unlock full text, audio, and video from chapter 4

Plans

Ch.1-3 FreeCh.4-11 Premium

FAQ

What problem does addyosmani/agent-skills solve?

addyosmani/agent-skills makes AI-assisted engineering more inspectable by turning implicit judgment into workflows, boundaries, checkpoints, and evidence.

Does addyosmani/agent-skills guarantee correct production software?

No. addyosmani/agent-skills cannot guarantee semantic correctness, host equivalence, production safety, or durable cross-session memory.

When should a team adopt the full addyosmani/agent-skills lifecycle?

A team should use the full lifecycle when a change has meaningful blast radius, crosses public or trust boundaries, or needs a shared path from intent to launch.

How does addyosmani/agent-skills use tests and operational evidence?

addyosmani/agent-skills uses failing tests, focused verification, review, telemetry, staged rollout, and rollback as different evidence for different risks.