TL;DR
Anthropic has published lessons from using hundreds of Claude Code Skills across its engineering organization. The company presents Skills as folder-based reusable work units, not saved prompts, and says verification Skills had the strongest measured effect on output quality.
Anthropic has published lessons from using hundreds of Claude Code Skills across its engineering organization, offering a clearer view of how the company turns repeated agent instructions into shared, reusable workflows that can be versioned, run, and improved over time.
The source post, Lessons from building Claude Code: How we use skills, was written by Thariq Shihipar and published on Anthropic’s Claude blog on June 3, 2026, according to the provided source material. A July 1, 2026 write-up from Thorsten Meyer AI framed the post as more than a coding guide: it described Skills as a way to turn ad hoc prompting into durable institutional capability.
The confirmed core of Anthropic’s explanation is definitional. A Skill is a folder, not simply a markdown prompt. That folder can include SKILL.md instructions, references, runnable scripts, assets, templates, configuration files, hooks, and memory. The agent can discover the folder, read the root instructions, and pull in deeper material only when the task calls for it.
Anthropic’s internal Skills clustered into nine categories, according to the source summary: library and API references, product verification, data fetching and analysis, business-process automation, code scaffolding and templates, code quality and review, CI/CD and deployment, runbooks, and infrastructure operations. The company said product verification Skills, which check whether work was done correctly, produced the largest measured improvement in output quality.
A Skill is a folder, not a prompt
Anthropic published what it learned running hundreds of Skills across its own engineering org. Read as a business memo, the point is bigger than a coding trick: this is how ad-hoc prompting becomes durable institutional capability — the SOPs your agents actually follow, versioned and shared.
“A Skill is just a clever markdown prompt you save in a file.”
A folder the agent can discover, read & run — instructions, scripts, references, templates, config & on-demand hooks.
The knowledge of how your organization actually operates can be captured, versioned, shared & executed — and the thing capturing it is a humble folder with a script and a gotchas list inside. For the builder, that’s context engineering with real tools attached. For whoever owns the budget, it’s the difference between AI that starts from zero every morning and an asset that compounds. Caveats: best practices are still evolving, checked-in Skills cost context, and curation beats accumulation. Start with one Skill, one gotcha, and the category that catches your mistakes.
Skills Turn Prompts Into Assets
The development matters because many teams using AI coding agents still rely on repeated instructions typed into chats or pasted into prompts. Anthropic’s approach treats those instructions, examples, scripts, and checks as shareable software assets that can be reviewed and maintained like other engineering materials.
For engineering leaders, the claim is that Skills can make agent work more consistent across teams and reduce dependence on undocumented personal habits. For developers, the practical difference is that a Skill can contain real executable code, reusable templates, and guardrails rather than prose alone.
The strongest business implication is in the verification category. If Anthropic’s measurement holds outside its own environment, Skills that test, inspect, or validate AI-generated work may offer more value than Skills that only generate drafts. That makes the approach relevant for teams trying to reduce review burden while keeping control over output quality.
AI for Project Managers: A Desk Reference & Field Guide: Use Artificial Intelligence to Streamline Workflows, Automate Tasks, and Make Smarter Decisions with Practical Tools and Ethical Insights
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
How Anthropic Describes Skills
The provided source says Anthropic’s recommended Skill structure starts with a SKILL.md file containing root instructions and a description written for the model, which acts as the trigger. Deeper materials can live in references, scripts, assets, and configuration files so the agent does not have to load every detail at once.
This design uses what the source calls progressive disclosure: the agent reads the main instructions first, then reaches into supporting files only when needed. The Thorsten Meyer AI analysis compares that to giving a new employee a short guide that points to detailed documentation when the work gets more specific.
The source also lists several craft lessons from Anthropic’s experience: describe Skills for the model rather than for people, avoid stating the obvious, include scripts instead of only prose, use on-demand guardrail hooks, let Skills remember useful history, and leave the agent enough room to adapt to the task.
50 AI Workflows for Engineers: From Debugging to System Design, Code Review & Engineering Automation
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Limits Outside Anthropic Remain Open
Several details remain unclear from the provided source material. Anthropic’s exact measurement method, sample size by Skill category, and the size of the reported quality gains are not specified in the summary. It is also unclear how well the same results would transfer to smaller teams, non-engineering departments, or organizations using different AI agents.
The source also flags limits in the approach. Best practices are still evolving, checked-in Skills can add context cost, and a growing library may become less useful without active curation. The practical risk is that teams could collect folders faster than they improve or retire them.
Technical Writing for Engineers: Documentation, Diagrams & API References
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Teams May Start With Checks
The clearest next step for readers is to watch whether teams begin building verification-first Skill libraries rather than large collections of generation prompts. The source recommends starting with one Skill, one hard-earned caveat, and the category most likely to catch mistakes.
Anthropic’s public documentation at code.claude.com/docs/en/skills is the next reference point for teams evaluating the format. What happens next will depend on whether organizations can turn scattered operating knowledge into small, maintained Skill folders that agents actually use in daily work.
Mastering GitHub Actions: Advance your automation skills with the latest techniques for software integration and deployment
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic announce about Claude Code Skills?
Anthropic published lessons from using hundreds of Skills across its engineering organization, explaining how Skills package reusable instructions, scripts, references, templates, and checks for Claude Code.
Is a Skill just a saved prompt?
No. According to the source material, a Skill is a folder that can contain a SKILL.md file, supporting references, scripts, assets, configuration, hooks, and memory. The agent can read and run parts of that folder as needed.
Which type of Skill had the biggest impact?
The provided source says Anthropic found product verification Skills had the largest measured impact on output quality. Those Skills focus on checking work rather than only producing new output.
Why does this matter for companies using AI agents?
It gives companies a way to turn repeated instructions and internal know-how into versioned workflow assets. That may help teams make agent work more consistent, easier to review, and less dependent on one person’s private prompting habits.
What is still unknown?
The source summary does not provide Anthropic’s full measurement details or show whether the same quality gains would appear in other companies. It is also unclear how much maintenance a large Skills library requires over time.
Source: Thorsten Meyer AI
