TL;DR
Anthropic published lessons from using hundreds of Claude Code Skills across its engineering organization, according to a July 1 analysis by Thorsten Meyer AI. The key finding: Skills are better understood as reusable folders containing instructions, scripts, references and checks, not as saved prompts.
Anthropic has published an engineering account of how its Claude Code team uses Skills, after running hundreds across its engineering organization, and the central point is that a Skill is a reusable folder of instructions, scripts and references rather than a saved prompt. The finding matters for teams adopting coding agents because it reframes repeated AI instructions as shared operating assets that can be versioned, tested and reused.
The first confirmed point is definitional. Anthropic’s Claude Code Skills documentation says Skills extend Claude by adding a SKILL.md file with instructions, plus optional supporting files. The docs describe Skills as discoverable by Claude when relevant or invocable directly with a /skill-name command, with a description field used to help the model decide when to load them.
According to the Thorsten Meyer AI analysis, Anthropic’s internal Skills clustered into nine categories: library and API reference, 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 same analysis says Anthropic’s own measurement found verification Skills, which check work after generation, had the largest effect on output quality.
The report distinguishes between what is confirmed and what is claimed. It is confirmed that Claude Code supports Skill folders, supporting files, scripts, frontmatter, project and personal locations, and related controls in current public docs. Claims about the scale of Anthropic’s internal library and the relative performance gains from specific Skill categories come from Anthropic’s engineering account as summarized by Thorsten Meyer AI; the material is vendor engineering reporting, not peer-reviewed research.
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 Become Operating Assets
For engineering leaders, the shift is about repeatability. If a team keeps pasting the same checklist, deployment rule, testing recipe or review policy into an agent, a Skill turns that recurring guidance into a shared workflow. That can reduce variation between users and help new staff or new agents follow the same process as experienced engineers.
For developers, the practical change is that a Skill can include runnable scripts, templates and references instead of prose alone. That matters because a coding agent can compose around existing tools rather than recreating boilerplate each time. The reported strength of verification Skills also points to a near-term lesson for AI coding adoption: teams may get more value from agent checks and run recipes than from longer instruction files.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL500_.jpg)
Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Prompts To Folders
Anthropic’s public docs say Skills may live at enterprise, personal, project or plugin scope, and project Skills can be committed in a repository. The docs also say Claude Code can discover nested Skills in subdirectories, allowing monorepos to attach different instructions to different packages.
The Skill authoring guidance stresses concise instructions, clear descriptions and progressive disclosure, where SKILL.md serves as an overview and points Claude to detailed files only when needed. Current Claude Code docs also list bundled Skills such as /code-review, /run and /verify, with run and verify support tied to Claude Code version requirements in the documentation.
“Create a SKILL.md file with instructions, and Claude adds it to its toolkit.”
— Claude Code documentation
The AI Retainer Playbook: How to Build Predictable, Recurring Income Selling AI Workflows to Local Businesses | No Coding Required | Includes Outreach Scripts, Pricing Calculator & 90-Day Launch Plan
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Adoption Metrics Still Missing
The public material does not disclose enough detail to independently evaluate the internal results. It is not yet clear what sample size, baseline, grading method or task mix Anthropic used when measuring the reported quality gains from verification Skills.
It is also unclear how well the pattern transfers outside Claude Code, Anthropic’s engineering culture or teams with mature internal tooling. The analysis flags several limits: best practices are still changing, checked-in Skills can add context cost after loading, and a large Skill library may require active curation rather than simple accumulation.
Visual Studio Code USER GUIDE For Beginners 2026: Learn Project Setup, Code Editing, Testing, Version Control, Productivity Tools, and Modern Development Workflows
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Teams Test Verification Skills
The next practical step for companies is likely small-scale testing. The Thorsten Meyer AI analysis recommends starting with one Skill, one known caveat and the category that catches mistakes, especially verification workflows that confirm whether an agent’s output works.
The next milestone to watch is whether teams publish more measured comparisons of agent performance with and without Skills. Wider adoption will also depend on governance around ownership, security review, script permissions and how shared Skills are updated when underlying systems change.
The Human-Agent Orchestrator: Leading and Scaling AI-Driven Organizations
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did Anthropic report about Claude Code Skills?
Anthropic reported lessons from using hundreds of Skills across its engineering organization, according to the July 1 Thorsten Meyer AI analysis. The main takeaway is that Skills are reusable folders that can include instructions, references, scripts and templates.
Is a Skill just a saved prompt?
No. A saved prompt is mainly text, while a Skill folder can contain SKILL.md, reference files, scripts, templates and configuration. Claude Code’s public docs confirm that supporting files can make Skills more powerful than a single instruction file.
Which Skill category had the biggest reported effect?
According to the source analysis, Anthropic’s own measurement found verification Skills had the largest impact on output quality. Those Skills focus on checking whether generated work actually passes project-specific tests, launch steps or review rules.
Are Anthropic’s findings peer-reviewed?
No. The material is company engineering reporting and product documentation, not peer-reviewed research. Readers should treat the internal measurements as Anthropic’s reported results unless more detailed methodology is published.
What should teams do next if they use AI coding agents?
Teams should identify one repeated agent task, especially a verification or review step, and turn it into a small Skill with clear instructions and any needed script. The near-term test is whether that Skill improves consistency without adding too much maintenance burden.
Source: Thorsten Meyer AI
