Skill patterns and examples

The easiest way to understand skills is to look for work that has repeatable judgment. If the task is only a single command, a tool is enough. If the task requires a particular sequence, policy, voice, output format, or validation loop, it is a skill candidate.

For Claude-specific skill browsing, use Claude Has Skills. This page stays cross-ecosystem and focuses on patterns that can be expressed in the open skill shape.

Document production skills

Document work is a natural fit because success is rarely "write text." The agent needs the right format, template, citations, rendering checks, and sometimes a repair loop for generated files. A document skill can include brand templates under assets/, formatting rules in SKILL.md, and validation scripts that inspect the output.

Brand and voice skills

Brand skills are useful when a team has a voice that cannot be reduced to "professional" or "friendly." The skill should include examples, forbidden phrases, rhythm, vocabulary, and a review checklist. It should not include every page of the brand book unless the task routinely needs every page.

The durable value is consistency. A user can ask for launch copy, a support response, or an executive summary without re-teaching the agent the organization's voice every time.

Code review skills

Code review skills work when they encode local risk. A general model already knows to look for bugs. The skill should tell it which contracts matter in this repository, which tests are meaningful, which generated files should not be touched, how to inspect live review threads, and what kind of final verdict the team expects.

---
name: repo-review
description: Reviews pull requests in this repository for regressions, missing tests, API contract drift, generated file churn, and unresolved review threads. Use when asked to review or merge-gate a PR.
---

That kind of skill is only portable if the repository-specific bits are either references or clearly named assumptions. A generic review skill can travel; a repo review skill is intentionally local.

Release readiness skills

Release skills are valuable because they turn a fragile endgame into an explicit checklist: build, run focused tests, inspect generated artifacts, deploy, verify the live surface, and write the final report. They are especially useful for static sites and small apps where the agent can complete the whole loop.

MCP development skills

MCP server development is a strong example of skills and MCP cooperating. The MCP docs describe a reference set of MCP development skills that guide an AI coding assistant through design, deployment model, tool patterns, auth, and packaging decisions. The skills do not replace MCP; they help an agent build or reason about MCP systems more reliably.

A practical MCP development skill might ask what system the server connects to, whether it is local or remote, who will use it, which auth flow applies, and whether the output should be plain text, structured data, elicitation, or a richer app surface.

Data analysis skills

Data analysis skills are useful when the hard part is not plotting a chart but knowing the organization's data shape. The skill can point to schema references, canonical metric definitions, approved filters, quality checks, and export formats. Scripts can handle repetitive extraction or validation without asking the model to invent command-line glue every time.

Anti-examples

Not every repeated instruction deserves a skill. Avoid skills that merely say "be concise" or "write good code." Those belong in general instructions, if anywhere. Avoid skills that wrap a single command with no judgment. Make a shell alias or tool instead. Avoid skills that depend on private services without stating that dependency in frontmatter or documentation.

The best skill examples feel like small operating manuals: enough structure to improve the work, not so much structure that the agent spends the task reading instead of doing.