Skip to content

Agent Skills: your own procedures for an AI agent

AI tools · Automation
Not studied

Added 2026-10-29.

Problem

You explain the same procedure to the agent over and over — an incident review, a repository check, release prep. The result comes out slightly different every time.

Solution

"Skills are folders of instructions, scripts, and resources that Claude loads dynamically to improve performance on specialized tasks. Skills teach Claude how to complete specific tasks in a repeatable way." Package the procedure as a skill: a folder with SKILL.md (a description of when to use it, plus the instructions) and scripts if needed. The agent loads the skill when the task matches the description. Official examples and a template are in Anthropic's repository; the open standard is described at agentskills.io.

Example

skills/repo-check/SKILL.md:

---
name: repo-check
description: Check a GitHub repository before adding it to the knowledge base. Use when a tool needs to be assessed by task, tests, CI and maintenance.
---

# Repository check

1. Record HEAD, the date of the last commit, the latest tag and the license.
2. Find tests and CI in the tree, save links pinned to the commit.
3. Run OpenSSF Scorecard.
4. Write down the task, the reason for picking it, the limitation and the next step.

The frontmatter format follows the repository's template/SKILL.md.

Limitations

The README describes the skills as demonstrations. The document skills (docx, pdf, pptx, xlsx) are source-available, not open source. A skill with scripts executes code: review it and run agent-scan before installing someone else's.

Source

How to apply it yourself

  1. Pick one procedure you repeat more than once a week.
  2. Package it as a skill and test it on three real tasks.
  3. Compare the result with working without the skill.
  4. Mark it "Studied" once the skill gives a consistent result.