Skip to content

Claude Code guide in three levels, Beginner to Advanced: hands-on lessons with checked exercises, every volatile fact cited to its source. Independent community guide.

5.3 KB
Raw Blame

Skill patterns: live data, supporting files and a context of its own

Intermediate · Elective · about 20 minutes · Needs: Turn a repeated workflow into a skill · Verified against Claude Code v2.1.292 (stable) on 2026-10-15

Goal

By the end of this Elective you can write a skill that pulls live data into its prompt, keeps long reference material in a separate file, and runs in its own subagent.

You need

  • A repository with some uncommitted changes to review, and the skills from lesson 1.
  • Usage: light.

The idea

Lesson 1's skills were instructions and $ARGUMENTS. A few more patterns make a skill do more with less context (Extend Claude with skills):

Pattern Written as What it does
Live data !`git diff HEAD` in the body Claude Code runs the command before Claude reads the skill, and puts its output in its place
Supporting files Files beside SKILL.md, linked from it Long reference material that Claude reads only when it needs it
A context of its own context: fork in the frontmatter The skill becomes a subagent's prompt, and the subagent runs in the background. It doesn't see your conversation, and its result arrives when it finishes
Named arguments arguments: [focus], then $focus Each word you type after the name fills its own placeholder
Tool grants allowed-tools: Read, Grep Claude may use these tools without asking, during the turn that runs the skill. It removes no other tool

Keep SKILL.md short: once a skill runs, its whole text stays in the conversation. The docs suggest under 500 lines, with detail moved to supporting files.

Worked example

  1. Save this as .claude/skills/changes-review/SKILL.md:

    ---
    name: changes-review
    description: Reviews the uncommitted changes in this repository against the team checklist. Use when the user asks for a review before committing.
    context: fork
    allowed-tools: Bash(git diff *)
    ---
    
    ## Changes
    
    !`git diff HEAD`
    
    ## Your task
    
    Review the changes above against [checklist.md](checklist.md). Report each problem with its file and line, then one line on whether the change is ready to commit.
    
  2. Save the checklist beside it, as .claude/skills/changes-review/checklist.md:

    - Every changed function has a test.
    - Errors are reported, not swallowed.
    - No secrets in code or config.
    
  3. Make a small change in a source file, start claude and run /changes-review. Claude Code runs git diff HEAD and puts its output in the skill, then starts a subagent with it. The subagent reads the checklist, and its report arrives in your conversation when it finishes. The allowed-tools rule pre-approves git diff commands. git diff HEAD is read-only and would run without it, but outside auto mode an injected command your permission rules don't allow stops the skill.

  4. For another skill that pre-approves tools, read this guide's claude-md-review. Its allowed-tools: Read, Glob, Grep lets it read files without prompts; it doesn't stop Claude from using other tools (Pre-approve tools for a skill).

Your turn

Make the review take a focus:

  1. Add arguments: [focus] to the frontmatter of changes-review.
  2. Change the task to put the review's focus first: Review the changes above against checklist.md, looking hardest at $focus.
  3. Run /changes-review tests, then /changes-review secrets, and compare the two reports.

Check

You are done when all of these are true:

  • head -8 .claude/skills/changes-review/SKILL.md shows context: fork and arguments: [focus].
  • grep -c 'git diff HEAD' .claude/skills/changes-review/SKILL.md prints 1, and checklist.md sits beside SKILL.md.
  • Self-check (not tested): the two reports looked hardest at different things.

If /changes-review stops with Shell command failed or Shell command permission check failed, run git diff HEAD yourself, and check the allowed-tools line: an injected command that fails, or that your permission rules don't allow, aborts the whole skill (Inject dynamic context).

Go further

  • Skill authoring best practices: how to write descriptions, structure supporting files and test a skill.
  • anthropics/skills: Anthropic's example skills. Some are source-available rather than open source, so read each one's license before you copy it.
  • Troubleshooting: skills that don't trigger, trigger too often, or stop being followed.

Sources: Extend Claude with skills · anthropics/skills

Back to the Intermediate index · Topic: Skills · Stuck on this lesson?