How to Install Distilled Skills to Claude Code or Cursor Skills Directory

Copy your generated skill folders from books/<book-slug>/<skill-slug>/ into ~/.claude/skills/, <project>/.claude/skills/, or .cursor/skills/ to activate them for your AI agent.

The kangarooking/cangjie-skill repository converts long-form content—books, video transcripts, podcasts, or courses—into atomic, reusable skills. After the pipeline finishes distilling the content, each skill resides in its own folder under the books/ directory, waiting to be moved into your Claude Code or Cursor skills directory to become accessible to your agent.

Where Generated Skills Are Stored

After the delivery stage completes, cangjie-skill outputs atomic skill bundles to a predictable path structure:

books/<book-slug>/<skill-slug>/

Each skill folder contains two critical files:

• SKILL.md — The canonical skill definition containing instructions, examples, and metadata
• test-prompts.json — Validation prompts used to verify the skill triggers correctly

According to the delivery stage documentation in methodology/07-stage5-deliver.md (lines 46-52), the pipeline treats these folders as the source of truth before they are distributed to your local agent directories.

Installation Path Options

The cangjie-skill pipeline supports three distinct installation targets, asking you once during the delivery stage which destination to use rather than prompting per skill (methodology/07-stage5-deliver.md, lines 48-52):

Scope	Claude Code Path	Cursor Path
User-level (global access)	~/.claude/skills/<skill-slug>/	Not applicable
Project-level	<project>/.claude/skills/<skill-slug>/	.cursor/skills/<skill-slug>/

Choose user-level installation when you want the skill available across all projects. Select project-level when the skill is context-specific to a single codebase.

Installing Skills to Claude Code

To make a skill available to Claude Code, use the cp -r command to copy the entire skill folder into the appropriate directory.

For user-level access (recommended for general-purpose skills):

cp -r books/my-book/awesome-framework ~/.claude/skills/

For project-level access (recommended for domain-specific skills):

cp -r books/my-book/awesome-framework ./.claude/skills/

These commands mirror the template provided in templates/INDEX.md.template (lines 63-66), which generates the copy instructions dynamically based on your selected destination during the pipeline's delivery stage.

Installing Skills to Cursor

Cursor follows a similar convention but uses a .cursor directory structure within your project:

cp -r books/my-book/awesome-framework .cursor/skills/

Ensure your Cursor settings are configured to recognize the .cursor/skills/ directory, as Cursor will scan this location to load available skill definitions alongside your project context.

Bulk Installation and Verification

When processing multiple skills from a single source, you can automate the installation by checking for the presence of test-prompts.json to ensure only verified skills are copied:

# From the repository root

for d in books/*/*; do
    if [ -f "$d/test-prompts.json" ]; then
        cp -r "$d" ~/.claude/skills/
    fi
done

After copying, the pipeline validates installation by verifying that at least one skill responds correctly to its should_trigger prompt (step 4 of the delivery stage). You can manually verify this by running a test query:

cd ~/.claude/skills/awesome-framework
echo "I need a framework for rapid product experimentation" | claude-cli --run

If the skill is properly installed, Claude Code should recognize the context and invoke the skill's instructions from SKILL.md.

Alternative: Repository-Only Distribution

If you prefer to keep skills as repository artifacts—for example, when publishing to GitHub without activating them locally—you can skip the copy step entirely. In this mode, skills remain in books/<book-slug>/ and function as documentation or shareable resources rather than active agent capabilities.

Summary

• Generated skills live in books/<book-slug>/<skill-slug>/ and contain SKILL.md plus test-prompts.json
• User-level installation uses ~/.claude/skills/<skill-slug>/ for global availability in Claude Code
• Project-level installation uses .claude/skills/ for Claude Code or .cursor/skills/ for Cursor, scoped to specific repositories
• Verification requires testing the should_trigger prompt to ensure the skill activates correctly
• Skip installation entirely if you only need to distribute skills as static repository artifacts

Frequently Asked Questions

Can I use symlinks instead of copying skill folders?

Yes. The delivery stage supports both copying and symlinking folders into the target directory. Symlinks are useful when you want to edit skills in the original books/ location while keeping them active in your agent's skills directory, though you must ensure the source path remains stable.

What happens if I skip the installation step?

If you skip copying skills to the Claude Code or Cursor directories, the skills remain as static artifacts in the books/ folder. They will not be active in your AI agent, but you can still version them in GitHub, share them with others, or manually reference the SKILL.md files as documentation.

How do I verify that a skill was installed correctly?

Navigate to the installed skill directory in your skills folder and test whether the agent recognizes the should_trigger prompt defined in test-prompts.json. The cangjie-skill pipeline automatically performs this verification during step 4 of the delivery stage, confirming the skill responds to at least one trigger sentence before marking installation complete.

Does the installation process differ between Claude Code and Cursor?

The process is nearly identical, but the destination paths differ. Claude Code uses .claude/skills/ or ~/.claude/skills/, while Cursor uses .cursor/skills/. Both agents expect the same folder structure containing SKILL.md, so you can install the same skill to both directories if you use both tools on the same project.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}