How to Integrate Distilled cangjie-skill Skills with Claude Code or Cursor

Copy or symlink the generated skill directories from books/<book-slug>/<skill-slug>/ into your host's skills folder—either ~/.claude/skills/ for user-wide access or <project>/.claude/skills/ / <project>/.cursor/skills/ for project-scoped availability.

The cangjie-skill pipeline transforms books, video transcripts, podcasts, and courses into atomic, executable Claude skills that AI agents can invoke in real-world scenarios. After the pipeline completes, these skills live as artifacts in the books/ directory tree. To make them discoverable by Claude Code or Cursor, you must complete a manual integration step that copies or symlinks each verified skill into the host environment's designated skills location. This article explains the exact paths, commands, and verification steps derived from the kangarooking/cangjie-skill source code.

Where Distilled Skills Live After Generation

The cangjie-skill pipeline produces self-contained skill directories during its delivery stage. According to methodology/07-stage5-deliver.md (lines 44-52), each skill is packaged as:

books/<book-slug>/<skill-slug>/
├── SKILL.md              # Canonical skill definition with RIA++ front-matter

└── test-prompts.json     # Darwin-skill compatible test cases

These artifacts are not automatically installed. The pipeline stops at generation—you must explicitly move them into Claude Code or Cursor's load path.

Choosing Your Installation Target

The integration supports two scopes, selected via user prompt as described in methodology/07-stage5-deliver.md (lines 48-51):

Scope	Claude Code Path	Cursor Path	Use Case
User-wide	~/.claude/skills/<skill-slug>/	~/.cursor/skills/<skill-slug>/	Available across all projects
Project-wide	<project>/.claude/skills/<skill-slug>/	<project>/.cursor/skills/<skill-slug>/	Scoped to single codebase

The user-wide location keeps skills globally accessible. The project-wide location versions skills with your repository, making them portable for team members who clone the project.

Step-by-Step Integration Commands

The templates/INDEX.md.template file (lines 59-68) provides ready-to-paste commands for both hosts.

Installing for Claude Code

# User-wide installation

cp -r books/my-book/effective-learning/ ~/.claude/skills/

# Project-wide installation

cp -r books/my-book/effective-learning/ ./.claude/skills/

Installing for Cursor

# User-wide installation

cp -r books/my-book/effective-learning/ ~/.cursor/skills/

# Project-wide installation

cp -r books/my-book/effective-learning/ ./.cursor/skills/

Using Symlinks for Development

For skills you actively iterate on, symlink instead of copy to avoid manual re-sync:

# Claude Code project-wide symlink example

ln -s "$(pwd)/books/my-book/effective-learning" ./.claude/skills/effective-learning

Verifying Successful Integration

After copying, confirm the host loads the skill correctly. As noted in methodology/07-stage5-deliver.md (lines 53-55), run a trigger test:

# In Claude Code console

should_trigger "我想学习如何高效记忆"

If the skill's SKILL.md description matches your query, the agent will indicate it triggered the skill. A positive response confirms the integration succeeded.

Automating Bulk Installation

For pipelines generating multiple skills, automate installation with a Python script that filters for verified skills only:

import pathlib
import shutil

book_dir = pathlib.Path("books/my-book")
install_dir = pathlib.Path("~/.claude/skills").expanduser()

for skill_dir in book_dir.glob("*/"):
    # Only install skills with passing tests

    if (skill_dir / "test-results.md").exists():
        shutil.copytree(
            skill_dir,
            install_dir / skill_dir.name,
            dirs_exist_ok=True
        )
        print(f"Installed: {skill_dir.name}")

This script checks for test-results.md as a quality gate before copying, ensuring only validated skills reach your agent.

Optional: Registering with Darwin-Skill for Evolution

The generated test-prompts.json files enable automatic skill improvement via the Darwin framework. From templates/INDEX.md.template (lines 72-78):

darwin evolve books/my-book/

This command consumes the test suites in each skill directory to iteratively refine trigger accuracy and response quality.

Key Files Reference

File	Purpose	Location
SKILL.md	Skill definition with RIA++ structure	books/<slug>/<skill>/SKILL.md
test-prompts.json	Test cases for validation and evolution	books/<slug>/<skill>/test-prompts.json
methodology/07-stage5-deliver.md	Official delivery stage specification	methodology/07-stage5-deliver.md
templates/INDEX.md.template	Copy commands and Darwin integration	templates/INDEX.md.template

Summary

• cangjie-skill generates skills as directories under books/<book-slug>/<skill-slug>/ with SKILL.md and test-prompts.json.
• Integration requires manual copying or symlinking into ~/.claude/skills/, ./.claude/skills/, or ./.cursor/skills/ depending on scope.
• Verify installation with should_trigger "<test query>" to confirm the host loads the skill.
• Automate bulk installation by checking for test-results.md as a quality filter.
• Use darwin evolve with the included test-prompts.json files to continuously improve skill performance.

Frequently Asked Questions

What files must be present for a skill to work in Claude Code or Cursor?

A complete skill requires SKILL.md (the canonical definition with RIA++ front-matter) and optionally test-prompts.json for testing and evolution. Both are generated automatically by the cangjie-skill pipeline in each skill directory under books/<book-slug>/.

Can I use the same skill directory for both Claude Code and Cursor simultaneously?

Yes. The skill format is host-agnostic. Install the same directory to both ~/.claude/skills/ and ~/.cursor/skills/, or use symlinks pointing to a single source of truth in your books/ tree.

How do I update a skill after the source material changes?

Re-run the cangjie-skill pipeline to regenerate the skill directory, then repeat the copy or symlink step. If you used symlinks, the update is automatic; if you copied, overwrite the previous installation.

Does cangjie-skill support automatic installation without manual copying?

No. According to methodology/07-stage5-deliver.md, the pipeline deliberately stops at artifact generation. The user must explicitly choose the installation target and execute the copy/symlink step, ensuring conscious control over which skills enter the agent environment.

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