How cangjie-skill Integrates with Claude Code and Cursor Skills Directories

cangjie-skill generates Claude Code-compatible skill packages and installs them into the host's .claude/skills/ or Cursor skills directory, enabling AI agents to invoke the extracted knowledge.

The cangjie-skill repository provides a pipeline that transforms books, videos, podcasts, and courses into executable AI skills. According to the source code in kangarooking/cangjie-skill, the integration relies on a explicit installation step that copies generated skill artifacts from a temporary build directory into the host environment's skills directory.

Build and Installation Pipeline

The integration follows a five-stage workflow defined in the methodology documentation. Each stage prepares the skill for eventual execution within Claude Code or Cursor.

Stage 1: Build Skill Artifacts

The pipeline first extracts principles, frameworks, and cases from source material. It writes these into a temporary books directory at books/<book-slug>/<skill-slug>/, containing SKILL.md, DIGEST.md, and supporting assets.

As documented in methodology/07-stage5-deliver.md, the build output location is explicitly not the final destination: "宿主 (Claude Code / Cursor 等) 不会从这里加载 skill。必须安装" (The host will not load skills from here; installation is required).

Stage 2: Convert to Claude Code Format

Each extracted unit becomes a Claude Code skill with standardized front-matter.

According to methodology/04-stage2-ria-plus.md, the pipeline "把阶段 1.5 通过的每个方法论单元,构造成一个符合 Claude Code skill 规范的 SKILL.md" (constructs each methodology unit into a SKILL.md conforming to Claude Code skill specifications). This includes mandatory YAML front-matter with description and title fields.

Stage 3: Install into Host Skills Directory

The generated skill folder must be copied into the user's local skills directory. The templates/INDEX.md.template file contains the specific installation command:

cp -r {{skill-slug-1}} <project>/.claude/skills/    # Claude Code

For Cursor, the equivalent directory follows a similar pattern (typically ~/.cursor/skills/ or project-specific paths).

Stage 4: Skill Activation

Claude Code reads the description field in the skill's front-matter to determine activation context.

The methodology/04-stage2-ria-plus.md notes: "A2 的产出直接写入 skill frontmatter 的 description 字段 — Claude 据此决定是否激活 skill" (A2's output is written directly to the skill frontmatter's description field—Claude uses this to decide whether to activate the skill).

Stage 5: Runtime Execution

Once installed, the skill becomes an executable agent callable from any Claude Code session or Cursor's skill registry. The repository README confirms this delivers "把通过测试的 skill 安装到 Claude Code / Cursor 的 skills 目录，让它们真正可被调用" (installs tested skills into Claude Code / Cursor skills directories, making them truly callable).

Why cangjie-skill Uses a Copy-Based Integration

Claude Code and Cursor enforce strict directory boundaries for security and organization. They do not traverse external paths to load skills from arbitrary build locations.

This architectural constraint necessitates the explicit copy step. The build process remains independent from the execution environment, while the delivery stage bridges the gap by placing skills where the host environment expects them.

Key Files Driving the Integration

File	Purpose
templates/INDEX.md.template	Emits the copy command targeting .claude/skills/
methodology/07-stage5-deliver.md	Documents the requirement to install skills into the host directory
methodology/04-stage2-ria-plus.md	Defines the Claude Code skill format and front-matter specification
README.md	Summarizes the delivery step and platform compatibility

Practical Installation Example

After running the cangjie-skill pipeline, install the generated skill manually or via automation:

# Build the skill (example pipeline execution)

cangjie-skill build ./my-book-source.md

# Install into Claude Code

cp -r books/my-book-title/skill-name ~/.claude/skills/

# Or for Cursor

cp -r books/my-book-title/skill-name ~/.cursor/skills/

# Verify installation

ls ~/.claude/skills/skill-name/

# Expected: SKILL.md, DIGEST.md, assets/

Once copied, launch Claude Code or Cursor. The skill will appear in the available tools list based on its description field matching your current context.

Summary

• cangjie-skill converts long-form content into AI executable skills through a multi-stage pipeline.
• Skills are constructed in a temporary books/ directory before installation into the host environment.
• The templates/INDEX.md.template provides the canonical copy command for moving skills to .claude/skills/.
• Claude Code activates skills based on the description field in SKILL.md front-matter.
• Cursor integration follows an identical pattern using its respective skills directory path.

Frequently Asked Questions

Where does cangjie-skill place generated files before installation?

The pipeline writes all artifacts to books/<book-slug>/<skill-slug>/ during the build phase. This directory contains SKILL.md, DIGEST.md, and asset subdirectories. According to methodology/07-stage5-deliver.md, Claude Code cannot load skills from this location and requires an explicit copy step.

Can I install cangjie-skill outputs directly into Cursor?

Yes. While the templates reference .claude/skills/ specifically, Cursor uses a similar directory structure (typically ~/.cursor/skills/ or project-local .cursor/skills/). Copy the generated skill folder to the appropriate Cursor skills directory to enable the same functionality.

What determines when Claude Code activates a cangjie-skill skill?

Activation depends on the description field in the skill's YAML front-matter. As defined in methodology/04-stage2-ria-plus.md, Claude Code parses this field to decide whether the skill's context matches the current conversation. A precise, keyword-rich description increases activation accuracy.

Is the installation step automated or manual?

The repository provides the installation command in templates/INDEX.md.template as a manual copy instruction (cp -r). Production deployments typically automate this step in CI/CD pipelines or local build scripts following the successful generation and testing of skill artifacts.

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