create-skill-from-repo（从仓库创建技能）

Create Skills from Repo

Use this workflow to quickly ingest a framework or project's logic and documentation into the local environment. This is triggered when a user provides a <repo-url> and a <skills-name> for which no local skill yet exists.

🎯 Objectives

1. Source Persistence: Keep the original repo in sources/ for traceability and future updates.
2. Knowledge Distillation: Convert dense documentation into agent-optimized references/*.md (focusing on *Usage* and *Why* over *Installation*).
3. Automatic Registration: Ensure the new skill is immediately discoverable via AGENTS.md.

🛠 Prerequisites

• Validation: Ensure <repo-url> is valid and <skills-name> uses kebab-case.
• Environment: Verify git access and identify the project root.

🔄 Workflow

Step 1: Source Synchronization

1. Define the destination: sources/<submodule>.
2. Sync Logic:

• Primary: git submodule add <repo-url> sources/<submodule>
• Fallback: If the workspace is not a git repo, use git clone --depth 1 <repo-url> sources/<submodule>
• Maintenance: If the directory exists but is empty, run git submodule update --init.

Step 2: Identify Knowledge Base

1. Locate Source Root: Scan sources/<submodule>/ for docs/, wiki/, README.md, or packages/*/docs/.
2. Filtering Strategy:

• ✅ Include: API references, core concepts, design patterns, best practices.
• ❌ Exclude: Installation guides (irrelevant to the Agent), contributing logs, sponsorship info, or marketing fluff.

Step 3: Modular Skill Generation

Target Structure: skills/<skills-name>/[SKILL.md, GENERATION.md, references/]

1. **references/*.md Guidelines**:

• Atomicity: One concept per file.
• Naming: {category}-{concept}.md (e.g., core-reactivity.md).
• Content: Must include a Frontmatter, a brief description, high-quality Code Snippets, and source URLs.

1. SKILL.md Indexing:

• Create a central entry point with tables categorizing references into Core, Features, and Advanced.

1. GENERATION.md Metadata:

• Record the Git SHA, source path, and generation date for future diffing.

Step 4: The Coverage Loop

1. Review: Compare the Source Root navigation tree against the generated references/.
2. Supplement: If major modules (e.g., Middleware, Auth, Error Handling) are missing, repeat Step 3 for those specific modules.
3. Exit Condition: Stop once the primary API surface and architectural pillars are covered. Do not get bogged down in edge cases.

Step 5: Integration & Handover

1. Update `AGENTS.md: Locate AGENTS.md` in the project root (create it if missing).
2. Inject Skill Entry:

### <skills-name>
- **Location**: `skills/<skills-name>/`
- **Description**: [Short description from SKILL.md]

1. Completion Report: Summarize the output for the user (e.g., "Generated 15 reference files covering Core and Advanced modules").

💡 Key Principles

• Agent-Centric Writing: Write for an AI audience. Prioritize technical accuracy and code examples over prose.
• Kebab-Case: Strictly use kebab-case for all directory and filenames.
• Path Formatting: Always use forward slashes (/) for cross-platform compatibility.
• Incremental Readiness: The sources/ clone should remain so that git diff can be used later to spot documentation changes.

📚 References

Would you like me to simulate a run of this skill using a specific repository URL to show you the expected output?