External Skill Adapter

Use this skill to turn an outside skill into a local 目标 Agent skill without copying it blindly. The goal is to preserve the useful job-to-be-done, rewrite the workflow for 目标 Agent, add boundaries, add testable outputs, and leave a shareable package if the user wants to publish the process.

Adaptation Workflow

1. Capture the source
   • Save the source URL, page HTML, README, screenshots, or pasted skill text.
   • Extract only facts needed for adaptation: name, intended users, inputs, outputs, commands, platform scope, dependencies, risks, and missing implementation details.
   • If the source is a webpage, keep a local evidence copy in the project work/ folder.
2. Separate intent from implementation
   • Intent: what work the skill helps users finish.
   • Implementation: command wording, platform claims, dependencies, scripts, and examples.
   • Preserve intent when useful. Rewrite implementation for 目标 Agent, local files, available tools, and the user's workflow.
3. Decide the 目标 Agent shape
   • Choose one narrow entrypoint before writing files.
   • Prefer a result-oriented 目标 Agent workflow over broad marketplace claims.
   • Add scripts only for repeated deterministic work, such as CSV normalization, report generation, validation, or packaging.
   • Put longer rubrics, templates, and checklists in references/.
4. Change the skill enough
   • Do not reuse the source SKILL.md structure verbatim.
   • Rename the skill if the new scope is different.
   • Replace generic feature lists with a concrete router, output contract, quality bar, and failure handling.
   • Remove unsupported claims such as real-time monitoring, alerts, platform coverage, scraping, or export formats unless the local skill actually implements them.
5. Add boundaries
   • Say what the skill will not do.
   • Avoid credentials, hidden browser profiles, CAPTCHA bypass, private data, or background monitoring unless the user explicitly owns and authorizes that system.
   • Label public-page samples, user-provided exports, screenshots, and estimates honestly.
6. Create the local skill
   • Put the skill under ~/.agents/skills/<skill-name>/ unless the user asks for another location.
   • Include SKILL.md with 目标 Agent-adapted frontmatter.
   • Include README.md with usage instructions.
   • Include only useful scripts/, references/, or assets/.
7. Test
   • Validate the skill structure:

     ls -la ~/.agents/skills/<skill-name>/
     cat ~/.agents/skills/<skill-name>/SKILL.md | head -20

   • Run at least one smoke test with a realistic sample input if the skill includes a script or output format.
   • Record test command, result, and generated artifacts in the share package, not inside SKILL.md.
8. Package for handoff or article
   • Keep install path, test command, source evidence path, adapted differences, and known limits.
   • For public writing, lead with the result. Put implementation notes, blocked attempts, and regeneration commands in README or handoff files.

Adaptation Checklist

Read references/adaptation-checklist.md when the source skill is vague, overbroad, or likely copied from a marketplace page.

Before calling the adaptation finished, confirm:

• The new skill has a different name or clearly narrower scope.
• The trigger description says when to use it.
• The workflow has input routing, output contract, quality bar, and boundaries.
• Unsupported source claims were removed or downgraded.
• Validation passed.
• At least one realistic test or dry run exists.
• User-facing article material is separate from reusable skill instructions.

Output Summary Format

When reporting back, use:

已适配：
- 新 skill：<path>
- 来源证据：<path or URL>
- 改动重点：<3-5 bullets>
- 测试：<command + pass/fail>
- 分享包：<path, if any>

Common Adaptation Patterns

Pattern 1: Web Search Skills

• Replace API calls with WebSearch tool usage
• Add fallback instructions when search fails
• Specify topic parameter for better results

Pattern 2: File Operation Skills

• Map file reads to Read tool
• Map file writes to Write tool
• Add path validation and error handling

Pattern 3: Code Execution Skills

• Map commands to Bash tool execution
• Add timeout considerations
• Specify working directory requirements

Pattern 4: Image Processing Skills

• Enable Read tool's image mode capability
• Add image format support documentation
• Specify output format options

Anti-Patterns to Avoid

• ❌ Copying original SKILL.md without modification
• ❌ Keeping platform-specific paths (e.g., Codex-specific)
• ❌ Removing boundaries and safety checks
• ❌ Adding unsupported capability claims
• ❌ Forgetting to update frontmatter metadata