Skill Recommender

Agent skill recommender. Input a user need, task description, or existing skill list; output best matching skills, install rationale, duplicate/merge candidates, and gaps for new skill briefs. Runtime-neutral and not tied to one agent platform.

Use this skill to recommend the best matching skills from an existing skill collection.

Default workflow

1. Identify the user's target intent:
   • exact function
   • adjacent use case
   • duplicate/overlap check
   • best skill recommendation
2. Scan available skills from a given directory.
3. Match by:
   • name
   • description
   • keywords
   • domain words
4. Group similar skills into clusters when needed.
5. Return a ranked recommendation with reasons.

Use the bundled scripts

1. Recommend similar skills

Use scripts/recommend_skills.js when the user gives a query such as:

• “找同类 skill”
• “有没有类似的 skill”
• “这些 skill 哪个更适合”
• “帮我筛选同功能的 skills”

Example:

bash

node scripts/recommend_skills.js '{
  "query": "家庭采购 库存 提醒",
  "skills_dir": "/Users/jianghaidong/.openclaw/workspace/skills",
  "limit": 8
}'

2. Cluster related skills

Use scripts/cluster_skills.js when the user wants a grouped view of similar skills, duplicates, or overlapping categories.

3. Avoid duplicate skill development

Use scripts/check_skill_dedup.js when the user is about to build a new skill and wants to know:

• whether similar skills already exist
• whether the request should reuse an existing skill
• whether it is better to extend or merge current skills instead of building a new one

Typical triggers:

• “这个 skill 会不会重复开发”
• “帮我查重一下 skill”
• “我想做一个新 skill，先看看有没有类似的”
• “这是不是应该复用现有 skill”

Output contract

Default output sections:

• 目标功能理解
• 推荐 skill
• 推荐理由
• 相近但次优 skill
• 可合并/避免重复开发的 skill
• 下一步建议

Recommendation rules

• Prefer concrete task fit over keyword overlap.
• Distinguish “direct match” from “adjacent match”.
• If several skills overlap heavily, say so explicitly.
• If a new skill request is already covered by existing skills, recommend reuse before new development.
• If confidence is low, surface the ambiguity.

Read references as needed

• Read references/recommendation-rules.md for ranking heuristics.
• Read references/output-schema.md for the output schema.
• Read references/dedup-mode.md when the user wants to avoid duplicate skill development.