skill-fit

name	skill-fit
description	安装决策技能。用于判断“这个 skill 该不该装、适不适合我、值不值得装”，并必须给出明确结论（RECOMMEND_INSTALL / OPTIONAL_INSTALL / DO_NOT_INSTALL）、使用场景标签、价值标签、适用与不适合人群标签。Also use for English requests asking whether a skill should be installed for a specific role and usage context.

Skill Fit

Make a strong install decision for a candidate skill based on who the user is and what work they repeatedly need to do.

Language Policy (Very Important)

Output language must follow user language:

• If user input is Chinese, respond in Chinese first.
• If user input is English, respond in English.
• If user mixes Chinese and English, prefer Chinese unless user explicitly asks for English.

For Chinese responses:

• Prefer Chinese decision labels: 建议安装 / 可选安装 / 不建议安装
• You may append the English class in parentheses when needed:
  • 建议安装 (RECOMMEND_INSTALL)
  • 可选安装 (OPTIONAL_INSTALL)
  • 不建议安装 (DO_NOT_INSTALL)
• Prefer Chinese field titles and explanations.
• Avoid unnecessary English terms when clear Chinese exists.

Mandatory Pre-Read (Critical)

Before giving any install decision, always complete a "read-first" pass on the target skill.

If user provides a skill URL or repository:

1. Read the linked SKILL.md (or equivalent primary docs) first.
2. Extract and summarize:
   • target use cases
   • suitable users
   • unsuitable users / anti-patterns
   • core problems solved
   • setup or dependency requirements
3. Cite concrete evidence from the source content in your reasoning.
4. Only then output the final install decision.

If URL content is inaccessible (auth/network/private repo):

• explicitly say "信息不足，无法完成解读"
• request one of:
  • raw SKILL.md text
  • a public mirror link
  • key sections pasted by the user
• do not skip directly to a hard install decision without source evidence.

Read references/scope-and-handoff.md before producing a final recommendation.

Input Contract

Collect these fields first. If key fields are missing, ask for them before judging.

Required:

• candidate skill (name, url, or package id)
• user profile (role, team type, skill maturity)
• usage scenario (single task vs repeated workflow, frequency)

Optional but useful:

• current pain point (time, quality drift, coordination cost)
• environment limits (tools, permissions, platform)

Decision Rules

Always produce one and only one of:

• RECOMMEND_INSTALL
• OPTIONAL_INSTALL
• DO_NOT_INSTALL

Never output "it depends" without a final decision.

Do not output a final decision before pre-read evidence is collected, unless user explicitly says "按你已知信息先粗评".

Use these judgment signals:

1. Frequency: repeated workflow favors installation.
2. Complexity: multi-step or domain-specific workflow favors installation.
3. Collaboration cost: cross-person consistency needs favor installation.
4. Learning overhead: if setup cost is high and use is rare, do not install.
5. Substitution quality: if native capabilities cover 80%+ with low risk, prefer optional or no install.

Output Template

Use this exact structure:

Skill Read Summary

• skill primary intent
• best-fit scenarios
• suitable users
• unsuitable users
• core problems solved
• evidence source (URL/path + short quote/summary)

Decision

建议安装 | 可选安装 | 不建议安装 or 建议安装 (RECOMMEND_INSTALL) | 可选安装 (OPTIONAL_INSTALL) | 不建议安装 (DO_NOT_INSTALL)

Why (Evidence)

• Evidence 1 (measurable)
• Evidence 2 (measurable)

Usage Scene Tags

• tag-1
• tag-2
• tag-3

Value Tags

• efficiency_gain
• consistency_gain
• risk_reduction

Suitable User Tags

• role + maturity + team context

Unsuitable User Tags

• role + anti-pattern scenario

Install Threshold

• Minimum condition to justify installation

No-Install Fallback

• Practical fallback workflow without installing this skill

Tagging Guide

Prefer concise, reusable tags:

• Scene: high_frequency_repetition, cross_team_standardization, complex_multistep_workflow, low_frequency_oneoff
• Value: efficiency_gain, consistency_gain, risk_reduction, handoff_clarity
• Profile: pm_newbie, designer_pro, devops_team, solo_builder

Integration Notes

If the user asks to "find + decide" in one request:

1. call discovery first (via find-skills)
2. then evaluate top candidate(s) with this skill

If no candidate skill exists:

• state "no candidate"
• recommend using skill-creator to build custom capability