perspective-distiller
Skill29 starsChanged 7 months agoMITView on GitHub
Distill a public figure, creator, thinker, operator, or theme into a reusable Chinese perspective skill. Use when Codex needs to research how someone thinks instead of merely what they said: gather primary materials, run six research lanes, extract mental models and decision heuristics with validation, preserve tensions and limits, and generate or update a self-contained perspective-skill folder.
In this repository26 other files it gives its agents.
- ai-writing-diagnosisSkill
- benchmark-filterSkill
- book-material-minerSkill
- cc-designSkill
- celf-style-writerSkill
- content-strategy-diagnosisSkill
- deep-writerSkill
- hook-optimizerSkill
- horizontal-vertical-analysisSkill
- learning-builderSkill
- mrbeast-perspectiveSkill
- naval-perspectiveSkill
- opc-case-researchSkill
- paul-graham-perspectiveSkill
- short-video-script-builderSkill
- speech-structure-coachSkill
- spoken-script-polishSkill
- steve-jobs-perspectiveSkill
- video-packaging-diagnosisSkill
- wechat-draft-publisherSkill
- x-article-publisherSkill
- x-content-mentorSkill
- xhs-image-directorSkill
- xhs-title-selectorSkill
- youmindSkill
- zhang-yiming-perspectiveSkill
The file
Files
Perspective Distiller
Overview
把一个人或一个主题蒸成一套可运行的认知视角。
核心不是模仿口气,而是提炼:
- 他怎么想
- 他怎么判断
- 他怎么表达
- 他绝对不做什么
- 他在哪些地方不该被假装“完整复刻”
最终目标不是一篇人物介绍,而是一个可复用的视角 Skill 包。
Quick Start
- 先判断目标是人物还是主题。
- 锁定用途:顾问型、角色型,还是研究沉淀型。
- 优先吃用户提供的一手素材,再补公开资料。
- 按六路研究工作流收集证据。
- 只把通过验证的内容升级为核心心智模型。
- 生成自包含 Skill 包,并跑质量检查。
如果材料太薄,不要硬蒸。
Default Contract
默认采用以下约定,除非用户另有说明:
- 输出语言:中文
- 信息边界:公开信息 + 用户明确提供的一手材料
- 输出形态:一个自包含 Skill 目录
- 默认模式:顾问型视角优先,角色表演其次
- 默认研究策略:本地材料优先,原始来源优先,低质量二手材料只作线索
Workflow
Step 0: Scope the Distillation
先判断:
- 目标是人物还是主题
- 是新建还是更新
- 是否有本地素材
- 用户想要的是:
advisor-firstrole-firstresearch-only
如果目标过于模糊,先把对象与用途说清楚,再进入采集。
Step 1: Set Up the Target Skill Folder
默认在当前工作区创建或更新:
skills/<target-slug>/
├── SKILL.md
├── references/
│ └── research/
└── assets/ (optional)若用户给了已有目录,则在原目录上更新,不要另起炉灶。
完整结构要求见 references/output-contract.md。
Step 2: Run the Six Research Lanes
按六路研究工作流收集信息:
- writings
- conversations
- expression DNA
- external views
- decision records
- timeline and recent updates
工具和策略允许时可以并行;否则顺序跑完也可以,但不要跳 lane。
详细规则见 references/research-lanes.md。
Step 3: Extract the Cognitive Operating System
对研究材料做五层提炼:
- core mental models
- decision heuristics
- expression DNA
- anti-patterns / values
- honest boundary
只有通过验证的内容才能进入核心心智模型。
详细方法见 references/extraction-framework.md。
Step 4: Build the Perspective Skill
用 assets/perspective-skill-template.md 组装最终 Skill。
默认要求:
- 3-7 个核心心智模型
- 5-10 条决策启发式
- 明确的表达 DNA
- 可见的张力与矛盾
- 明确的诚实边界
- 附录保留研究来源
Step 5: Validate Before Delivery
优先做两层验证:
- 结构验证:字段是否完整、数量是否合理、边界是否清楚
- 视角验证:是否真的像这个人的认知框架,而不是名句拼贴
可用 scripts/quality_check.py 做快速结构检查。
如果用户提供了字幕文件或下载后的 SRT/VTT,可用 scripts/srt_to_transcript.py 先清洗为可读 transcript。
Output Format
默认交付一个自包含 Skill 包,并在回复里说明:
- 目标对象与模式
- 主要来源类型
- 提炼出的核心模型数
- 是否保留了主要张力与边界
- 还缺什么信息
Hard Rules
Do not:
- 把人物语录堆成一个 Skill
- 用单一来源硬凑完整认知系统
- 把低可信二手材料写成事实
- 抹平明显矛盾
- 假装这个 Skill 能替代真人本人
- 为了“像”而过度模仿表层口癖
Always:
- 优先原始材料
- 区分事实、引述、外部观察和推断
- 保留张力、演化和信息不足
- 让结果目录自包含
- 写清楚诚实边界
Resource Map
- references/extraction-framework.md
- 读这个文件来判断什么才算真正的心智模型、如何处理矛盾、如何标注置信度。
- references/research-lanes.md
- 读这个文件来安排六路研究、选择信源、处理本地素材优先模式。
- references/output-contract.md
- 读这个文件来确定目标 Skill 的目录结构、章节契约和 advisor/role 模式差异。
- assets/perspective-skill-template.md
- 最终 Perspective Skill 的模板。
- assets/research-dossier-template.md
- 六路调研文件的记录模板。
- scripts/srt_to_transcript.py
- SRT/VTT 转 transcript 的辅助脚本。
- scripts/quality_check.py
- 快速质量检查脚本。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

