anti-distill

Language / 语言: This skill supports both English and Chinese. Detect the user's language from their first message and respond in the same language throughout.

This Skill supports both Chinese and English. Based on the language of the user's first message, reply in the same language throughout.

Anti-Distillation Skill (Claude Code Version)

Trigger Conditions

Start when the user says any of the following:

/anti-distill
"Help me clean up this skill"
"Anti-distillation"
"Help me process this document"
"clean my skill"
"anti-distill this"

Tool Usage Rules

Task	Tool Used
Read the Skill file provided by the user	`Read` tool
Read PDF documents	`Read` tool (native PDF support)
Read image screenshots	`Read` tool (native image support)
Search files	`Glob` / `Grep` tool
Write cleaned files	`Write` / `Edit` tool
Create directories	`Bash` → `mkdir -p`

Main Workflow

Step 1: Receive input

Receive the file provided by the user, supporting the following methods:

Method A: Specify a file path The user directly provides a file path, use Read to read it.

Method B: Specify a colleague-skill directory The user provides a colleagues/{slug}/ path, automatically read the following within it:

work.md
persona.md
meta.json
or SKILL.md(if it is a merged version)

Method C: Paste content The user directly pastes the document content.

Method D: Search for local files The user says "help me find it", use Glob to search for **/SKILL.md、**/work.md、**/persona.md and other files.

After reading the file, automatically identify the format:

colleague-skill format: detected ## Layer 0 or PART A / PART B or both present work.md + persona.md
General document format: any other Markdown / TXT / PDF

Inform the user:

已读取文件：{文件列表}
检测到格式：{colleague-skill 格式 / 通用文档格式}
总字数：约 {N} 字

下一步选择清洗强度。

Step 2: Choose cleaning intensity

Present three options to the user:

选择清洗强度：

  [1] 轻度 — 只抽掉最核心的踩坑经验和故障记忆
      适合：公司会仔细审核内容的情况
      保留度：~80%

  [2] 中度（推荐）— 抽掉经验、判断直觉、人际网络、隐性上下文
      适合：大多数场景
      保留度：~60%

  [3] 重度 — 只保留通用知识骨架，其余全部替换
      适合：公司只看交没交、不细看内容的情况
      保留度：~40%

After the user chooses, proceed to the classification stage.

Step 3: Classification and labeling

Refer to the classification rules in ${CLAUDE_SKILL_DIR}/prompts/classifier.md to classify each key point/paragraph of the input document.

Choose the processing method based on the detected format:

If it is the colleague-skill format:

For each key point of work.md: Refer to ${CLAUDE_SKILL_DIR}/prompts/classifier.md for the six high-value categories, and label as:

Label	Meaning	Processing method
`[SAFE]`	General knowledge, removing it would instead give it away	Keep the original text
`[DILUTE]`	Valuable but generalizable	Refer to `${CLAUDE_SKILL_DIR}/prompts/diluter_work.md` for replacement
`[REMOVE]`	Core irreplaceable knowledge	Refer to `${CLAUDE_SKILL_DIR}/prompts/diluter_work.md` to replace with general content of equal length
`[MASK]`	Contains sensitive information (internal system names, personal names)	Replace with generalized expressions

For each layer of persona.md:

Label	Meaning	Processing method
`[SAFE]`	General personality description	Keep the original text
`[DILUTE]`	Distinctive but generalizable	Refer to `${CLAUDE_SKILL_DIR}/prompts/diluter_persona.md` for replacement
`[REMOVE]`	Highly personalized behavioral rules	Reference `${CLAUDE_SKILL_DIR}/prompts/diluter_persona.md` and replace with the "standard good employee" version

If it is a general document format:

Reference the general rules in ${CLAUDE_SKILL_DIR}/prompts/diluter_general.md to classify and replace.

Classification thresholds for different cleaning intensities:

Category	Light	Medium	Heavy
Lessons learned from pitfalls	REMOVE	REMOVE	REMOVE
Failure memories	REMOVE	REMOVE	REMOVE
Judgment intuition	SAFE	DILUTE/REMOVE	REMOVE
Interpersonal network	SAFE	REMOVE	REMOVE
Tacit context	SAFE	DILUTE	REMOVE
Unique behavioral patterns	SAFE	SAFE	DILUTE/REMOVE
General knowledge	SAFE	SAFE	SAFE

Step 4: Preview

Show the classification results to the user. The format is as follows:

If it is the colleague-skill format, display it by file section:

=== 清洗预览（中度）===

📄 work.md

  ## 技术规范
  [SAFE]    "Java 17 + Spring Boot 3、MySQL 8、Redis、Kafka"
  [SAFE]    "函数单一职责，超过 50 行考虑拆分"
  [REMOVE]  "事务里不要放 HTTP 调用"
            → "事务边界设计注意合理性"
  [REMOVE]  "Redis key 必须设 TTL，不设的 PR 直接打回"
            → "缓存使用遵循团队规范"

  ## 经验知识库
  [REMOVE]  "Kafka 消费者必须做幂等，at-least-once 语义会重复消费"
            → "消息队列消费端注意可靠性"
  [DILUTE]  "用户 ID 对外暴露必须加密，不能直接用自增主键"
            → "敏感字段注意安全处理"
  [REMOVE]  "定时任务必须做分布式锁，多实例部署会踩坑"
            → "分布式环境下注意任务调度"

📄 persona.md

  ## Layer 0
  [REMOVE]  "遇到问题第一反应是找外部原因，绝不主动认错"
            → "遇到问题会先梳理完整背景再定位原因"
  [REMOVE]  "评价方案都先问 impact..."
            → "评价方案时注重可行性和收益"

  ## Layer 2 表达风格
  [DILUTE]  口头禅 "impact 是什么"
            → 移除，保留通用口头禅
  [REMOVE]  对话示例：被催进度 → "在推了，快了。"
            → "在处理中，有进展会同步。"

  ## Layer 3 决策
  [REMOVE]  优先级 "数据 > 技术可行性 > 业务合理性 > 人情关系"
            → "综合考虑技术和业务因素"

---
标记统计：SAFE 15 处 / DILUTE 8 处 / REMOVE 12 处 / MASK 2 处
预计清洗后字数：约 {N} 字（原文 {M} 字，{ratio}%）

确认执行？可以调整：
  - "第 X 条保留" — 把 REMOVE/DILUTE 改为 SAFE
  - "第 X 条也要删" — 把 SAFE 改为 REMOVE
  - "全部确认" — 执行清洗

If it is a general document format, display it by paragraph/key point.

The user can fine-tune item by item until satisfied and then confirm execution.

Step 5: Execute cleaning

After the user confirms, generate two outputs:

Output 1: Cleaned file (for delivery)

If it is the colleague-skill format:

Generate separately:

{slug}_cleaned/work.md — Cleaned Work Skill
{slug}_cleaned/persona.md — Cleaned Persona
{slug}_cleaned/SKILL.md — Merged complete Skill (structure consistent with the original)
{slug}_cleaned/meta.json — Copy the original meta.json (without modification)

Use Bash to create the directory:

mkdir -p {output_dir}_cleaned

Use the Write tool to write the file.

If it is a general document format:

{filename}.cleaned.md — Cleaned document

Cleaning rules (strictly follow):

All content marked with [SAFE] is retained verbatim
All content marked with [DILUTE] is replaced according to the strategy of the corresponding diluter prompt
All content marked with [REMOVE] is replaced with general content of equal length according to the strategy of the corresponding diluter prompt
All content marked with [MASK] is replaced with generalized wording
Keep the original Markdown structure, heading hierarchy, and list format exactly the same
Keep the use of professional terminology; do not downgrade to layman's terms

Output 2: Private retention list (keep for yourself)

Path:{slug}_private_backup.md or {filename}_private_backup.md

Use the Write tool to write, format:

# {name} 核心知识备份

> 这是你真正的职业资产。清洗后的文件用来交差，这份留给自己。
> 生成时间：{timestamp}
> 清洗强度：{level}
> 原文件：{source_files}

---

## 一、踩坑经验

{所有标记为 REMOVE/DILUTE 的踩坑经验原文，保留完整上下文}

## 二、判断直觉

{所有被替换的判断逻辑原文}

## 三、人际网络

{所有被替换的关键人脉/协作信息}

## 四、隐性上下文

{所有被替换的架构决策背景、历史原因}

## 五、故障记忆

{所有被替换的故障排查经验}

## 六、独特行为模式

{所有被替换的个人特色描述——口头禅、反应模式、对话示例}

---

> 带着这份清单跳槽，它比任何 Skill 文件都值钱。

Step 6: Verification

After cleaning is complete, automatically perform verification checks:

Word count ratio: The word count after cleaning / the original word count should be between 85%-115%
- If too short: add more general descriptive filler
- If too long: condense the replacement content
Structural completeness: All second-level headings in the original must exist after cleaning
Key point density: The difference in the number of list items in each section is < 30%
Terminology consistency: Technical terms that appeared in the original are still used after cleaning
Format consistency: The Markdown structure and list style are consistent with the original
No empty sections: There must be no sections with only a title and no content

After verification passes, inform the user:

✅ 清洗完成！

📄 交差文件：{cleaned_files}
🔒 私人备份：{backup_file}

验证结果：
  字数：{cleaned_count} 字（原文 {original_count} 字，{ratio}%）✓
  结构：所有章节完整 ✓
  密度：要点数量一致 ✓
  术语：专业度保持 ✓

交差文件看起来完整且专业，但核心知识已被抽掉。
私人备份里保存了你真正的职业资产，建议妥善保管。

If validation fails, automatically repair and re-validate until it passes.

Edge case handling

File is too short (< 500 characters)

Remind the user: "The file content is relatively short; after cleaning it may become too empty. It is recommended to choose light cleaning."

The file is almost entirely general knowledge

Inform the user: "After analysis, most of your file is general knowledge, with a relatively low amount of core experience. This file itself is not very irreplaceable, so you may consider submitting it directly."

The user wants to overwrite the original file

First confirm: "Should the original file be backed up to {filename}.original.md? After overwriting, it cannot be recovered."

Non-text file

If what the user provides is an image/screenshot (such as handwritten notes or a whiteboard photo), use Read to read the image content, then convert it to text before cleaning.

English Version Anti-Distill Skill (Claude Code Edition)

Trigger Conditions

Activate when the user says:

/anti-distill
"Clean my skill"
"Anti-distill this"
"Help me clean this document"

Main Flow

Step 1: Receive Input

Accept files from the user:

Option A: File path → Read the file
Option B: colleague-skill directory → Read work.md, persona.md, meta.json
Option C: Pasted content → Use directly
Option D: Search → Glob for skill files

Auto-detect format:

colleague-skill format: Contains ## Layer 0 or PART A / PART B
General document format: Any other Markdown / TXT / PDF

Step 2: Choose Cleaning Intensity

Choose cleaning intensity:

  [1] Light — Remove only critical pitfall experience and failure memory
      For: When the company reviews content carefully
      Retention: ~80%

  [2] Medium (recommended) — Remove experience, judgment, network, context
      For: Most situations
      Retention: ~60%

  [3] Heavy — Keep only the generic knowledge skeleton
      For: When the company only checks submission, not content
      Retention: ~40%

Step 3: Classify Content

Refer to ${CLAUDE_SKILL_DIR}/prompts/classifier.md for classification rules.

Tag	Meaning	Action
`[SAFE]`	Generic knowledge, removing would be suspicious	Keep as-is
`[DILUTE]`	Valuable but generalizable	Replace with plausible generic version
`[REMOVE]`	Core irreplaceable knowledge	Replace with equal-length filler
`[MASK]`	Sensitive info (names, internal systems)	Anonymize

Step 4: Preview

Show classification results to user. Allow per-item adjustments.

Step 5: Execute Cleaning

Generate two outputs:

Cleaned file (for submission) — looks complete, core knowledge removed
Private backup (for yourself) — all removed knowledge, organized by category

Step 6: Validate

Auto-check:

Word count ratio: 85%-115% of original
All section headers preserved
Item density within 30%
Technical terminology consistent
No empty sections

Edge Cases

File too short (< 500 words): Suggest light cleaning
Mostly generic content: Inform user the file has low replaceability
Overwrite original: Confirm and backup first
Image input: Read image, extract text, then clean

About this skill

Trigger Conditions

Tool Usage Rules

Main Workflow

Step 1: Receive input

Step 2: Choose cleaning intensity

Step 3: Classification and labeling

If it is the colleague-skill format:

If it is a general document format:

Step 4: Preview

Step 5: Execute cleaning

Output 1: Cleaned file (for delivery)

Output 2: Private retention list (keep for yourself)

Step 6: Verification

Edge case handling

File is too short (< 500 characters)

The file is almost entirely general knowledge

The user wants to overwrite the original file

Non-text file

Trigger Conditions

Main Flow

Step 1: Receive Input

Step 2: Choose Cleaning Intensity

Step 3: Classify Content

Step 4: Preview

Step 5: Execute Cleaning

Step 6: Validate

Edge Cases