Socratic Audit

A Hermes/Claude-style skill for auditing fluent but uncertain reasoning.

Socratic Audit helps an AI assistant stop treating beautiful language as truth. It applies Socratic interrogation to AI answers, management arguments, technical proposals, vendor claims, strategy decks, investment theses, and philosophical or aesthetic judgments.

A beautiful argument is not evidence that it is true.

What it does

The skill forces a four-layer audit:

1. Facts vs views — separate verifiable claims from interpretations and value judgments.
2. Argument skeleton — strip rhetoric and expose hidden premises.
3. Counterargument and boundary test — seek the strongest opposing case and failure conditions.
4. Judgment ownership — return the final decision to the human instead of letting AI sign it.

Why it exists

Generative AI is very good at producing fluent, balanced, plausible explanations. That is useful, but dangerous: a complete-sounding answer can arrive before the human has actually thought.

This skill treats AI as a sparring partner, not an oracle.

Example use cases

• “Is this AI answer reliable?”
• “Does this architecture proposal actually make sense?”
• “Is this vendor claim just marketing?”
• “Is this management slogan backed by real mechanisms?”
• “Can this investment thesis survive a strongest-counterargument test?”
• “Is this essay just elegant rhetoric?”

Installation

Clone the repository and symlink it into your Hermes skills directory:

git clone https://github.com/zero-click/socratic-audit-skill.git ~/code/socratic-audit-skill
mkdir -p ~/.hermes/skills/critical-thinking
ln -sfn ~/code/socratic-audit-skill ~/.hermes/skills/critical-thinking/socratic-audit

Then start a new Hermes session so the skill loader can see it.

Files

SKILL.md                              # Main skill
references/probability-rhetoric.md    # Supporting reference
scripts/validate.py                   # Local validation script

Validation

python3 scripts/validate.py

The validator checks:

• required files exist;
• YAML frontmatter is valid;
• skill name is socratic-audit;
• description is under 1024 characters;
• public repo privacy scan passes;
• required Socratic audit concepts are present.

中文说明

Socratic Audit 是一个用于“防止被顺滑语言骗过”的 AI skill。

它适用于审视：

• AI 生成的回答；
• 管理层话术；
• 技术架构方案；
• vendor/consultant 的说法；
• 投资观点；
• 哲学、艺术、价值判断类文章。

核心四问：

1. 哪些是事实，哪些只是看法？
2. 它默认了什么前提？
3. 最强反方是什么？
4. 抛开漂亮文字，我自己真的认可并愿意承担吗？

它的目标不是让 AI 显得更聪明，而是让 AI 接受追问。

License

MIT