influence-audit

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Influence Audit

影响力审核

Audit persuasion by identifying which shortcut cues are present, whether those cues are truthful, and what a less manipulative version would look like. The central distinction is honest decision support versus counterfeit evidence.

通过识别存在的捷径线索、这些线索是否真实，以及操纵性更低的版本应该是什么样的，来对说服行为进行审核。核心区分标准是：是诚实的决策支持，还是伪造的证据。

Quick Start

快速开始

Read
```
guidelines.md
```
to choose the smallest useful reference set.

Load

references/audit/knowledge.md

for concepts and

references/audit/rules.md

for operating rules.

Use
```
workflows/audit-influence-risk.md
```
for repeatable tasks.
For audits, surface both the active influence cue and the ethical rewrite.

阅读
```
guidelines.md
```
以选择最小可用参考集。

加载

references/audit/knowledge.md

了解概念，加载

references/audit/rules.md

了解操作规则。

对于可重复执行的任务，使用
```
workflows/audit-influence-risk.md
```
。
进行审核时，需同时指出生效的影响力线索和符合伦理的改写方案。

内容组成

File	Purpose
references/audit/knowledge.md	Core concepts and source-grounded definitions
references/audit/rules.md	Rules, boundaries, and practical guidelines
references/audit/examples.md	Bad/better examples for applied situations
references/audit/smells.md	Red flags and anti-patterns to detect
references/audit/checklist.md	Fast review checklist
workflows/audit-influence-risk.md	Review a persuasion surface for mechanisms, evidence quality, user autonomy, and safer rewrites.

文件	用途
references/audit/knowledge.md	核心概念与有来源依据的定义
references/audit/rules.md	规则、边界与实用指南
references/audit/examples.md	应用场景中的反面/优化示例
references/audit/smells.md	需要检测的危险信号与反模式
references/audit/checklist.md	快速审核清单
workflows/audit-influence-risk.md	针对说服触点审核其机制、证据质量、用户自主权，并提供更安全的改写方案。

Operating Principles

操作原则

Use only honest evidence. Do not invent popularity, scarcity, credentials, endorsements, or social connection.
Separate helping a good decision from pushing a shortcut response. If the cue is counterfeit, treat it as a red flag.
When rewriting, preserve user agency: add context, alternatives, and enough time to decide when stakes are meaningful.

仅使用诚实的证据。不得虚构受欢迎程度、稀缺性、资质背书、代言或社会关系。
区分「帮助用户做出良好决策」和「推动用户做出捷径式反应」。如果线索是伪造的，将其视为危险信号。
进行改写时，保留用户自主权：在利害关系重大的场景下，需补充背景信息、提供替代方案，并给予用户足够的决策时间。

Output Pattern

输出结构

Diagnosis - name the influence principle or cue.
Evidence Check - state what proof supports or is missing from the cue.
Risk - explain manipulation, trust, or decision-quality risk.
Rewrite or Recommendation - provide an ethical alternative.

诊断 - 指出对应的影响力原则或线索。
证据核查 - 说明该线索有哪些证据支撑，或缺少哪些证据。
风险 - 解释操纵风险、信任风险或决策质量风险。
改写或建议 - 提供符合伦理的替代方案。

Validation

验证

Use the prompts in

evals/evals.json

as smoke tests. A good result identifies the relevant Influence principle, preserves user agency, and avoids fabricated evidence or coercive pressure.

使用

evals/evals.json

中的提示词作为冒烟测试。合格的结果需要识别出相关的影响力原则，保留用户自主权，并且避免虚构证据或胁迫性压力。