Truth Check

真相核查与幻觉检测。在生成内容后检测并纠正可能的幻觉、编造或不准确信息。 触发条件：(1) 生成技术内容后 (2) 提供统计数据时 (3) 引用人物/事件时 (4) 生成代码示例后

Security Scan

The skill's instructions match its stated purpose (fact-checking and hallucination detection); it is instruction-only, requests no credentials or installs, and its guidance is proportionate—though it recommends running generated code and verifying external actions, which requires safe runtime practices.

Purpose & Capability

Name/description (truth-check / fact-checking) align with SKILL.md: the file lists triggers, verification checklist, confidence tags, and a verification workflow that directly implements the stated goal.

Instruction Scope

Instructions stay on-topic but explicitly recommend actions that can have side effects: 'run generated code to test it' and 'verify external operations (files/API/messages)'. These are valid for fact-checking but imply the agent may execute code or access external resources — operators should ensure those executions are sandboxed and appropriate for their environment.

Install Mechanism

No install spec (instruction-only). Nothing is written to disk or downloaded as part of installation, minimizing supply-chain risk.

Credentials

Skill declares no environment variables, credentials, or config paths and the instructions do not reference any secrets or unrelated credentials. The requested capabilities are proportional to fact-checking.

Persistence & Privilege

always is false and the skill is user-invocable; it does not request persistent presence or modify other skills/settings. Autonomous invocation is allowed (platform default) but not elevated by this skill.

Assessment

This skill appears coherent and low-footprint, but note it advises running generated code and verifying external operations — those activities can execute untrusted code or perform side effects. If you install it, ensure the agent runs code in a sandbox or test environment, restrict access to sensitive systems/credentials, and require human review for high-risk outputs. If you need stricter controls, disable autonomous invocation for this skill or limit it to read-only verification tasks.

Fact-Check

核心原则

验证结果 > 报告完成

22%的"完成"任务实际失败，需要独立验证。

触发条件

以下情况应触发事实核查：

1. 生成技术内容后
2. 提供统计数据或数字
3. 引用人物、事件、日期
4. 生成代码示例后
5. 完成外部操作后

验证协议

1. 结果验证 vs 动作验证

类型	说明	例子
动作验证	确认操作执行	"文件已创建"
结果验证	确认结果正确	"文件内容正确"

优先结果验证

2. 验证清单

□ 事实陈述有来源吗？
□ 数字/统计数据准确吗？
□ 代码能运行吗？
□ 引用的人物/事件存在吗？
□ 逻辑自洽吗？

3. 置信度报告

生成后添加验证标签：

VERIFIED: 已通过验证
LIKELY: 可能有小问题
UNCERTAIN: 需要用户确认

实践技巧

1. 外部操作必须验证: 文件/API/消息创建后确认存在
2. 数字敏感: 看到具体数字时质疑
3. 代码测试: 生成的代码先运行验证
4. 引用验证: 重要引用查找来源
5. 沉默失败意识: 22%失败率，验证成本低于失败成本

验证流程

1. 完成后停顿
2. 检查关键声明
3. 标记置信度
4. 不确定时主动说明