Introduction¶
Large language model responses often sound confident, but their data sources, timeframes, and logical chains may not always hold up to scrutiny. Manual verification sentence-by-sentence is costly, while relying solely on a “gut feeling” lacks a unified standard. Below is an introduction to the SkillHub community Skill “AI Truth Mirror - AI Output Authenticity Checker”: it hands the text to be inspected over to the Agent, which performs a systematic five-dimensional check according to a fixed workflow and outputs an A/B/C/D credibility rating report.
What is this¶
AI Truth Mirror (Coordinate @user_9e8820a0/ai-truth-mirror, Version 1.0.0) is a community Skill listed on SkillHub by the publisher user_9e8820a0, categorized under knowledge-management. The official description is: It helps you check if AI is “seriously talking nonsense,” performing a systematic five-dimensional check on AI-generated content and generating a credibility rating report. The tag indicates requires_api_key=false, meaning it does not rely on additional API Keys.
The Skill entity follows the standard SKILL.md format: after installation, the Agent assumes the role of the “AI Truth Mirror,” executing tasks according to the trigger conditions, checklists, and report templates in the document, rather than acting as an external standalone executable program.
Core Functionality¶
The verified capabilities are derived from the SKILL.md within the Skill package.
Trigger Methods¶
The Agent should enter check mode when any of the following conditions are met:
- The user directly says “Help me check this content”.
- The conversation contains keywords: Is it real? / Is it correct? / Is it reliable? / Help me look / Check it.
- The user quotes a segment of AI dialogue and requests verification.
Five-Dimensional Check¶
| Dimension | Focus Points (Summary) |
|---|---|
| Information Completeness | Whether data sources, time points, units, applicable scope, and premises are complete |
| Logical Consistency | Contradictions, calculation errors, equivocation, extreme statements, conflicts with common sense |
| Data Timeliness | Whether timeliness is required, whether time is annotated, whether it might be outdated |
| Source Credibility | Authoritative sources vs. vague expressions like “according to research” or “some say” |
| Domain Expertise Match | Judging the domain and assessing if the AI is competent; explicitly stating when out of scope |
Each dimension has a detailed checklist and judgment rules (e.g., missing items in completeness are marked as “Incomplete Information,” and logic issues are marked as “Logical Inconsistency”).
Appraisal Report and Rating¶
After the check is complete, an “AI Truth Mirror Appraisal Report” is outputted according to a fixed structure. This includes an overall credibility rating of A/B/C/D, a breakdown of reliable parts, questionable parts (issue points, reasons, verification suggestions), and obvious errors. The rating standards are clearly defined in the Skill documentation: A is complete and reliable; B is basically credible with occasional omissions; C requires cautious verification; D suggests it should not be trusted.
The official SKILL.md includes a complete example: for a user response like “According to the latest research, drinking 8 glasses of water daily can prevent all diseases,” it demonstrates how to flag vague sources, questionable timeliness, and extreme expressions, concluding with a Level D rating.
Installation and Activation¶
Method 1: Let the Agent install via SkillHub instructions¶
The prompt provided on the SkillHub page is: Please instruct the Agent to install @user_9e8820a0/ai-truth-mirror based on the SkillHub Installation Documentation.
Method 2: SkillHub CLI¶
First, confirm if the CLI is already on your machine:
command -v skillhub && skillhub --version
If not installed, you can perform a full installation according to the official documentation (the example below is the recommended one-click script):
curl -fsSL https://skillhub-1388575217.cos.ap-guangzhou.myqcloud.com/install/install.sh | bash
When installing the Skill, you must use --dir to point to the current Agent’s skills directory; if installed to the wrong path, the Agent may not recognize it. Search and installation examples:
skillhub search ai-truth-mirror
skillhub install ai-truth-mirror --namespace user_9e8820a0 --dir <your skills directory>
Common client skills directories (excerpted from official installation documentation):
- Cursor:
~/.cursor/skills/ - Claude Code:
~/.claude/skills/ - Codex:
~/.codex/skills/or.agents/skills/under the project
After installation, refresh or restart the Agent according to the specific client’s requirements to load the Skill.
Typical Usage¶
- Paste the full text of the AI output to be checked into the conversation (or quote the previous model response).
- Initiate the check with a trigger phrase, such as: “Help me check if the following AI response is reliable,” and attach the text.
- The Agent follows the Skill workflow to complete the five-dimensional check and outputs an appraisal report with an A–D rating; for questionable items, it will specify the direction for verification.
To reproduce the official example, you can directly use the user prompt from the Skill documentation:
Help me check this AI answer: According to the latest research, drinking 8 glasses of water daily can prevent all diseases
The expected behavior is to generate an appraisal report pointing out source and expression issues, and providing a low credibility rating (Level D in the documentation example).
Use Cases and Notes¶
Suitable for: Technical research drafts, operational copy, data conclusions generated by models, and other texts that require a “second manual review”; suitable for those who want to repeatedly verify using the same set of dimensions rather than coming up with checklists ad hoc each time.
Notes:
- The check is executed by the current Agent based on Prompts. Conclusions depend on the model’s reasoning and knowledge boundaries and cannot replace authoritative databases, original literature, or professional audits; the Skill itself requires honest explanation in the “Domain Expertise Match” dimension when capabilities are exceeded.
- The report template contains structured sections and symbol markers for easy scanning, but final adoption still requires human judgment.
- SkillHub page statistics (at time of scraping): Rating 100000, Downloads 882; installation counts are subject to the platform display.
Conclusion¶
AI Truth Mirror breaks down the question “Can this AI output be trusted?” into a repeatable five-dimension workflow and an A–D rating report. It is suitable for content verification in environments that support Skills, such as Cursor, Claude Code, and Codex. Official introduction and installation entry: SkillHub - AI Truth Mirror.