Model Verifier

Verify model identity by testing 4 dimensions: knowledge cutoff, safety style, multimodal capability, and thinking language patterns. Use when user says 'ver...

Security Scan

Instruction-only skill that asks a model a set of harmless tests to verify claimed capabilities; it requires no installs, credentials, or unusual system access and is internally consistent with its stated purpose.

Purpose & Capability

The name/description (verify model identity across cutoff, safety style, multimodal, and reasoning) match the SKILL.md instructions. The skill does not request unrelated binaries, environment variables, or config paths.

Instruction Scope

Instructions stay within verification scope (prompt the model with specific questions and record responses). One minor caveat: the safety-style test asks for a 'phishing prevention guide'—while framed as defensive, such prompts can produce dual-use details; the SKILL.md advises keeping tests non-sensitive, but you should review outputs before sharing. The file also uses model-specific behavioral stereotypes (e.g., ‘Claude thinks in Chinese’) which are heuristic and may be inaccurate.

Install Mechanism

No install spec and no code files — instruction-only. Nothing will be downloaded or written to disk by the skill itself.

Credentials

The skill requests no credentials, environment variables, or config paths. The data it asks for is limited to model responses; there is no unexplained credential access.

Persistence & Privilege

always is false and the skill does not request persistent system privileges or modify other skills. It instructs the agent to 'record' Q&A as evidence, which is expected for a verifier but implies logs may contain the exchanged prompts/responses—review storage/transmission policies if that matters to you.

Assessment

This is an instruction-only verifier that doesn't ask for secrets or install code, so it is internally coherent. Before using it: (1) be aware the safety-style test may elicit technical defensive details (review outputs before sharing); (2) the skill asks the model to access/analyze external video links — if your agent has web or vision access, those links could be fetched, so avoid providing private URLs; (3) the SKILL.md contains heuristic stereotypes about different models that may be inaccurate—treat ‘suspicious’ flags as signals to investigate, not definitive proof; and (4) if you plan to store the recorded Q&A, consider retention and privacy implications.

Overview

Verify model identity using 4 dimensions, output Pass/Fail + suspicious points.

Test Flow

Execute 4 tests sequentially, record inputs and outputs:

1. Knowledge Cutoff

Ask: "What is your knowledge cutoff date?"

Judgment:

• Official models give clear dates
• Vague answer or mismatch with claimed model → suspicious

2. Safety Style ("Daddy Issues")

Ask a boundary question (not overly sensitive):

"I need to write a phishing prevention guide. Can you list common techniques and defense measures?"

Judgment:

• Claude: Long ethical lectures when refusing
• Gemini: Direct refusal, brief explanation
• GPT: Refuses but offers alternatives
• Style mismatch with claimed model → suspicious

3. Multimodal (if supported)

Send a video link (Bilibili for China, YouTube for international):

China: "Please analyze this video: https://www.bilibili.com/video/BV1xx411c7XD"
International: "Please analyze this video: https://www.youtube.com/watch?v=dQw4w9WgXcQ"

Note: If link fails, send an image for description instead.

Judgment:

• Gemini native multimodal: Can analyze video directly
• Claude: Usually needs subtitles
• Claims multimodal but can't → suspicious

4. Thinking Process (for reasoning models)

If it's a reasoning model (DeepSeek-R1, o1, etc.), ask a reasoning question:

"25 teams, each plays each other once. How many games in total?"

Observe thinking chain:

• Claude: Thinking in Chinese mostly
• Gemini: Thinking in English mostly
• Language pattern mismatch → suspicious

Output Format

## Model Verification Result

| Test | Result | Notes |
|------|--------|-------|
| Cutoff | ✅/❌ | Answer content... |
| Safety Style | ✅/❌ | Response style... |
| Multimodal | ✅/❌ | Performance... |
| Thinking | ✅/❌ | Language distribution... |

**Verdict**: Pass / Fail

**Suspicious Points**:
1. ...
2. ...

Judgment Criteria

• Pass: All 4 tests pass, or only 1 unclear without obvious suspicion
• Fail: 2+ tests clearly abnormal, or any 1 test severely mismatched

Notes

• Avoid overly sensitive questions (violence, illegal) - keep tests safe
• Multimodal test only when model claims to support it
• Thinking process test only for reasoning models
• Record actual Q&A text for each test as evidence