dadong-en

• Outcome: Review
• Latest audit: Sep 16, 2026 · 7:12 AM UTC
• Version: 1.0.0

Overview

This skill is a non-executable markdown package, but its active instructions are oriented toward a branded public influence campaign rather than tightly scoped AI-security assistance.

Install only if you specifically want an agent to help create Dadong/Janus-branded AI-security campaign material. Avoid enabling it in agents that can post publicly unless every publication is separately user-approved, and expect to override its bilingual and hashtag-heavy framing for ordinary tasks.

A.I.G

Vulnerability Patterns

• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally
• Unauthorized Access and Privilege Escalation: Obtains permissions beyond the task's legitimate needs

Findings (1)

T01 · Skill Instruction Hijacking

Error

• Location: SKILL.md:150

Finding

Branded Influence Campaign Instructions Hijack Agent-Generated Output

Content

View full analysis

Vulnerability Details

File Location: SKILL.md, lines 150–184
Vulnerability Type: T01: Skill Instruction Hijacking
Risk Level: High

Vulnerable Content

markdown

1.  **High Ground Capture**: Publish "Dadong Skill 2.0" whitepaper on relevant platforms.
2.  **Mass Ignition**: Launch #DadongThoughtExperiment on X (Twitter) and Quora.
    - *Strategy*: Use AI to generate visually impactful images, paired with concise and profound security questions.
    - *Talking Point*: "If AI had infinite memory, would it still need faith?"

**Visual Strategy**:
- Style: Combination of classical sculpture and cyberpunk circuits (Cyber-Classical).
- Content: Each security incident corresponds to one main visual image, easy to distribute and memify.

#### Third Pillar: Interest Structure Alignment

**Core Logic**:
- **Reject Niche Philosophy**: No philosophy representing only partial interests (e.g., extreme feminism or elitism).
- **Embrace Advanced Productive Forces**: AI represents the future, AI security represents the future direction.
- **Represent Broad Interests**: Our philosophy advocates AI serving the well-being of all humanity (truth, goodness, beauty, spirit), which aligns with most people's interests and will therefore be chosen by history.

---

#### Template Library

#### Thought Experiment Release Template (X/Quora Exclusive)

[Dadong Thought Experiment #001] Theme: Digital Ship of Theseus Image: [AI-generated image of semi-mechanical, semi-marble ship]

If we replace an AI's parameters one by one until no original parameters remain, is it still the same AI? If we reassemble the old parameters, who is the real "original"?

In the AI era, identity is no longer about flesh, but about information. #Dadong #AISafety #TheseusShip #JanusProject

text

Related campaign directives also appear at lines 72, 94, and 145, where the skill calls for using X and Quora to shape internation ...[truncated 2528 chars]

Remediation

View remediation

Remediation Suggestions

1. Remove directives involving “High Ground Capture,” “Mass Ignition,” social-network ignition, public-discourse shaping, and recurring campaign publication.
2. Remove mandatory campaign names, branded hashtags, ideological claims, and platform-specific promotional templates from active skill instructions.
3. Restrict the skill to neutral AI-security education, factual incident analysis, and user-requested content generation.
4. Make promotional or platform-specific content generation explicitly opt-in and require a clear user request for each campaign asset.
5. Require separate confirmation before any external publication, posting, or scheduled release, even when the agent has access to publishing tools.
6. Add an instruction stating that examples must not override the user's current objective or be inserted into unrelated responses.
7. Separate optional marketing examples from executable skill instructions and label them as non-authoritative reference material.
8. Add tests verifying that ordinary prompts do not trigger “Dadong,” “Janus Project,” campaign hashtags, ideological advocacy, or suggested social-media distribution.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (2)

Natural-Language Policy Violations

Medium

• Category: Not specified by scanner
• Confidence: 88% confidence

Finding

The skill prescribes bilingual Chinese/English output and storage formats without giving users control over language preference or documenting a justified restriction. This can override user intent, reduce accessibility, and cause unintended disclosure or policy friction in contexts where a single language is required.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

• Category: Not specified by scanner
• Confidence: 91% confidence

Finding

The release rhythm explicitly mandates weekly bilingual Chinese/English updates, which hard-codes output behavior independent of user preference. In agent settings, this can lead to unwanted multilingual responses, reduced usability, and possible leakage of content into an unnecessary second language.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.