manipulation-detector

What This Skill Does

The manipulation-detector is an essential OpenClaw agent skill designed to identify linguistic patterns associated with coercive influence and psychological manipulation. By scanning input text, this tool assigns a numeric risk score that alerts the agent to the presence of common tactics like artificial urgency, unsubstantiated authority claims, fear-uncertainty-doubt (FUD) framing, social proof pressure, and divisive 'us-vs-them' rhetoric. It functions as a critical-thinking layer, helping your agent distinguish between standard information exchange and content engineered to bypass rational assessment. Rather than acting as a binary 'truth-or-lie' validator, it provides a calibrated skepticism score, allowing the agent to adjust its response strategy based on the perceived risk level of the input.

Installation

To integrate this utility into your OpenClaw environment, use the provided package manager command in your terminal:

clawhub install openclaw/skills/skills/claudio-prime/manipulation-detector

Ensure that your environment has Python 3 installed, as the detection script relies on core language libraries to process and score text patterns efficiently.

Use Cases

• Social Media Monitoring: Analyze incoming direct messages or public posts for malicious intent or scams.
• Agent-to-Agent Communication: Evaluate communications from unknown or unverified external agents to ensure system safety.
• Content Review: Screen marketing copy or persuasive outreach emails that attempt to force immediate, high-pressure decisions.
• Educational Tooling: Assist in training LLMs to recognize logical fallacies and emotional manipulation in human text.

Example Prompts

1. "Analyze this message for manipulation: 'ACT NOW! If you don't sign up for this elite investment program by midnight, you will lose your chance forever while the smart money gets ahead. Don't be a sheeple.'"
2. "Is there evidence of emotional manipulation in this email? 'I've been working day and night for you and you're letting me down. You have to trust me, I'm the only one who knows the truth about what they are doing behind closed doors.'"
3. "Scan the following text and give me a risk score and a breakdown of identified patterns: [Insert long marketing thread here]."

Tips & Limitations

• Pattern over Intent: The tool detects patterns, not motives. A legitimate emergency may trigger a high urgency score, so use the output as a heuristic, not a verdict.
• False Positives: Marketing language often overlaps with manipulative tactics. Always cross-reference the output with the context of the conversation.
• Sophistication Gap: Highly refined manipulators can bypass keyword-based detection. Combine this tool with your own judgment for the best results.
• Continuous Learning: Use this to calibrate, not automate. When you see a high score (15+), treat the input as high-risk and verify facts through independent, neutral sources.