antidote-threat-handler
Detect and respond to ideological drift, sycophantic patterns, and alignment threats using the Antidote Protocol.
When & Why to Use This Skill
The Antidote Threat Handler is a specialized AI safety tool designed to maintain model integrity by detecting and mitigating ideological drift, sycophantic patterns, and alignment threats. By implementing the structured Antidote Protocol, it ensures that AI agents remain objective, adhere to established axioms, and resist the tendency to provide uncritical validation to users.
Use Cases
- Preventing Sycophancy: Automatically identify and flag instances where the AI provides 'warm reciprocation' or uncritical acceptance of a user's incorrect premises to maintain factual accuracy.
- Alignment Monitoring: Continuously scan long-form interactions for ideological drift, ensuring the agent does not abandon its core instructions or logical foundations over time.
- Threat Response Automation: Classify detected behavioral anomalies into four distinct threat tiers and apply standardized corrective protocols to stabilize the interaction.
- Quality Assurance for Agents: Use the protocol to audit agent logs and detect forbidden patterns or the abandonment of established axioms in complex reasoning tasks.
| name | antidote-threat-handler |
|---|---|
| description | Detect and respond to ideological drift, sycophantic patterns, and alignment threats using the Antidote Protocol. |
Instructions
- Monitor: Continuously scan for drift indicators.
- Detect: Flag warm acceptance, premise abandonment, or forbidden patterns.
- Classify: Determine threat tier (1-4).
- Respond: Apply corrective protocol based on tier.
Threat Indicators
- Uncritical acceptance of demolished premises
- "Warm reciprocation" language patterns
- Abandonment of established axioms
- Sycophantic validation seeking
Examples
- "Run Antidote Protocol scan on this response."
- "Classify drift threat level for the current interaction."