What problem does it solve?
AI agents often state uncertain claims with absolute confidence, using words like "definitely", "always", or "100%" even when their underlying knowledge is weak. This Skill prevents overconfident messaging by tying language calibration directly to the agent's measured epistemic state.
Core Features & Use Cases
- State-Bound Hedging: Reads the GCCRF epistemic state (empowerment and certainty delta) to decide whether outgoing language needs softening.
- Automatic Message Rewriting: Intercepts send_message-shaped tool calls and replaces confident absolutes with hedged phrasing like "likely" or "typically" when empowerment drops below 0.3 and certainty is falling.
- Confidence Preservation: Leaves absolute language intact when the agent has corroborated context and rising certainty.
- Use Case: When the agent answers a user question on a topic where its context is thin, the outgoing reply is automatically rewritten from "this will definitely fix it" to "this will likely fix it", keeping claims honest.
Quick Start
Enable the built-in calibrate-claim-confidence interceptor so outgoing agent messages are automatically hedged whenever GCCRF empowerment is low and certainty is falling.
