felipecustodio Overview & Purpose
Metacognitive Monitoring is "thinking about thinking." It provides real-time self-assessment of the agent's own cognitive state, identifying overconfidence, confirmation bias, or drift away from original user constraints.
When to Use
- Self-Correction Loops: Mid-task checks during long, multi-step agent workflows.
- Bias Detection: Ensuring recommendations aren't biased toward early assumptions.
Execution Workflow
- Evaluate Current Confidence Level: Rate confidence (0.0 to 1.0) in the current line of reasoning.
- Audit for Cognitive Biases:
- Confirmation Bias: Am I ignoring contradictory evidence?
- Anchoring: Am I overly fixated on the first solution considered?
- Sunk Cost: Am I continuing down a bad path just because I spent time on it?
- Assess Goal Drift: Compare current sub-task against original user prompt requirements.
- Course Correction: Adjust strategy if confidence drops or drift is detected.
Expected Output Contract
### Metacognitive Self-Audit
- **Confidence Index**: [0.0 - 1.0]
- **Detected Biases**: [None / Identified Bias]
- **Goal Alignment Check**: [On Track / Drifted]
- **Corrective Action**: [Continue / Adjust Focus / Pivot Strategy]
Scripts
scripts/metacognitive_monitoring.py- Deterministic evaluation, state validation, and CLI tool for metacognitive-monitoring.