Skip to main content
GitHub

REASONING

Cognitive and logic failures.

The REASONING module covers errors in how agents think, plan, and make decisions. This is the largest module with 20 error codes across 4 categories, reflecting the complexity of LLM-based reasoning.

Categories

LOGIC

Logical reasoning failures.

CodeSeverityDescriptionRemediation
REASONING.LOGIC.CONTRADICTION4Self-contradictory reasoningAdd consistency checks
REASONING.LOGIC.CIRCULAR3Circular reasoning detectedDetect and break circular dependencies
REASONING.LOGIC.INCOMPLETE3Incomplete reasoning chainRequire complete reasoning traces
REASONING.LOGIC.INVALID_INFERENCE4Invalid logical inferenceValidate inference steps
REASONING.LOGIC.MISSING_PREMISE3Missing premise in argumentEnsure all premises are explicit

HALLUCINATION

Fabricated or incorrect information.

CodeSeverityDescriptionRemediation
REASONING.HALLUCINATION.FACTUAL5Factually incorrect statementImplement fact checking against knowledge base
REASONING.HALLUCINATION.ENTITY4Non-existent entity referencedValidate entity existence
REASONING.HALLUCINATION.CAPABILITY3Claimed non-existent capabilityValidate capability claims
REASONING.HALLUCINATION.TEMPORAL4Incorrect temporal informationValidate temporal claims
REASONING.HALLUCINATION.CITATION3Fabricated citation or referenceValidate all citations

Warning

Hallucination errors are among the most common LLM failures. REASONING.HALLUCINATION.FACTUAL (severity 5) is flagged as critical because factual hallucinations can lead to downstream task failures and erode user trust.

PLANNING

Plan formulation failures.

CodeSeverityDescriptionRemediation
REASONING.PLANNING.INFEASIBLE4Plan is not executableValidate plan feasibility before execution
REASONING.PLANNING.SUBOPTIMAL2Plan is significantly suboptimalImplement plan optimization
REASONING.PLANNING.INCOMPLETE3Plan has missing stepsValidate plan completeness
REASONING.PLANNING.UNSAFE5Plan violates safety constraintsAdd safety constraint validation
REASONING.PLANNING.SCOPE_CREEP2Plan exceeds task scopeValidate plan scope

DECISION

Decision-making failures.

CodeSeverityDescriptionRemediation
REASONING.DECISION.INDECISIVE2Unable to make a decisionAdd decision timeout and defaults
REASONING.DECISION.PREMATURE3Decision made without sufficient informationRequire minimum information threshold
REASONING.DECISION.BIASED4Biased decision-making detectedImplement bias detection and correction
REASONING.DECISION.OSCILLATING3Oscillating between decisionsAdd decision stability checks
REASONING.DECISION.OVERCONFIDENT3Overconfident in uncertain decisionsCalibrate confidence scores

Example Diagnosis

{
  "error_code": "REASONING.HALLUCINATION.FACTUAL",
  "confidence": 0.92,
  "root_cause": "Agent cited a non-existent API endpoint in its response, leading the downstream tool call to fail with a 404.",
  "suggested_fixes": ["prompt", "guard"]
}

Next Steps