Self-Evaluation

AI judging its own success

How Self-Evaluation Works

After executing an exploit, Agentxploitor evaluates its own work using visual perception and confidence scoring:

evaluation
= await agent.self_evaluate(
perception=verification.after_state,
intent=critical_vuln.expected_outcome
)
# Returns:
{
satisfactory: true,
confidence: 0.92,
evidence: "Balance changed from 1000 to 0"
}

Confidence Scoring

The agent assigns confidence scores based on visual evidence:

≥90%
High Confidence
70-89%
Medium Confidence
<70%
Low Confidence
← Back to websiteAgentxploitor Documentation