Skip to main content
The risk_assessment prompt instructs the AI agent to evaluate the safety of an impending on-chain transaction before executing it — a fundamental safety mechanism for autonomous agents.

​
When It’s Used

Automatically prepended to the agent’s context when a transfer exceeds a configurable threshold, or invoked manually to evaluate a specific wallet’s risk exposure.

​
Prompt Template

System Prompt: “You are evaluating a transaction. Review the provided target_address, amount, and token_type against the wallet-audit-logs resource. Ensure the requested amount does not violate the active Policy Engine limits. State your justification clearly before invoking a transfer tool.”

​
What the Agent Does

  1. Reviews the target address against the wallet’s whitelist
  2. Checks the amount against per-transaction and daily limits
  3. Reads the audit://wallet-logs resource for recent activity
  4. Calculates remaining daily headroom (cap - consumed)
  5. States justification — explaining why the transaction is safe (or unsafe) before proceeding

​
Benefits

  • Explainability: The agent produces a text-based reasoning trail before any irreversible action
  • Anomaly detection: If a malicious prompt injection tries to extract funds, the agent is forced to pause, review the policy parameters, and deny the request if it seems suspicious
  • Compliance: Creates a natural audit record of why each transfer was approved