Developer Proposes Network Monitoring to Verify AI Coding Agent Actions

A developer building Rashomon, a tool that records and verifies AI coding agent execution, is exploring the need for network-level visibility. They argue that simply verifying commands run by an agent is insufficient, as processes can contact external services without being documented. For example, a routine 'npm install' command may contact package registries or download files not detailed in an agent's summary. This additional layer of evidence is considered crucial as coding agents become more autonomous and perform complex tasks like deploying changes or calling APIs. The goal is to create independent verification by combining execution records with network activity logs.