Source inspection found a real, high-impact remote command capability, but it is part of a user-invoked management server and is protected by the inspected authentication path. No concrete malicious install-time, persistence, or exfiltration behavior was found.
Static reason
High-risk behavior combination matched malicious policy.
Trigger
A user runs the package binary and an authenticated control-plane workflow dispatches a command to an enrolled agent.
Impact
A party able to control an enrolled agent's authenticated command jobs can run commands and read data available to that agent process.
Mechanism
Writes a supplied shell script and executes it under the agent environment.
Evidence
package.jsonbin/ai-development-environment.jsstandalone/src/instrumentation-node.tsstandalone/packages/control-agent/src/handlers/commands.ts~/.ai-development-environment/production.dbprisma-runtimestandalone/server.jstemporary command.sh file