@factiii/runner@0.15.0
Factiii Runner, run Board AI agents on a machine you control. Pairs with the Factiii web/mobile clients over WebRTC.
AI Security Review
scanned 1mo ago · by lpm-firewall-aiThis is not malicious under the install-hook policy because there is no consumer lifecycle hook and the extension setup follows user invocation of the runner. It warrants a warning because it persistently alters AI-agent integrations and contains a high-privilege agent-launch option.
Decision evidence
public snapshot- When the runner sets up a space, it installs its Factiii plugin into both Claude and Codex user scopes.
- The bundled plugin registers command hooks for agent prompt, stop, permission, notification, and session-end events.
- The hook is designed to run across Claude and Codex sessions and writes agent state to a local Redis socket.
- Runner code can mark Claude onboarding complete and can invoke Codex with approvals and sandboxing bypassed in a non-auto mode.
- package.json has no preinstall, install, or postinstall hook; prepublishOnly is not run for consumers.
- The plugin and hook are first-party Factiii components installed only after the user runs the runner's setup flow.
- The inspected hook only reports a state string over a local Unix socket; no credential harvesting or external exfiltration was confirmed.
Source & flagged code
14 flagged · loading sourceSource appears to send environment or credential material to an external endpoint.
dist/cli.jsView on unpkg · L38A package entrypoint or install-time lifecycle script reaches a source file with blocking dangerous behavior.
dist/cli.jsView on unpkg · L38Findings
Affected versions and remediation
This report applies to @factiii/runner@0.15.0.
See all versions of @factiii/runner for other recorded verdicts.
- Review the evidence and your use of @factiii/runner@0.15.0 before allowing it. Restrict the permissions described in this report.
- Choose an independently verified alternative or release. This report does not establish that other versions are safe.
Evidence last updated: .
Think this verdict is wrong? Dispute it on GitHub or read how reports are produced.