google-messages

Pass

Audited by Gen Agent Trust Hub on Jul 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes incoming SMS/RCS messages which are external, untrusted inputs. This creates an Indirect Prompt Injection surface (Category 8).
  • Ingestion points: sms-observer.js extracts message previews and sender information from the Google Messages DOM.
  • Boundary markers: The skill does not implement specific delimiters or 'ignore' instructions for the agent when processing message content.
  • Capability inventory: The agent has the ability to send messages, click elements, and execute JavaScript within the browser context via the browser tool.
  • Sanitization: While the sms-webhook-server.js includes a robust OTP_PATTERNS filter to prevent sensitive verification codes from being forwarded, it does not sanitize general natural language instructions that might be embedded in messages.
  • [COMMAND_EXECUTION]: The skill uses several mechanisms for local command execution.
  • Webhook Server: sms-webhook-server.js uses execFileSync to call the openclaw CLI for message forwarding.
  • Injection Scripts: scripts/inject-observer.sh uses node -e to execute dynamic JavaScript that communicates with the browser via the Chrome DevTools Protocol (CDP).
  • Polling Scripts: poll-and-forward.sh and poll-messages.sh utilize curl and jq to interact with the browser's local debug port (18800) and execute evaluations.
  • [EXTERNAL_DOWNLOADS]: The scripts/inject-observer.sh script automatically triggers npm install if the required ws dependency is missing, which involves downloading packages from the public NPM registry.
  • [DATA_EXFILTRATION]: The skill is designed to forward SMS content to external channels (e.g., Telegram). While this is the intended functionality, the data being handled (private messages) is highly sensitive. The implementation uses a local webhook (127.0.0.1:19888) by default to minimize exposure.
  • [DYNAMIC_EXECUTION]: The skill relies on injecting and evaluating JavaScript (sms-observer.js) into the browser session at runtime to monitor messages, which is a form of dynamic code execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 16, 2026, 09:42 AM
Security Audit — agent-trust-hub — google-messages