bumble-automation

Fail

Audited by Gen Agent Trust Hub on Jun 21, 2026

Risk Level: HIGHCREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: Hardcoded Anchor Browser API key ("sk-77fea665...") and multiple Bumble session cookies (including "session" and "HDR-X-User-id") are exposed in the skill markdown body.
  • [COMMAND_EXECUTION]: The skill provides a complete Node.js script ("like140.mjs") and instructs the agent to execute it using the bash tool.
  • [DATA_EXFILTRATION]: The provided automation script reads and logs private conversation history from the Bumble platform, exposing sensitive user data to the agent context.
  • [REMOTE_CODE_EXECUTION]: The script uses Playwright to connect to a remote Chrome DevTools Protocol (CDP) endpoint provided by the third-party service "anchorbrowser.io".
  • [EXTERNAL_DOWNLOADS]: The instructions suggest the global installation of the "playwright-core" package via npm if it is missing from the host environment.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection by reading raw user messages from Bumble to generate contextual replies without sanitization.
  • Ingestion points: Chat message history is ingested via the "openAndReply" function in the provided script.
  • Boundary markers: None provided to separate user-generated content from agent instructions.
  • Capability inventory: The skill has access to the "bash" tool and "desktop-commander" for system-level operations.
  • Sanitization: No filtering or escaping is applied to the retrieved messages before they are processed by the LLM.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Jun 21, 2026, 04:01 PM