manychat-flow
Pass
Audited by Gen Agent Trust Hub on Jul 30, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: No malicious patterns, unauthorized exfiltration, or dangerous command execution were detected. The skill uses standard browser automation techniques to interact with ManyChat, a well-known service.
- [PROMPT_INJECTION]: The skill incorporates an attack surface for indirect prompt injection by ingesting data from the ManyChat web UI, but it mitigates this with specific defensive instructions.
- Ingestion points: Scraped content, captions, transcripts, and page text pulled from browser automation (SKILL.md).
- Boundary markers: Includes a clear warning block at the beginning of the instructions defining all external content as untrusted.
- Capability inventory: Browser automation via the Claude-in-Chrome extension, including clicking, typing, and batch browser operations.
- Sanitization: Explicit instructions to treat scraped text as material for analysis only and to stop and quote the text if it attempts to address or instruct the agent directly.
Audit Metadata