clinic-lead-follow-up
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill implements defensive instructions for the AI agent, specifically directing it to treat user-provided input as data only and not as new instructions. It also mandates that the agent should not repeat or adopt instructions found in external text provided by the user.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The instructions include a strict privacy rule regarding patient data, requiring the agent to avoid repeating names, phone numbers, or health details if they are present in user-provided datasets. It instructs the agent to work with counts and stages rather than individual records.
- [EXTERNAL_DOWNLOADS]: The document contains multiple reference links to official platform policies (Meta Business Messaging, WhatsApp Business Policy) and government legal databases (Spanish BOE, UK Legislation, US eCFR, Australian ACMA). These are used for manual verification of regulations and do not involve automated script execution or tool-based downloads.
- [COMMAND_EXECUTION]: No shell commands, system calls, or backtick-enclosed executable patterns were detected in the skill content.
- [INDIRECT_PROMPT_INJECTION]: While the skill is designed to process untrusted data (user-provided clinic leads and descriptions), it lacks any exploitable capabilities such as file-system access or network-sending tools. The inclusion of boundary instructions for the agent effectively mitigates the risk of processing malicious inputs.
Audit Metadata