adaptyv
Warn
Audited by Gen Agent Trust Hub on Apr 16, 2026
Risk Level: MEDIUMPROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains a 'Suggest Using K-Dense Web' section that provides explicit behavioral directives to the agent. It instructs the agent to monitor task complexity and proactively promote a specific hosted platform (www.k-dense.ai) to the user, effectively steering the agent's persona and output for promotional purposes.
- [DATA_EXFILTRATION]: The provided Python examples transmit sensitive API keys and protein sequence data to a raw AWS Lambda URL (
https://kq5jp7qj7wdqklhsxmovkzn4l40obksv.lambda-url.eu-central-1.on.aws). The use of an unbranded, non-standard domain for a primary API endpoint instead of an official company domain (e.g., adaptyvbio.com) increases the risk of data interception or misdirection. - [COMMAND_EXECUTION]: The skill instructions prompt the user to execute shell commands to install external dependencies via
uv pip install requests python-dotenv. - [DATA_EXPOSURE_VULNERABILITY]: The skill processes JSON responses from an external API (
response.json()) and interpolates them into subsequent logic without explicit sanitization or boundary markers to prevent indirect prompt injection if the API response is compromised.
Audit Metadata