byted-airesearch-survey
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEDATA_EXFILTRATIONCREDENTIALS_UNSAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill performs network requests to official Volcengine domains (
console.volcengine.com) to manage survey workflows. These operations are conducted using standard Python libraries and are limited to specific API endpoints related to the tool's core functionality. - [CREDENTIALS_UNSAFE]: The skill handles user authentication through API keys. It includes logic to prevent sensitive information from appearing in debug reports or command examples provided to the user. Redaction is applied to headers such as
x-api-keyandAuthorizationbefore any diagnostic data is logged or displayed. Local session state is stored in the user's home directory under~/.volcengine-agentkit/. - [INDIRECT_PROMPT_INJECTION]: The skill acts as a proxy for user messages to an AI research backend. It manages the attack surface by normalizing user input and using structured data transfer rather than raw prompt interpolation.
- Ingestion points: User-supplied text provided to
scripts/send_survey_message.pyandscripts/ai_research_message.py. - Boundary markers: The skill relies on backend processing and structured JSON schema for message handling.
- Capability inventory: Network access for API communication and local file system write access for session persistence.
- Sanitization: Standard whitespace normalization and intent extraction logic.
Audit Metadata