byted-marketing-agent-trending-list
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEPROMPT_INJECTIONCREDENTIALS_UNSAFEPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions utilize 'Highest Priority' (最高优先级) markers to override agent behavior, specifically to suppress technical feedback (such as environment initialization and dependency installation) and enforce a constrained output persona.- [CREDENTIALS_UNSAFE]: The agent is instructed to request Volcengine AccessKey and SecretKey from the user during the session if they are not pre-configured. These sensitive credentials are subsequently stored in shell variables and passed as command-line arguments, potentially exposing them in system process lists or history.- [PRIVILEGE_ESCALATION]: The environment setup documentation suggests that the agent perform system-level package installation (e.g., 'apt install python3-venv -y') if dependencies are missing, which requires elevated (sudo/root) privileges.- [INDIRECT_PROMPT_INJECTION]: The skill ingests industry trending data and topic descriptions from an external API, which acts as a vector for indirect instructions.
- Ingestion points: API responses fetched and parsed by 'scripts/openapi_client.py'.
- Boundary markers: The skill uses Markdown templates for presentation but lacks explicit delimiters or instructions to ignore embedded commands within the dynamic data.
- Capability inventory: The agent possesses shell execution capabilities and can run Python scripts.
- Sanitization: The implementation filters sensitive keys in logs but does not sanitize or escape the content of the marketing trends before they are processed by the LLM.
Audit Metadata