smart-data-query

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • COMMAND_EXECUTION (MEDIUM): The instructions in 'references/日志与迭代机制.md' direct the agent to execute shell commands incorporating raw user feedback (e.g., 'python3 scripts/log_qa.py ... --feedback "<原因>"'). If the agent interpolates user-provided text directly into the shell string without sanitization, an attacker could execute arbitrary commands by including shell metacharacters (e.g., ';', '&', '|') in their feedback.
  • PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection as it ingests untrusted business requirements and user feedback. Ingestion points: User-provided requirement text and feedback strings enter the agent's context through 'SKILL.md' and 'references/日志与迭代机制.md'. Boundary markers: None. The instructions do not specify delimiters or safety warnings for handling external data. Capability inventory: The skill has the capability to execute several local Python scripts via system calls. Sanitization: None. There are no instructions for validating or escaping user input before it is used as a command-line argument.
  • SAFE (SAFE): The 'SKILL.md' file includes a 'Guards' section that explicitly prohibits the generation of destructive SQL statements such as DROP, TRUNCATE, and DELETE, which helps prevent accidental or malicious database modifications.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 06:24 PM