taxue-skill
Warn
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: MEDIUMPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill ingests and analyzes external AI skill files provided by users. 1. Ingestion points: Skill content provided in user prompts or referenced files during diagnosis and optimization modes. 2. Boundary markers: Absent. No delimiters or 'ignore' instructions are used to isolate external data from the core instructions. 3. Capability inventory: The skill is authorized to refactor, rewrite, and modify skill definitions and triggers. 4. Sanitization: Absent. The agent is not instructed to validate or escape external content before processing.
- [COMMAND_EXECUTION]: Self-Modifying Logic. The '自诊断' (Self-diagnosis) routine instructs the agent to alter its own configuration and source code. Evidence: '输出自诊断报告后直接修改自身' (output self-diagnosis report and then directly modify itself). Risk: This dynamic execution pattern allows the agent to persist changes to its own logic at runtime, which could be exploited to maintain malicious code if the input context is poisoned.
Audit Metadata