skill-forge-base
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONOBFUSCATIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from quiz questions and user queries by interpolating them directly into AI prompt templates. Ingestion occurs in lib/server.js via the /api/ask-ai and /api/analyze-question API endpoints. Boundary markers consist of basic Markdown headers (e.g., '## 测验信息'), which are not resistant to adversarial prompt injection. Capability inventory includes the ability to execute shell commands via child_process.spawn and perform local file operations. Sanitization is absent as input is used directly in prompt strings.
- [OBFUSCATION]: The file tmp_old_server.js is encoded in UTF-16 Little Endian, which obfuscates its logic from tools expecting standard UTF-8 or ASCII character sets. Analysis confirms it is a legacy artifact of the main server script.
- [DYNAMIC_EXECUTION]: The server script (lib/server.js) dynamically executes shell commands using child_process.spawn with the shell: true option. While this is used for legitimate AI CLI integration, the reliance on shell execution increases the risk of command injection if the local configuration file is compromised.
Audit Metadata