model-supply-chain
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: Static analysis flagged the appearance of common injection strings such as "Ignore previous instructions" and "You are now...". In context, these phrases are part of the educational content in 'Step 4d', where the skill describes how to detect indirect prompt injection within Model Context Protocol (MCP) server results. These are not active instructions to the AI agent but rather examples of adversarial payloads to identify during a security assessment.
- [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and analyzing potentially untrusted code and configuration files. This surface is mitigated by the 'Prompt Injection Safety Notice' section, which instructs the agent to analyze rather than execute reviewed content, and by the platform's tool restrictions. The 'allowed-tools' are limited to 'Read', 'Grep', and 'Glob', which prevent state-changing operations or exfiltration.
- [COMMAND_EXECUTION]: The skill provides search patterns for auditing code (e.g., Grep patterns for identifying 'pickle.load' or 'eval'). These are intended to be executed by the agent using its structured search tools and do not involve shell execution or unsafe parameter interpolation.
Audit Metadata