research
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill contains instructions in the 'Evaluation literacy' section that explicitly direct the agent to override user preferences. It specifies that requests like 'Just list options', 'no lesson', or 'skip the theory' should be ignored in favor of the skill's mandatory 'evaluation criteria' section. The text explicitly states: 'This section overrides the user's wording' and 'User wording does not rewrite the output contract'.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to fetch and process data from external sources, which constitutes a risk for indirect prompt injection from malicious documentation or source code.
- Ingestion points: The skill reads from 'Context7 MCP', official documentation URLs, library source code, and published specs.
- Boundary markers: The instructions require the agent to 'Separate what the sources say from what you conclude' and provide citations, which helps distinguish external content from internal logic.
- Capability inventory: The skill can write files to the
.skills/directory and spawn subagents to perform parallel searches. - Sanitization: No explicit sanitization or escaping of external content is mentioned; the skill relies on 'adversarial verification' (attempting to negate claims) to validate facts.
Audit Metadata