cpp-core-guidelines-review
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The sub-agent template used for code reviews (
agents/guideline-section-reviewer.md) is susceptible to indirect prompt injection because it processes untrusted source code without sufficient isolation or instruction-data separation. - Ingestion points: Untrusted data enters the sub-agent's context through the C++ source files specified in the
{{TARGET_FILES_PATTERN}}variable. - Boundary markers: The prompt template lacks delimiters (such as XML tags or unique markers) to encapsulate the source code being analyzed, and it does not include explicit instructions to the agent to disregard any commands or system-like instructions found within the code's comments or string literals.
- Capability inventory: The agents have the capability to read reference documentation from the skill's folder and write findings to markdown files in the
review/directory. A malicious source file could potentially craft a payload that tricks the agent into writing deceptive findings or summaries. - Sanitization: There is no evidence of sanitization or filtering of the source code content before it is interpolated into the sub-agent's instructions.
Audit Metadata