requesting-code-review
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to invoke sub-agents (e.g.,
code-reviewer) using thetasktool and provide them with verbatim requirement text from external files, creating a surface for embedded instructions in requirements to influence sub-agent behavior. - Ingestion points: The
PLAN_OR_REQUIREMENTSsection of thetasktool prompt described inSKILL.mdingests untrusted text from plan or requirement documents. - Boundary markers: The instructions do not define clear delimiters or "ignore embedded instructions" warnings for the sub-agent when processing the interpolated requirement text, increasing the risk of the sub-agent obeying instructions hidden within the requirements.
- Capability inventory: The skill utilizes
gitcommands (e.g.,rev-parse,log) to determine diff ranges and dispatches sub-agents with specific reviewer roles via thetasktool. - Sanitization: No sanitization, escaping, or validation of the requirement text is implemented before it is passed to the sub-agent context.
Audit Metadata