authoring-signals-scouts
Pass
Audited by Gen Agent Trust Hub on Jul 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documentation includes a dedicated safety section addressing the risk of indirect prompt injection for agents that process external or untrusted data. It provides specific remediation strategies, such as treating ingested content as data rather than instructions, sanitizing personal information (PII), and employing "quote, don't act" patterns to prevent malicious steering of agent behavior.
- [COMMAND_EXECUTION]: The instructions describe scout architectures that utilize standard shell utilities, including
git,curl,npx,node,python, andrg, within a sandboxed environment. These tools are documented as part of the intended functionality for code analysis and data fetching from public repositories. - [EXTERNAL_DOWNLOADS]: The documentation outlines scout patterns involving data retrieval from external sources, such as cloning public GitHub repositories and interacting with HTTP APIs. The skill includes guidance on managing these operations safely, focusing on rate limits and verification of tool availability within the execution environment.
Audit Metadata