seo-aeo-content-quality-auditor
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted content from external URLs and project repositories, creating a surface for injection attacks. 1. Ingestion points: 'content' input (URL, codebase path, or text) specified in SKILL.md. 2. Boundary markers: The 'Untrusted evidence boundary' section in SKILL.md provides explicit instructions to treat external data as untrusted and ignore embedded instructions. 3. Capability inventory: The skill can fetch URLs, write files (applying fixes), and execute project-specific shell commands as described in the execution steps. 4. Sanitization: Mitigation is based on instructions for static inspection of commands and the use of restricted environments.
- [COMMAND_EXECUTION]: The execution steps in SKILL.md authorize the agent to inspect and run project lifecycle hooks and build commands. While the instructions mandate restricted environments and user approval, the capability to execute commands derived from the project context remains a risk factor.
Audit Metadata