figure-rhetoric
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret data from untrusted external sources (manuscripts), which introduces a susceptibility to indirect prompt injection.
- Ingestion points: The audit process involves reading compiled PDFs, figure image files (.png, .pdf, .jpg), and LaTeX source text to identify visual takeaways and prose claims.
- Boundary markers: The instructions do not specify any markers or system directives to isolate or ignore potentially malicious instructions embedded within the manuscript text or captions.
- Capability inventory: The skill utilizes file-reading capabilities to inspect documents. It does not explicitly define or call for network access, subprocess execution, or file-writing tools.
- Sanitization: There are no instructions for sanitizing, escaping, or validating the content extracted from processed documents before it is used in the report prompts.
- [NO_CODE]: The skill is composed strictly of markdown-based instructions and YAML configuration files, containing no scripts, builds, or binary executables that could lead to direct code execution.
Audit Metadata