parse-reviews
Pass
Audited by Gen Agent Trust Hub on May 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to executelatexmk, which compiles a generated LaTeX document containing transcribed reviewer comments. - [EXTERNAL_DOWNLOADS]: The skill connects to the Elsevier Serial Title API to retrieve academic journal metrics, which requires a user-provided
SCOPUS_API_KEY. - [PROMPT_INJECTION]: The skill ingests untrusted reviewer comments from PDFs, creating an indirect prompt injection surface. 1. Ingestion points: Reviewer reports are read in Phase 2 using the
Readtool. 2. Boundary markers: The instructions mandate verbatim transcription ("Verbatim means verbatim"), which prevents the agent from filtering or escaping malicious LaTeX macros. 3. Capability inventory: The agent can write.texfiles and execute thelatexmkcommand in Phase 4. 4. Sanitization: No escaping or sanitization of LaTeX-sensitive control characters is performed on the extracted text before compilation.
Audit Metadata