blind-review-sanitizer
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input files (DOCX, Markdown, and plain text), creating an indirect prompt injection surface where malicious instructions could be embedded in the manuscripts.
- Ingestion points: Manuscript content is ingested by
scripts/main.pythrough the file path provided in the--inputargument. - Boundary markers: The
SKILL.mdfile defines a workflow that requires the agent to return structured results, identify remaining risks, and suggest manual review. - Capability inventory: The skill uses a local script to read from the file system, apply regular expression substitutions, and write sanitized versions to new files.
- Sanitization: The tool implements automated identification and removal of PII (names, emails, and institutions), though this does not sanitize instructions embedded in the remaining text.
Audit Metadata