academic-aio
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is designed to ingest and process untrusted medical manuscripts, repository READMEs, and metadata files to generate automated checklists.
- Ingestion points: The bundled scripts, such as check_summary_box.py and batch_metadata_audit.py, read external files (manuscripts, CITATION.cff, and Hugging Face cards) provided by the user into the execution context.
- Boundary markers: No explicit delimiters or boundary markers are defined in the instructions to separate untrusted text from the agent's instructions during script execution.
- Capability inventory: The skill possesses filesystem capabilities (Read/Write/Edit/Glob) and can execute shell commands (Bash). The Python scripts generate reports and write files to the local qc/ directory.
- Sanitization: Processing relies on regular expressions and JSON parsing to extract specific sections. While this enforces structural integrity, it does not prevent malicious instructions embedded within the processed text from potentially influencing the agent's analysis of the document.
Audit Metadata