sciwrite-manuscript
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests text from research documents provided by the user (e.g., .tex, .docx, .md files) to perform structural reviews. This creates a surface for indirect prompt injection, where instructions hidden within a paper could attempt to bypass agent constraints or bias the review.\n
- Ingestion points: User files are loaded into the agent's context using the
extract_text.pyscript.\n - Boundary markers: No specific delimiters or safety warnings are implemented to separate ingested document text from agent instructions.\n
- Capability inventory: The skill has the ability to run shell commands and present processed text to the user.\n
- Sanitization: No sanitization of document content is described in the provided instructions.\n- [COMMAND_EXECUTION]: The skill executes local Python scripts using shell commands to extract content and check for numerical or keyword consistency.\n
- Evidence: Commands such as
python "$SCIWRITE/scripts/extract_text.py" paper.texare used in the analysis workflow.\n - Path Resolution: The location of helper scripts is resolved dynamically based on environment variables or relative directory searches to find the skill root.
Audit Metadata