scholar-evaluation
Warn
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONNO_CODE
Full Analysis
- [COMMAND_EXECUTION]: The documentation in
SKILL.mdinstructs the agent to execute a shell commandpython scripts/generate_schematic.py "your diagram description". This pattern is susceptible to command injection if the "diagram description" (which is derived from untrusted user content) is not properly sanitized or escaped before execution. - [NO_CODE]: The file
scripts/generate_schematic.pyis referenced in the documentation and designated for execution but is missing from the provided skill files. - [PROMPT_INJECTION]: The skill is designed to process untrusted external data (scholarly work).
- Ingestion points: Research papers, proposals, and literature reviews processed by the agent.
- Boundary markers: None provided in the instructions to separate untrusted content from commands.
- Capability inventory: Execution of Python scripts via subprocess calls (
calculate_scores.pyand the missinggenerate_schematic.py). - Sanitization: No sanitization logic is present to handle malicious instructions embedded in the documents being evaluated.
Audit Metadata