scientific-critical-thinking
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze untrusted external data like research papers and scientific claims, which serves as a significant attack surface.\n
- Ingestion points: Scientific documents, claims, and study designs loaded into the agent context via the
Readtool or user prompts (SKILL.md).\n - Boundary markers: Absent; the instructions do not specify delimiters or warnings to ignore embedded instructions within processed data.\n
- Capability inventory: The skill has access to the
Bashtool to run local scripts and instructs the agent to invoke a secondaryscientific-schematicsskill based on content descriptions (SKILL.md).\n - Sanitization: Absent; there is no guidance on escaping or validating natural language descriptions derived from external content before they are used in shell commands or sub-prompts.\n- [COMMAND_EXECUTION]: The skill contains specific instructions to run
python scripts/generate_schematic.pywith natural language arguments. If these arguments are automatically generated based on the content of an untrusted research paper, it creates a vector for command injection or unintended system behavior.
Audit Metadata