scientific-critical-thinking
Pass
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data sources such as research papers and media reports, creating a vulnerability to indirect prompt injection attacks.
- Ingestion points: Untrusted data enters the agent context through the analysis of research methodology, scientific claims, and media reports (SKILL.md).
- Boundary markers: Absent. The instructions provide no clear delimiters or directives to the agent to distinguish between its own logic and instructions that might be embedded within the documents it reads.
- Capability inventory: The skill utilizes the
Bash,Write, andEdittools and is explicitly instructed to execute shell commands to generate diagrams based on processed content (SKILL.md). - Sanitization: Absent. There is no mention of sanitizing, escaping, or filtering content retrieved from external sources before it is interpreted by the agent.
- [COMMAND_EXECUTION]: The skill instructions direct the agent to execute a shell command (
python scripts/generate_schematic.py) using natural language input provided in the document context. This pattern introduces a risk of command injection if the underlying Python script does not strictly sanitize the arguments passed to it via the shell.
Audit Metadata