file-converter
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill is designed to generate Python scripts at runtime based on user requests and then execute them.
- Evidence: "Generates conversion code dynamically based on the specific request." (SKILL.md)
- Evidence: "Execute conversion and report results" (SKILL.md Workflow step 6).
- The instructions provide multiple code templates for various conversion types (JSON to YAML, CSV to JSON, Markdown to HTML, PNG to WebP, SVG to PNG) that the agent is expected to use to assemble and run executable scripts.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from various file formats and has the capability to write files and execute commands, creating a surface for indirect prompt injection.
- Ingestion points: Processes external files including PDF, DOCX, CSV, XML, TOML, and YAML as defined in SKILL.md and various reference files.
- Boundary markers: No explicit instructions or delimiters are provided to ensure the agent treats file content strictly as data and ignores any embedded natural language instructions.
- Capability inventory: The skill has the capability to write to the local filesystem, execute shell commands (e.g., libreoffice, potrace), and run dynamically generated Python code.
- Sanitization: While yaml.safe_load is recommended for YAML handling, there is a lack of guidance on sanitizing or escaping content from other formats like PDF or XML, which could be used to deliver malicious instructions to the agent.
Audit Metadata