synthetic-data
Fail
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: HIGHCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
generateWithSDVfunction inreferences/patterns.mdis vulnerable to command injection. It constructs a Python script by interpolating variables likeconfig.synthesizerandinputPathdirectly into a string template which is then executed viaspawn("python3", ["-c", pythonScript]). If these variables are influenced by an attacker, they can break out of the string literal to execute arbitrary code within the Python process. - [DYNAMIC_EXECUTION]: The skill implementation patterns include the assembly of executable Python strings at runtime which are subsequently passed to a system shell execution environment (
python3 -c). This practice of building and running dynamic code from string segments is a significant security risk when those segments include external inputs. - [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for synthetic data generation and augmentation that interpolate untrusted input strings (
context,examples) directly into system and user prompts without delimiters or sanitization. - Ingestion points:
references/patterns.mdin functionsgenerateSyntheticData(via thecontextargument) andaugmentTextData(via theexamplesarray). - Boundary markers: No boundary markers, delimiters, or isolation instructions are present in the prompt templates.
- Capability inventory: The skill features capabilities for command execution via
child_process.spawn, file system manipulation viafs/promises, and external API communication through AI SDKs. - Sanitization: There is no evidence of input validation, escaping, or sanitization applied to user data before it is interpolated into LLM requests.
Recommendations
- AI detected serious security threats
Audit Metadata