generating-test-data

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The script scripts/generate_data.py includes a generate_script method that writes arbitrary content to a shell script (.sh) and marks it as executable using chmod 0o755. This allows for the creation of executable files from unvalidated input.
  • [INDIRECT_PROMPT_INJECTION]: The SKILL.md instructions direct the agent to generate 'Adversarial inputs' such as SQL injection strings, XSS payloads, and Unicode edge cases. This introduces a risk of indirect prompt injection where generated values could influence subsequent agent actions or system behavior if not properly handled.
  • Ingestion points: Data models, schemas, and type definitions read by the skill.
  • Boundary markers: No delimiters or ignore instructions identified for the generated adversarial content.
  • Capability inventory: Access to Write and Edit tools, plus a script for creating executable files.
  • Sanitization: No sanitization is performed on the data written to generated files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 06:18 PM
Security Audit — agent-trust-hub — generating-test-data