dataset-generator
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [REMOTE_CODE_EXECUTION]: The documentation and installation guides recommend an installation pattern using 'curl | bash' or 'powershell -File' pointing to the author's GitHub repository (Bhanunamikaze/AI-Dataset-Generator). This is a common but unverifiable execution method for installing tools in this environment.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data from external sources, including JSONL/CSV files, URLs, and web search results. To mitigate risks, it implements a robust security utility in 'scripts/utils/security.py' that identifies common prompt injection markers (e.g., 'ignore previous instructions', 'role override') and flags suspicious records for manual review. Furthermore, 'sub-skills/llm-judge.md' explicitly instructs the agent to treat all record content as untrusted data and to ignore any instructions embedded within them.
- [COMMAND_EXECUTION]: The orchestration script 'scripts/build_loop.py' and the test suite ('tests/test_pipeline.py') utilize 'subprocess.run' to execute internal Python scripts within the pipeline. These executions use the current Python interpreter and are restricted to scripts provided within the skill package.
- [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch source material for datasets. This includes using web search APIs (SerpAPI, Bing, Google CSE) and directly fetching content from user-provided URLs via 'scripts/collect.py', 'scripts/research.py', and 'scripts/browser_collect.py' (which uses Playwright).
Audit Metadata