karpathy-jobs-bls-visualizer
Pass
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches occupational handbook data from the official Bureau of Labor Statistics website (bls.gov) using Playwright browser automation.
- [REMOTE_CODE_EXECUTION]: Clones the source code from the 'karpathy/jobs' GitHub repository and executes it locally as part of the data processing pipeline.
- [COMMAND_EXECUTION]: Uses the 'uv' runner to execute multiple Python scripts (
scrape.py,process.py,make_csv.py,score.py,build_site_data.py) and launches a local HTTP server for the visualization frontend. - [DATA_EXFILTRATION]: Utilizes an OpenRouter API key provided by the user in a local environment file for LLM scoring; operations are consistent with the documented purpose of the tool.
- [PROMPT_INJECTION]: Identified as having an indirect prompt injection surface due to the ingestion and processing of external web content from the BLS website for LLM-based scoring.
- Ingestion points:
scrape.pydownloads HTML pages from bls.gov, which are then cleaned byprocess.py. - Boundary markers: Data is transitioned from raw HTML to Markdown before being sent to the LLM scoring pipeline in
score.py. - Capability inventory: The skill has the ability to write to the local file system (saving HTML, CSV, and JSON data) and execute shell commands via 'uv'.
- Sanitization: The
process.pyscript performs Markdown conversion on raw HTML data prior to LLM interaction.
Audit Metadata