data-scraper-agent

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection where untrusted data from external websites is processed by an LLM.
  • Ingestion points: Data is fetched from arbitrary external URLs in scraper/sources/.
  • Boundary markers: The _build_prompt function in ai/pipeline.py lacks robust delimiters or instructions to ignore embedded commands within the scraped text.
  • Capability inventory: The agent can write data to Notion databases and update its own repository state via GitHub Actions.
  • Sanitization: No explicit sanitization or filtering of the scraped content is performed before interpolation into the prompt.
  • [COMMAND_EXECUTION]: The GitHub Actions workflow defined in .github/workflows/scraper.yml executes shell commands to install dependencies, run the scraper, and perform git operations (commit and push) to persist feedback data.
  • [DATA_EXFILTRATION]: By design, the skill collects data from public sources and transmits it to the Google Gemini API (a well-known service) for analysis and subsequently to Notion or other storage providers. This behavior is consistent with the skill's stated purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 04:00 AM
Security Audit — agent-trust-hub — data-scraper-agent