The Agent Skills Directory

[PROMPT_INJECTION]: The skill framework is vulnerable to indirect prompt injection because it is designed to scrape arbitrary external data and interpolate it directly into LLM prompts.\n
Ingestion points: External data is fetched from public websites and APIs via the sources defined in scraper/sources/ and aggregated in scraper/main.py.\n
Boundary markers: The prompt template in ai/pipeline.py uses markdown headers to organize items but does not include explicit instructions to the AI model to ignore potential commands embedded within the scraped text.\n
Capability inventory: The agent possesses the capability to write to external databases (Notion/Supabase) and perform automated commits and pushes to its host repository.\n
Sanitization: No sanitization, validation, or escaping of the scraped content is performed before it is added to the AI analysis prompt.\n- [COMMAND_EXECUTION]: The skill uses GitHub Actions with elevated permissions to automate tasks and maintain state.\n
The .github/workflows/scraper.yml file specifies permissions: contents: write and executes git push commands to save the feedback history in data/feedback.json.\n- [EXTERNAL_DOWNLOADS]: The skill relies on several external Python packages and binary installers for its operation.\n
Dependencies include well-known libraries such as requests, beautifulsoup4, playwright, and notion-client.\n
The setup instructions include the automated download of browser binaries via python -m playwright install to facilitate scraping of JavaScript-heavy websites.

data-scraper-agent