The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes well-known, industry-standard Python libraries for its core functionality, including requests for networking, beautifulsoup4 for HTML parsing, playwright for browser automation, and notion-client for database integration.
[COMMAND_EXECUTION]: The skill uses GitHub Actions for workflow automation and task scheduling. The configuration includes standard steps for environment setup and a git-based mechanism for persisting user feedback, which is clearly documented and within the expected scope of a scraper tool.
[CREDENTIALS_UNSAFE]: The implementation follows security best practices by using python-dotenv and os.environ to manage API keys (e.g., GEMINI_API_KEY, NOTION_TOKEN). It explicitly warns against hardcoding secrets and provides a .env.example file for safe local development.
[SAFE]: Indirect Prompt Injection Surface. As a data scraping tool, the skill naturally ingests untrusted content from the web.
Ingestion points: External data is retrieved via HTTP requests in the scraper/sources/ modules.
Boundary markers: The code in ai/pipeline.py uses json.dumps() to serialize scraped items before interpolating them into the LLM prompt, providing a clear boundary between data and instructions.
Capability inventory: The LLM's role is restricted to scoring and summarizing data; its output is not used to trigger system commands or dynamic code execution.
Sanitization: Structured JSON formatting is used to maintain separation of concerns between external data and the agent's internal logic.

data-scraper-agent