The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It is designed to scrape arbitrary data from public websites and APIs and immediately process that data through an LLM for 'enrichment' (scoring, summarizing, and classifying).
Ingestion points: The fetch function in scraper/sources/my_source.py and various scraping patterns (RSS, HTML) ingest untrusted content from external URLs.
Boundary markers: The _build_prompt function in ai/pipeline.py interpolates the scraped data (as JSON-stringified items) directly into the LLM prompt. There are no clear boundary markers or system instructions provided to the LLM to ignore potentially malicious instructions embedded in the scraped data.
Capability inventory: The agent scripts have capabilities to perform network requests (requests), write to external databases (notion-client), and commit changes to the repository (git commit via GitHub Actions).
Sanitization: No sanitization, filtering, or validation of the scraped text is performed beyond a length limit (context[:800]) and standard JSON encoding.
[EXTERNAL_DOWNLOADS]: The skill requires several external Python packages to function. It provides a requirements.txt template and a GitHub Actions workflow that performs automated installations.
Python packages: requests, beautifulsoup4, lxml, python-dotenv, pyyaml, notion-client, and playwright.
The GitHub Actions workflow also includes a step to download and install Playwright browser binaries (python -m playwright install chromium).
[COMMAND_EXECUTION]: The skill includes a GitHub Actions workflow (.github/workflows/scraper.yml) that executes shell commands on a schedule. This includes installing dependencies, running the Python agent, and using git commands to push feedback data back to the repository.
[DATA_EXFILTRATION]: By design, the skill collects data from external sources and transmits it to third-party storage providers (Notion, Google Sheets, or Supabase) using API tokens provided by the user in environment variables.

data-scraper-agent