The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides instructions for saving sensitive session data, such as browser cookies and local storage state, to local files (e.g., cookies.json and state.json) to facilitate authenticated scraping. While this is a standard operational requirement for the skill's purpose, it creates a potential risk of exposing session credentials if the local environment is not properly secured.
[EXTERNAL_DOWNLOADS]: The skill recommends installing and using several third-party libraries and tools, including playwright, httpx, beautifulsoup4, scrapy, tls-client, and frida-tools. These are well-known and widely used tools in the web scraping and security research communities.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of ingesting and processing untrusted data from external websites. Ingestion points: Website content is fetched from arbitrary URLs via httpx and playwright, as documented in SKILL.md and references/static-scraping.md. Boundary markers: The skill employs markdownify to strip non-content HTML elements and uses explicit instructions within LLM prompts to enforce structured output (JSON). Capability inventory: The skill possesses capabilities for network requests, browser automation, local file writes, and invocation of LLM APIs. Sanitization: It implements a Pydantic-based validation and retry loop (extract_with_validation) in references/llm-extraction.md to ensure that data extracted by the LLM conforms to a predefined schema, reducing the risk of processing malicious instructions embedded in web content.
[COMMAND_EXECUTION]: The documentation includes shell commands for environment setup, such as package installation via pip and the execution of security research tools like frida for mobile app analysis.

executing-web-scraping