The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's core workflow involves fetching and analyzing content from arbitrary, user-supplied URLs using the WebFetch tool. This external content is untrusted and serves as the primary input for the agent's logic in Step 1. A maliciously crafted website could embed instructions designed to override the agent's behavior or manipulate the generated scraping logic.
Ingestion points: Target website content fetched via WebFetch and WebSearch as described in Step 1.
Boundary markers: The instructions do not define boundary markers or clear 'ignore' directives to prevent the agent from obeying instructions found within the scraped data.
Capability inventory: The skill utilizes Bash, Write, Edit, and Read tools, providing a broad surface for file system and command-line operations.
Sanitization: There is no requirement for the agent to sanitize or filter the fetched HTML/JavaScript content before using it to generate scraper code.- [DYNAMIC_EXECUTION]: The skill is designed to generate Python scripts (scrape.py) and dependency lists (requirements.txt) at runtime. These scripts are then executed by the agent to perform data extraction.
Risk Factor: While templates are provided, the specific scraping logic is synthesized based on untrusted external data. This creates a chain where remote content influences locally executed code.
Execution Method: The generated code is intended to be executed via the Bash tool to perform the scraping and data cleaning tasks.

web-scraper-as-a-service