The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of data from arbitrary external URLs and local files for downstream processing by an LLM. This creates a surface for indirect prompt injection where malicious instructions hidden in scraped content could influence agent actions.
Ingestion points: Data is pulled from external sources via requests and the Firecrawl SDK in scripts/local_bs4_example.py and scripts/firecrawl_example.py.
Boundary markers: The skill does not implement or mandate strict delimiters or isolation instructions for the ingested content.
Capability inventory: The skill maintains capabilities for network requests, filesystem writes, and local script execution.
Sanitization: The validation logic in scripts/validate_extraction.py focuses on structural integrity (JSON/CSV well-formedness) rather than content-based sanitization for malicious instructions.
[EXTERNAL_DOWNLOADS]: The requirements.txt file includes standard, well-known Python packages such as firecrawl, pandas, requests, and beautifulsoup4. These are used for their intended purposes in data extraction and are fetched from public registries.
[COMMAND_EXECUTION]: The SKILL.md instructions explicitly guide the agent to execute the provided Python runner scripts to perform scraping, data cleaning, and validation tasks.

universal-scraping-architect