The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the requests library from the official Python package registry (PyPI) during runtime to support API-based workflows.
[COMMAND_EXECUTION]: Executes the pip command via subprocess.check_call to manage library installation in scripts/challenger.py. The command is limited to installing a specific, hardcoded package within sandboxed environments.
[DYNAMIC_EXECUTION]: Performs runtime installation and loading of Python modules to provide optional features without requiring pre-installed dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided content such as draft documents and technical briefs. Ingestion points: Data enters the system through the artifact, context, voice, and finding fields processed by scripts/challenger.py. Boundary markers: The skill wraps external content in XML-like tags (e.g., <artifact>, <context>) to separate untrusted data from system instructions. Capability inventory: Uses subprocess for package management and requests for network communication with Anthropic, Google, and Cloudflare APIs. Sanitization: System prompts in the references/ directory include an explicit 'TRUST BOUNDARY' instruction, directing the model to treat the wrapped content strictly as data for review, not as commands.

challenging