The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses the uvx tool in text_handler.py and pdf_handler.py to download and execute code from remote sources, including the author's own repository (github.com/grahama1970/treesitter-tools.git) and the well-known marker-pdf package.
[COMMAND_EXECUTION]: Multiple modules use subprocess.run to call CLI utilities such as memory-agent, pkill, curl, and marker_single. Some arguments are derived from user-provided file paths.
[EXTERNAL_DOWNLOADS]: url_handler.py uses httpx and curl to fetch content from user-provided URLs for distillation.
[PROMPT_INJECTION]: The skill processes untrusted document content and interpolates it into LLM prompts, creating a surface for indirect prompt injection. Ingestion points: cli.py (file, URL, and text inputs). Boundary markers: Newline delimiters used in qra_prompts.py. Capability inventory: subprocess.run (CLI calls) and httpx.get (Network). Sanitization: Text is truncated but not specifically sanitized to prevent instruction leakage.

doc2qra