The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from user-provided URLs and utilizes uvx to retrieve the treesitter-tools utility from the vendor's GitHub repository. It also downloads well-known PDF extraction tools such as pymupdf4llm and marker-pdf if they are not already installed.
[REMOTE_CODE_EXECUTION]: The script uses uvx to dynamically run tools fetched from remote sources, specifically the treesitter-tools utility from the vendor's repository (github.com/grahama1970) and marker-pdf for enhanced PDF processing.
[COMMAND_EXECUTION]: The skill executes shell commands via subprocess.run to interact with PDF extraction tools, code symbol analyzers, and the memory-agent for data storage. It includes a custom function to stream long-running subprocess output to the terminal for monitoring.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs and PDF files and includes it in prompts sent to the LLM for knowledge extraction, creating a vulnerability surface for indirect prompt injection.
Ingestion points: fetch_url() and read_file() functions in distill.py which load content from external sources.
Boundary markers: The system prompt uses a 'Text:' label for content but does not employ strong delimiters or explicit instructions for the LLM to ignore potentially malicious embedded instructions.
Capability inventory: The skill possesses the ability to execute shell commands (subprocess.run), perform network requests (urllib.request), and write to the local file system.
Sanitization: Extracted content is not sanitized or filtered for prompt injection patterns before being interpolated into LLM requests.

distill