The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script law_to_markdown.py resolves and executes a JavaScript file using osascript. The location of this file is tied to an external skill (mineru-ocr) which the instructions guide the user to install from an untrusted third-party GitHub repository (github.com/cat-xierluo/legal-skills). This establishes a critical dependency on unverified code that runs with local user privileges.\n- [EXTERNAL_DOWNLOADS]: The documentation and source code provide explicit instructions and a shell command to download software from a non-trusted GitHub source (cat-xierluo). This represents a significant supply-chain risk as the external repository is not verified and is outside the control of the skill's author.\n- [COMMAND_EXECUTION]: The skill utilizes subprocess.run to invoke system commands (osascript). It is designed to execute scripts from a directory that the user is prompted to populate with code from an untrusted source, potentially facilitating arbitrary script execution.\n- [PROMPT_INJECTION]: The skill ingests untrusted external data which is then processed and analyzed by an LLM to make structural decisions.\n
Ingestion points: Document reading in law_to_markdown.py using pdfplumber and python-docx.\n
Boundary markers: No delimiters or instructions to ignore embedded commands are present in the output text.\n
Capability inventory: Subprocess execution via osascript and file system write access.\n
Sanitization: The skill does not escape or validate the content of processed documents before providing it to the model.

law-to-markdown