The Agent Skills Directory

[COMMAND_EXECUTION]: The script law_to_markdown.py uses subprocess.run to execute a JavaScript file using /usr/bin/osascript. This executes a script residing in a different skill folder (mineru-ocr), allowing for cross-skill logic execution.
[EXTERNAL_DOWNLOADS]: The skill's documentation and error logic direct users to install a dependency from an unverified GitHub repository (github.com/cat-xierluo/legal-skills). This source is not categorized as a trusted vendor.
[REMOTE_CODE_EXECUTION]: The main script dynamically resolves paths to, and then executes, code from the mineru-ocr skill folder. Because it recommends downloading this dependency from an external untrusted source, it facilitates the execution of unverified third-party code.
[PROMPT_INJECTION]: The skill processes untrusted document files (.txt, .docx, .pdf) which could contain adversarial instructions meant to manipulate the agent's behavior.
Ingestion points: The input parameter in law_to_markdown.py allows loading arbitrary external document files.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded prompts within the document content.
Capability inventory: The skill has the ability to read files, write to the local filesystem, and execute shell commands.
Sanitization: While the skill applies regular expressions for document normalization, it does not include sanitization to neutralize potential prompt injection attacks embedded in the documents.

law-to-markdown