The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run to call external binaries, specifically pandoc in scripts/contract_analyzer.py for text extraction and mmdc (Mermaid CLI) in scripts/mermaid_renderer.py for flowchart rendering. While these are executed with list-based arguments to prevent shell injection, they represent a dependency on local system tools.
[INDIRECT_PROMPT_INJECTION]: The core functionality involves the agent reading and analyzing untrusted text provided in contract files (.docx). Maliciously crafted contracts could contain embedded instructions designed to influence the agent's review logic, bypass the provided checklist, or bias the generated opinions and summaries. The skill lacks specific boundary markers or instructions to the agent to disregard such embedded commands.
[DYNAMIC_EXECUTION]: The scripts/mermaid_renderer.py module renders Mermaid diagrams using mmdc, which typically invokes a headless browser (Puppeteer). The script includes fallback logic to locate local Chrome or Chromium installations and generates a temporary Puppeteer configuration using the --no-sandbox flag, which is a common but less secure execution mode for browser processes.
[COMMAND_EXECUTION]: The scripts/ooxml/unpack.py utility uses zipfile.extractall() to decompress .docx files. This operation can be vulnerable to path traversal (ZipSlip) attacks if the input file contains malicious relative paths designed to overwrite files outside the intended target directory, although the workflow attempts to scope the extraction to a generated output folder.

contract-review