The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md suggest the installation of a forked version of the officecli tool from a personal GitHub repository (https://github.com/lidge-jun/OfficeCLI) via a shell script located in a global npm directory. While the skill appropriately requests user consent before installation, executing an external installation script to deploy a third-party fork represents a supply chain and code execution risk.
[REMOTE_CODE_EXECUTION]: Utility scripts such as scripts/docx_cli.py and tests/generate_fixtures.py employ dynamic path computation to modify the Python module search path (sys.path). This behavior is used to load the ooxml_core package from environment-defined locations or the user's home directory, which could lead to arbitrary module loading if those paths are compromised.
[COMMAND_EXECUTION]: The skill's primary functionality relies on executing external shell commands, specifically the officecli binary and soffice (LibreOffice). Furthermore, the test suite (scripts/run_tests.py) uses subprocess.run to execute internal scripts, representing a potential for misuse if arguments were to be influenced by untrusted data.
[PROMPT_INJECTION]: The skill processes content from external Word documents and provides functionality to chunk this text for the agent to read (scripts/docx_cli.py chunk). There are no explicit boundary markers or instructions to ignore embedded prompts within the extracted content. An attacker could craft a DOCX file containing hidden instructions that the agent might follow upon ingestion.
Ingestion points: Document text is read through officecli view and the chunking/searching subcommands in scripts/docx_cli.py.
Boundary markers: Absent; extracted content is presented to the agent without delimiters or safety warnings.
Capability inventory: Extensive file system modification and command execution capabilities across all skill scripts.
Sanitization: Although defusedxml is correctly used for secure XML parsing, the skill does not implement any sanitization or filtering of the natural language content extracted from the documents.

docx