The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted document content (DOCX and Markdown files), which introduces a surface for indirect prompt injection attacks.
Ingestion points: Documents are ingested as file paths and read in to-markdown/scripts/lib/converter.cjs and to-docx/scripts/lib/config-loader.cjs.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore natural language commands embedded within the document content.
Capability inventory: The skill utilizes file system write operations (fs.writeFileSync) and subprocess execution (spawnSync) to handle the conversion pipeline.
Sanitization: While the content is parsed and restructured into different formats (HTML, Markdown, OpenXML), there is no specific sanitization to filter out malicious prompt instructions in the text.
[COMMAND_EXECUTION]: The dispatcher script in scripts/convert.cjs uses spawnSync to route commands to specific conversion scripts.
Evidence: It executes process.execPath (the Node.js binary) with arguments derived from the skill's own internal directory structure and user-supplied CLI flags. While this is used for legitimate internal routing, it represents a command execution capability.

docx-convert