The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform automated setup and execution of a Node.js script. Specifically, it directs the agent to check for Node.js, install dependencies via npm install, and run the extraction script (extract-docx.mjs) without requiring explicit user consent for these operations. This behavior reduces user oversight of command execution.
[EXTERNAL_DOWNLOADS]: The skill triggers external downloads from the npm registry. It instructs the agent to run npm install for the packages mammoth, jszip, and xml2js. While these are well-known libraries, the automated nature of the installation process poses a supply chain risk if the environment is not restricted.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to process untrusted data (Word documents) and convert them into structured Markdown for further agent consumption. This creates a surface for indirect prompt injection, where malicious instructions hidden in a PRD document could influence the agent's behavior during the subsequent analysis or PRD generation steps.
Ingestion points: Reads and parses user-provided .docx files (untrusted data).
Boundary markers: Instructions tell the agent to use tags like [UNKNOWN] and [INFERRED], but there are no strict structural boundaries to prevent the agent from obeying instructions embedded within the extracted text.
Capability inventory: File writes to docs/prd/, directory creation, and command execution capabilities (via terminal).
Sanitization: Uses mammoth for HTML/Markdown conversion, which provides some structural sanitization, but does not filter for natural language instructions in the document body.

product-prd-extract