The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md reference an external text-polishing tool called shuorenhua. It provides a link to a public GitHub repository (github.com/MrGeDiao/shuorenhua) and suggests its use for final document refinement. This is presented as an optional dependency to be used when authorized.
[DATA_EXPOSURE]: The evals/evals.json file contains hardcoded local file paths (e.g., /Users/wayne/Downloads/FreiHAND-...pdf). This reveals the local username ('wayne') and directory structure of the developer's environment, which is common in development artifacts but constitutes a minor information disclosure.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external documents (PDFs) provided by the user. This creates a surface for indirect prompt injection if a research paper were to contain hidden instructions. The skill mitigates this risk by instructing the agent to 'freeze' technical content and facts before any stylistic polishing occurs and to prioritize inspected evidence over inference.

research-paper-deep-dive