The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the subprocess module in scripts/render_roadmap.py to execute external commands. Specifically, it calls the drawio CLI to render diagrams and can execute brew install --cask drawio to handle dependency installation on macOS systems.
[EXTERNAL_DOWNLOADS]: In scripts/render_roadmap.py, the skill includes functionality to automatically install the Draw.io application using the Homebrew package manager if configured by the user (auto_install_macos: true). While targeted at a well-known service, this involves downloading and executing third-party software at runtime.
[PROMPT_INJECTION]: The skill processes untrusted content from user-supplied research proposals via scripts/extract_proposal.py. This content is interpolated into prompts for the Gemini image generation model in scripts/generate_roadmap.py without robust boundary markers or sanitization.
Ingestion points: scripts/extract_proposal.py reads user-provided .tex and .md files.
Boundary markers: Absent in the prompt construction logic within _build_nano_banana_prompt.
Capability inventory: The skill can perform network operations (Gemini API) and write files to the output directory.
Sanitization: No specific sanitization or escaping is applied to the extracted text before it enters the LLM prompt context.

nsfc-roadmap