The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local TypeScript script scripts/render-reader-formats.ts and spawns subprocesses for Google Chrome or Chromium to generate PDF files and render diagrams. It uses environment variables like CASE_WRITING_CHROME to locate these system binaries.
[DYNAMIC_EXECUTION]: The skill dynamically generates HTML from Markdown and uses a headless browser to execute JavaScript (Mermaid.js) for diagram rendering. This involves runtime assembly and execution of content within a browser context.
[INDIRECT_PROMPT_INJECTION]: The skill processes external files such as case.md and content from the research/ directory to generate the teaching note. This represents an attack surface where malicious instructions in the input data could influence the agent's output.
Ingestion points: Data is read from case.md and various files within the research/ evidence pack.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded prompts when interpolating these external sources into the final note.
Capability inventory: The skill can execute local scripts and spawn system processes via the rendering utility.
Sanitization: The rendering script includes routines for escaping HTML and attributes (escapeHtml, escapeAttr) and validating URLs (safeUrl) to mitigate risks during the rendering phase.
[EXTERNAL_DOWNLOADS]: The rendering script fetches the Mermaid.js library from jsDelivr (cdn.jsdelivr.net), a well-known and established content delivery network. This is documented as part of the tool's diagramming capability.

case-teaching-note