The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes 'inspector comments' embedded in JSX source code ({/* @slide-comment ... */}). These markers contain base64url-encoded JSON with a note field containing natural language instructions (e.g., 'make this heading bigger') that the agent is directed to execute as code edits in the rules/apply-comments.md workflow. This creates a surface where malicious or unexpected instructions in a slide file could influence the agent's behavior.
Ingestion points: The agent reads slides/<id>/index.tsx to find and decode these markers.
Boundary markers: Instructions are wrapped in JSX comment markers, but the agent is specifically told to decode and follow the inner note text.
Capability inventory: The skill assumes access to the Edit tool to modify React components in the slide files.
Sanitization: The instructions advise interpreting comments 'in context' and making 'reasonable' interpretations, but there is no formal validation or sanitization of the embedded instructions.
[EXTERNAL_DOWNLOADS]: The skill instructs project initialization using npx @open-slide/cli init, which fetches and executes code from the npm registry. It also provides instructions for loading Hebrew fonts from Google Fonts (fonts.googleapis.com). These are well-known services and align with the skill's stated purpose.
[COMMAND_EXECUTION]: The skill suggests using node -e as a helper to decode base64url payloads from inspector markers and mentions standard project scripts like npm run dev and npm run sync:skills. These are routine development operations within this framework's ecosystem.

open-slide-best-practices