The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a system that ingests external Markdown files, creating a surface for indirect prompt injection.
Ingestion points: Slide frontmatter src attribute (e.g., src: ./other.md) documented in references/core-frontmatter.md and references/syntax-importing-slides.md.
Boundary markers: The documentation does not specify delimiters or warnings to prevent the agent from obeying instructions embedded in imported slides.
Capability inventory: Documents significant capabilities including local filesystem writing (monaco-write), code execution (monaco-run), and network tunneling (--tunnel).
Sanitization: No specific sanitization or validation mechanisms for imported content are described.
[COMMAND_EXECUTION]: The documentation includes various CLI commands for initializing, serving, and exporting presentations.
Evidence: SKILL.md and references/core-cli.md provide examples of pnpm and slidev commands for managing the presentation lifecycle.
[EXTERNAL_DOWNLOADS]: The skill references several external dependencies required for full functionality.
Evidence: references/core-exporting.md specifies the need for playwright-chromium for exports. references/style-icons.md and references/editor-prettier.md recommend installing various npm packages like prettier-plugin-slidev and @iconify-json collections.
[DYNAMIC_EXECUTION]: The skill documents features that allow for the interactive execution of code within the presentation environment.
Evidence: references/editor-monaco-run.md details the {monaco-run} feature for executing JavaScript or TypeScript code snippets directly in the browser.
[DATA_EXFILTRATION]: The documentation describes remote access features that could be leveraged to expose local data.
Evidence: references/presenter-remote.md explains the --remote and --tunnel flags, the latter of which uses Cloudflare Quick Tunnels to create public URLs for the local presentation server.

slidev