The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements patterns that process untrusted user input, creating a surface for indirect prompt injection.
Ingestion points: User-controlled data enters the application context via the markdown editor textarea in references/2024-10-08-turbo-frames-markdown-preview.md and the search form inputs in references/2024-12-10-stimulus-turbo-frames-faceted-search.md.
Boundary markers: The provided examples do not demonstrate the use of delimiters or specific instructions to the agent to differentiate between system instructions and user-provided data within these inputs.
Capability inventory: The skill utilizes Turbo Frames to render content dynamically based on this data, which could be exploited to influence agent behavior if the resulting content is not properly sanitized.
Sanitization: While the documentation notes that the server should transform data (e.g., markdown to HTML), the specific snippets do not include explicit sanitization logic or output encoding.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the use of established third-party libraries for DOM manipulation and animation.
References the morphdom library from a public GitHub repository for DOM diffing in references/2023-05-09-turbo-drive-custom-rendering.md.
References the gsap library for image transitions and animations in references/2023-05-09-turbo-drive-custom-rendering.md.
These are well-known and standard libraries within the web development ecosystem.

hwc-navigation-content