The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text content from the DOM to create animations. It mitigates potential injection risks by exclusively using textContent and createTextNode instead of innerHTML, ensuring that any HTML-like strings in the input text are rendered as literal text rather than being parsed as executable code.
Ingestion points: element.textContent in the splitWordReveal function (SKILL.md).
Boundary markers: Not applicable as it is a client-side UI utility.
Capability inventory: No subprocess calls, no eval(), no file-system access, and no network operations are present in any of the provided files.
Sanitization: Native browser sanitization is utilized via the textContent API.
[REMOTE_CODE_EXECUTION]: No remote scripts or external dependencies are referenced. All logic, including the complex canvas animation in the demo, is self-contained and implemented using native Web APIs.
[DATA_EXFILTRATION]: There are no network calls (fetch, XMLHttpRequest, etc.) or evidence of data being sent to external domains. The skill operates entirely within the local browser context.
[OBFUSCATION]: Analysis of the scripts and CSS found no evidence of Base64 encoding, zero-width characters, homoglyphs, or other techniques designed to hide malicious intent.

staggered-word-reveal