The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external web pages to collect performance metrics.
Ingestion points: Performance measurement functions in references/browser-profiling.md extract animation names, CSS classes, and element text from the live DOM.
Boundary markers: Collected metrics are returned as structured data; however, there are no specific instructions to ignore malicious directives that might be embedded in the audited page content.
Capability inventory: The skill allows the agent to modify source code, run build scripts (npm run build), and execute git commands (git commit).
Sanitization: Textual data extracted from the page is truncated using .slice() methods and whitespace is normalized, providing minimal protection against injection.
[COMMAND_EXECUTION]: The skill utilizes local shell commands to manage the repository and verify changes.
Evidence: Instructions in SKILL.md include running git status, npm run lint, npm run build, and a complex git commit sequence involving a temporary index file.
[DYNAMIC_EXECUTION]: JavaScript code is dynamically executed within the browser environment to perform audits.
Evidence: The references/browser-profiling.md file contains several tab.playwright.evaluate blocks that execute code at runtime to inspect the DOM and performance APIs.

optimize-web-animations