The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from web pages and local source code to perform audits, which creates a surface for indirect prompt injection. An attacker could embed malicious instructions in a website's HTML or a repository's source code to influence the agent's behavior.
Ingestion points: The skill reads DOM content, computed styles, and animation states via document.querySelectorAll and document.getAnimations in references/browser-profiling.md.
Capability inventory: The agent has access to shell commands (git status, git commit, npm run build) and browser control through playwright.evaluate.
Boundary markers: None identified in the provided instructions; the agent relies on its internal safety layers when processing the results of profiling.
Sanitization: Ingested DOM data and computed styles are processed directly without explicit escaping or filtering.
[COMMAND_EXECUTION]: The skill frequently uses repository-level commands, including git status, git read-tree, and git commit, as well as build tools like npm run build and npm run lint. These are standard for development tasks but represent a capability to modify the local environment.
[DYNAMIC_EXECUTION]: The skill executes complex JavaScript payloads within the browser context using Playwright's evaluate function to gather performance metrics. While the payloads are based on predefined templates in references/browser-profiling.md, they represent dynamic execution of code within the user's browser environment.

optimize-web-animations