The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the browse CLI and fetches the axe-core library from a public CDN (cdnjs.cloudflare.com) to perform accessibility audits. These resources are either vendor-owned or originate from well-known, trusted services.
[COMMAND_EXECUTION]: The skill executes a variety of system commands, including git for diff analysis, package managers (npm, yarn, pnpm) for dependency management, and the browse CLI for browser automation. It also includes instructions for users to whitelist these commands in their agent settings to reduce manual approval prompts.
[REMOTE_CODE_EXECUTION]: The skill utilizes the browse eval functionality to inject and execute JavaScript within the browser context. This is primarily used for running accessibility checks, capturing performance metrics, and monitoring console logs.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as git diffs and live web page content. This surface could theoretically be used for indirect prompt injection if malicious instructions were embedded in a codebase or on a website.
Ingestion points: git diff outputs and browser content retrieved via browse snapshot and browse eval in SKILL.md.
Boundary markers: Not explicitly defined for the processed data streams.
Capability inventory: Extensive use of Bash for command execution and the Agent tool for sub-agent coordination.
Sanitization: No specific sanitization or filtering of the ingested external content is mentioned.
[DYNAMIC_EXECUTION]: The skill dynamically generates JavaScript snippets at runtime to be executed via browse eval, enabling flexible assertions and data extraction based on the testing context.

ui-test