The Agent Skills Directory

[PROMPT_INJECTION]: Indirect Prompt Injection Surface
Ingestion points: Untrusted data is ingested from web pages via browse snapshot and browse eval (accessibility tree and DOM content) and from local code via git diff.
Boundary markers: The skill uses structured output markers (STEP_PASS/STEP_FAIL) and mandates explicit step budgets for sub-agents to restrict autonomous execution depth.
Capability inventory: The skill uses Bash to execute the browse CLI, handles file system operations in the .context directory, and delegates tasks using the Agent tool.
Sanitization: No specific sanitization of ingested web content is performed before analysis, although the skill includes tests specifically designed to detect and report XSS vulnerabilities in the target application.
[EXTERNAL_DOWNLOADS]: Fetches testing libraries from well-known services
The skill downloads the axe-core library from Cloudflare's CDN (cdnjs.cloudflare.com) for accessibility testing. This is a standard practice for web auditing tools and utilizes a well-known service.

ui-test