The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from arbitrary external websites, creating a surface for indirect prompt injection. \n
Ingestion points: mcp__chrome-devtools__navigate_page is used to load user-supplied URLs as seen in SKILL.md. \n
Boundary markers: The skill does not use specific delimiters or explicit instructions to the agent to ignore potentially malicious instructions embedded in the page text during the anti-bot check or synthesis phase. \n
Capability inventory: The skill utilizes mcp__chrome-devtools__navigate_page and mcp__chrome-devtools__evaluate_script across its workflow (documented in SKILL.md and references/analysis-guide.md). \n
Sanitization: Extraction scripts are structurally limited to CSS properties, but the initial anti-bot check reads document.body.innerText directly, which could contain adversarial content. \n
[COMMAND_EXECUTION]: The skill relies on executing JavaScript snippets in the browser context via the Chrome DevTools MCP. \n
Evidence: Multiple JavaScript blocks in references/analysis-guide.md are designed to be run using mcp__chrome-devtools__evaluate_script to programmatically gather page metadata, styles, and interaction states.

design-system-analyzer