The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection to execute shell commands when the skill is loaded.
Evidence: Found in SKILL.md under the Pre-loaded context section: !git branch --show-current and `!`git diff --name-only HEAD~1 -- '*.jsx' '*.tsx' '*.html' '*.css' 2>/dev/null | head -20.
These commands are used to provide the agent with relevant project context (branch name and recently modified files) to assist in the validation process.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it ingests and processes data from external web pages and browser logs.
Evidence: The workflow describes navigating to URLs, capturing page content, and listing console messages (Phase 1, Phase 2, and Phase 4 in SKILL.md).
Ingestion points: Browser DOM, console messages (list_console_messages), and network request logs.
Capabilities: The skill can perform browser interactions (click, fill, hover) and execute lighthouse audits.
Risks: If the target website or its console logs contain malicious instructions, the agent could potentially be manipulated during the evaluation phase.

visual-validate