The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill scans untrusted UI source code and instructs the agent to perform live verification on rendered pages using browser automation. This process creates a surface where malicious content within the analyzed files could potentially influence the agent's behavior through the generated QA reports.
Ingestion points: The 'design-scan.sh' script reads source files (TSX, HTML, etc.), and 'live-verification.md' provides JavaScript probes to extract text and attributes from rendered pages.
Boundary markers: Findings are reported in a structured Markdown format with clearly defined gates. The scanner normalizes content to prevent control character injection.
Capability inventory: The skill utilizes shell commands (ripgrep, jq, ast-grep) and instructs the agent to use 'agent-browser' to evaluate JavaScript.
Sanitization: The scanner uses 'jq' to normalize output and escapes newlines and tabs from finding snippets.
[DYNAMIC_EXECUTION]: The 'reference/live-verification.md' file contains several complex JavaScript snippets (probes) that the agent is instructed to execute using the 'agent-browser' tool's evaluation feature. These snippets are used for tasks like measuring layout characteristics, checking accessibility heading order, and calculating rendered contrast ratios.
[COMMAND_EXECUTION]: The 'design-scan.sh' script executes several command-line tools, including 'ripgrep' (rg), 'jq', and 'ast-grep' (sg), to perform static analysis on the target project directory. It follows best practices for shell execution, such as using temp directories and handling hostile paths.

design-qa