The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external source sites, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Source site screenshots, HTML specs, and manifest files are read from the outputDir and processed during the visual review and AI-driven rebuilding steps.
Boundary markers: No specific boundary markers or delimiters for the prompt interpolation are described in the subagent dispatch instructions.
Capability inventory: The skill can execute shell commands via Bash and modify files via Write and Edit tools.
Sanitization: The skill implements a robust mitigation strategy using the liberate_validate_artifacts tool, which explicitly checks for escaped text, prohibits raw PHP or script tags, and ensures content provenance to prevent prompt injection or stored XSS from being carried over from the source material.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to manage the local environment and facilitate the site replication workflow, including theme installation and visual diffing. These operations are scoped to the project's working directory and the intended automation tasks.
[EXTERNAL_DOWNLOADS]: The skill references legitimate services and tools including Automattic Studio for site previews and the pngjs library for image analysis. These resources are standard for the skill's documented purpose.

design-qa