The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted content from source websites, such as HTML and layout metadata, which constitutes an indirect prompt injection surface. This risk is effectively mitigated by the mandatory use of the liberate_validate_artifacts tool, which sanitizes output and verifies that all generated text is a subset of the captured source content, preventing malicious instructions from influencing the agent's code generation.
Ingestion points: SKILL.md (processing replicaBaseUrl, archetypeReps, source screenshots, and live replica DOM).
Boundary markers: The skill establishes a strict Honesty Discipline and structural separation between spec capture and pattern generation.
Capability inventory: Tools for file modification (Write, Edit) and an AI-driven section rebuilding capability (rebuild-section).
Sanitization: The liberate_validate_artifacts gate checks for WordPress escaping functions (esc_html, esc_attr, esc_url) and prohibits unauthorized script or PHP tags.
[COMMAND_EXECUTION]: The skill uses the Bash tool to manage theme installations and file system operations within the Automattic Studio environment. These operations are scoped to the site output directory and are a fundamental part of the design QA and remediation workflow.
[REMOTE_CODE_EXECUTION]: Although the skill involves generating and deploying new theme code and block markup, it operates within a local development runtime (Automattic Studio) and applies automated security gates. These gates ensure that no remote CDN URLs or unescaped executable scripts are introduced into the theme artifacts during the reconstruction process.

design-qa