The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input in the form of design briefs and existing codebases to generate new markup and perform research.
Ingestion points: The system ingests external data through design briefs (SKILL.md) and existing HTML/CSS source code during redesign tasks (redesign.md).
Boundary markers: The "Gate 0 — Brief Lock" (SKILL.md) establishes a checkpoint where design parameters must be formally recorded before code generation begins, providing a structural boundary for agent operations.
Capability inventory: The skill possesses the ability to generate HTML/CSS code, perform web searches using platform-specific tools, capture screenshots for critiques, and spawn sub-agents to explore design directions.
Sanitization: While the skill enforces "Absolute bans" and "AI-slop tests" to ensure design quality and prevent generic output, these are stylistic constraints rather than security-focused sanitization filters.
[COMMAND_EXECUTION]: The skill's instructions guide the agent to use command-line utilities for automated quality assurance.
Evidence: Technical audit procedures in references/moves/audit.md and references/register/copy.md utilize grep to scan generated files for forbidden fonts, marketing buzzwords, and placeholder company names. These commands are executed within the local environment against the project's own assets.

design-method