The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's architecture relies on reading external configuration and existing project files, creating a vector where untrusted data could influence the agent's behavior.
Ingestion points: The skill reads project-level config (ai/style.config.md), user-level config (~/.claude/style.config.md), and existing project source files during revamp or qa operations.
Boundary markers: The instructions lack specific requirements for using unique delimiters or "ignore embedded instructions" warnings when interpolating these configurations into the subagent tasks or analysis routines.
Capability inventory: The skill possesses the ability to write files (./proposals/), launch subagents with complex prompts, execute shell-based audits via grep, and render generated content using Puppeteer.
Sanitization: No explicit sanitization or validation logic is defined for the content extracted from the configuration files.
[COMMAND_EXECUTION]: The qa mode instructions include the execution of shell commands (grep, wc) to perform static analysis on the project's source code. While these are used for linting and auditing purposes, they provide a mechanism for shell interaction based on skill-driven logic.
[DYNAMIC_EXECUTION]: The skill generates self-contained HTML files containing design elements and layout logic. These files are subsequently executed/rendered using mcp__puppeteer__puppeteer_navigate, representing a workflow where the agent creates and then views dynamic web content.

frontend-design