The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bash script to perform environment discovery, checking for the presence of specific design tools and plugins within local directories such as .claude/skills and $HOME/.claude/skills. It also queries the agent's plugin configuration file (installed_plugins.json) to verify the installation status of design components.
[EXTERNAL_DOWNLOADS]: The skill identifies missing dependencies and provides the user with installation instructions. These instructions include references to official repositories from a trusted organization (Anthropics) and well-known services (such as Playwright and NPM). These are presented as suggestions for the user to action manually.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes untrusted user data.
Ingestion points: Untrusted data enters the context through the 'Required Inputs' section, specifically the 'Reference' and 'Constraints' fields provided by the user.
Boundary markers: None are present to delimit user-provided references from the agent's instructions.
Capability inventory: The skill has the capability to execute bash scripts for environment detection, write design specifications to the local file system (DIRECTION.md, MASTER.md), and interact with browsers via Playwright for verification.
Sanitization: No explicit sanitization or filtering of user-provided references is documented before they are incorporated into the design pipeline.
[SAFE]: The skill operates as intended for an orchestrator, using standard environment checks and referencing legitimate design tools without any signs of obfuscation, exfiltration, or persistence.

ui-design-pipeline