The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to ingest and process local project documents (e.g., vision/VISION.md, customer-blueprint/CUSTOMER.md) and existing brand assets to automatically populate design specifications. This creates an indirect prompt injection surface where instructions embedded in these source files could influence the agent's behavior during the interview or site generation process.
Ingestion points: Reads vision/VISION.md and customer-blueprint/CUSTOMER.md from the local workspace.
Boundary markers: Not explicitly implemented in the instructions.
Capability inventory: Performs file system writes (DESIGN.md, tokens.css, site code) and shell command execution (dev server).
Sanitization: The skill identifies missing values and uses explicit PLACEHOLDER markers rather than attempting to fabricate data, which is a defensive practice.
[COMMAND_EXECUTION]: The procedure includes a step to 'Run the dev server' to verify the functionality of the generated site (Vite/React). This involves shell command execution, which is an expected behavior for a developer-centric scaffolding skill.
[EXTERNAL_DOWNLOADS]: The skill's architecture utilizes well-known technology stacks and libraries, specifically the Vite build tool and the React framework. These are well-known services and do not represent a security risk in this context.

potatuhs-design