The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install standard GraphQL utility and development packages (such as graphql-request and @graphql-codegen/cli). It also suggests the use of @saleor/configurator via a package runner (pnpm dlx), which is an official tool from the skill's author.
[COMMAND_EXECUTION]: The agent executes shell commands to detect the project's package manager, install dependencies, run code generation tools, and check for the existence of other Saleor-related agent configuration files in the local repository.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it reads untrusted local files (e.g., package.json, framework configuration files) and environment variables to determine the project's architecture.
Ingestion points: package.json, nuxt.config.ts, next.config.js, svelte.config.js, remix.config.js, vite.config.js, and .env files.
Boundary markers: Absent in the instruction set for data processing.
Capability inventory: File system read/write, shell command execution (package installation, codegen), and environment variable modification.
Sanitization: Not explicitly defined, but the skill uses the extracted data to populate predefined code templates rather than directly executing it.
[CREDENTIALS_UNSAFE]: The skill explicitly includes a 'Two-client pattern' as a security rule to prevent sensitive credentials (SALEOR_APP_TOKEN) from being leaked to the browser. It correctly distinguishes between anonymous public clients and authenticated server-side clients.

storefront-builder