The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to utilize a local CLI tool (quilium) to manage the development lifecycle. Commands include project initialization (init), running a local preview server (run), deploying templates (push), and managing translations (dictionary). These operations involve shell execution and local process management, such as identifying and terminating processes holding network ports.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external content which presents a vulnerability surface for indirect prompt injection.
Ingestion points: The agent ingests data from two primary external sources: dynamic skill manifests/procedures via the MCP server (get-skills, get-skill) and static HTML prototypes provided for porting into the CMS (File: SKILL.md, references/from-a-prototype.md).
Boundary markers: The documentation explicitly instructs the agent to treat visitor-controlled inputs as untrusted data (File: references/engine-liquid.md).
Capability inventory: The skill allows for local file system modification, execution of the quilium CLI, and administrative updates to CMS site settings and content types.
Sanitization: The instructions mandate the use of the Liquid escape filter for any variable controlled by a visitor or remote API to prevent injection during rendering.
[EXTERNAL_DOWNLOADS]: The skill implements a dynamic loading protocol where it fetches additional instructions and site-specific procedures from a remote MCP server at runtime. This mechanism is identified as the source of truth for schema definitions and API shapes.
[DATA_EXPOSURE]: The skill provides technical transparency regarding the local environment, specifically noting that the CLI tool stores its authentication tokens in the ~/.quilium/ directory (File: references/local-dev.md).

quilium-builder