The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The inject-brand-context.cjs script reads markdown files such as docs/brand-guidelines.md to construct system prompt additions. This is a functional feature for brand adherence but creates a surface where instructions in project files can influence agent behavior.
Ingestion points: docs/brand-guidelines.md processed via inject-brand-context.cjs.
Boundary markers: The prompt content is grouped under a 'BRAND CONTEXT:' header, but lacks explicit guardrails against embedded directives.
Capability inventory: File system reads (fs) and execution of local scripts (execSync).
Sanitization: The script uses regex to select specific sections (e.g., color tables), which restricts the data to expected structures.
[COMMAND_EXECUTION]: The extract-colors.cjs script generates command strings for the magick CLI tool. It uses double-quote wrapping for the image path, though users and agents should be cautious of paths containing shell metacharacters in environments where these are evaluated.
[DYNAMIC_EXECUTION]: The sync-brand-to-tokens.cjs utility utilizes child_process.execSync to run an internal design system script at a hardcoded location within the project. This is a secure use of dynamic execution for build-time synchronization.

brand