The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Bash tool to recompile CSS variables and execute automated tests (referred to as "rendered gates") after design modifications. These actions are standard for a design-to-code workflow. Evidence: The instructions specify that the agent should "recompile and map the bundle's scoped CSS variables" and "run the affected code and rendered gates" using shell access.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from local project files to drive its design logic, which represents a potential injection surface. However, the skill implements strict logic ramps and guardrails to limit how this data influences the output. Ingestion points: The skill reads from .styleseed/project.json, .styleseed/artifacts/index.json, and associated markdown/manifest files. Boundary markers: The skill logic enforces strict resolution of artifact boundaries, reading only specific files related to the target artifact ID. Capability inventory: The skill has access to Bash, Write, Edit, Read, Grep, and Glob tools. Sanitization: The skill does not explicitly detail sanitization but relies on a deterministic ramp and predefined ramps for axis changes.

ss-dial