The Agent Skills Directory

[SAFE]: The skill instructions define a reasoning framework for UI analysis and design without introducing malicious patterns.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external data, such as live source code and visual references, creating an attack surface for indirect prompt injection.
Ingestion points: The 'Recover' process in SKILL.md instructs the agent to inspect route entries, composition-owning components, and rendered results.
Boundary markers: No specific delimiters or warnings for embedded instructions are defined in the output format.
Capability inventory: The skill primarily uses read-only tools for source inspection; it explicitly discourages creating or updating files unless requested by the user.
Sanitization: No sanitization or filtering of external content is described.
[REMOTE_CODE_EXECUTION]: The skill explicitly instructs the agent not to introduce new scanners, registries, or dependencies to produce the UI description, which reduces the risk of supply chain or remote execution attacks.

ui-grammar