The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the 21st CLI tool (in SKILL.md) to initialize project context, search for UI references, and generate variant implementations. These operations are core to the skill's purpose and utilize the vendor's own infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill processes project-specific data and tool outputs, creating a surface where malicious instructions could be ingested into the agent context.
Ingestion points: Data is read from .21st/design.json and the output of the 21st search command in SKILL.md.
Boundary markers: No specific delimiters or "ignore instructions" prompts are used to isolate ingested data from agent instructions.
Capability inventory: The agent can execute shell commands (21st tool) and write data to the local filesystem (.21st/design.json) as described in the workflow of SKILL.md.
Sanitization: The workflow does not include steps to validate or sanitize content from design files or search results before interpolation.

21st-ui-explore