The Agent Skills Directory

[COMMAND_EXECUTION]: The skill primarily operates by executing the 21st CLI (@21st-dev/cli) to interact with the 21st.dev registry. This involves commands for publishing, editing, and deleting items, as well as managing team libraries and profile configurations. This is the intended functionality of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to read user-provided code files (e.g., .tsx components, .css themes) to automatically detect names, slugs, and tags for publication. This ingestion of untrusted data from the filesystem into the agent's context constitutes a capability surface for indirect prompt injection.
Ingestion points: Local file paths for components, themes, and JSON profile data (board.json).
Boundary markers: The prompt instructions do not specify the use of delimiters or warnings to ignore embedded instructions within the ingested file content.
Capability inventory: The skill has access to read local files and execute shell commands via the 21st CLI.
Sanitization: No sanitization or validation logic is defined for the metadata auto-detection process.
[CREDENTIALS_UNSAFE]: While the skill involves managing API keys for authentication, it explicitly includes a security rule instructing the agent to never publish files containing API keys or secrets. This proactive check reduces the risk of accidental data exposure.

21st-registry