The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/search-components.py and scripts/pull-all-registries.py to interface with the shadcn CLI and manage its local cache. These operations are used for legitimate functionality, such as searching registries and installing components via the standard npx toolchain.
[EXTERNAL_DOWNLOADS]: Fetches registry and component metadata from https://ui.shadcn.com. This is a well-known service for the frontend development ecosystem. The downloads are handled by the Python urllib module and are used to populate an offline search cache in the scripts/.cache/ directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its ingestion of large amounts of third-party data.
Ingestion points: Component names and descriptions are fetched from https://ui.shadcn.com/r/registries.json and individual registry endpoints in scripts/pull-all-registries.py (line 118).
Boundary markers: The skill does not currently use explicit boundary markers or "ignore embedded instructions" warnings when presenting component data to the agent in scripts/search-components.py.
Capability inventory: The agent has the capability to execute shell commands and write to the filesystem through the npx shadcn tool via the provided search scripts.
Sanitization: While pull-all-registries.py (lines 25-115) includes extensive SKIP_KEYWORDS and SKIP_REGISTRIES filters to exclude low-quality or irrelevant content, it does not specifically sanitize for malicious prompt instructions hidden in registry metadata.

@tank/frontend-craft