The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill instructions include commands to verify the setup by echoing the API key to the console (echo $SHADCNBLOCKS_API_KEY or $env:SHADCNBLOCKS_API_KEY). This practice exposes sensitive credentials in the process output and history logs.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The workflow involves downloading and executing remote component code using the npx shadcn@latest add command from https://shadcnblocks.com/r/[block-id]. While this is a standard pattern for the Shadcn ecosystem and targets the vendor's domain, it represents the execution of remote code at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill ingests UI layouts and component metadata from an external API, which could be exploited to deliver malicious instructions to the agent if the source is compromised.
Ingestion points: Metadata and component identifiers are retrieved from api.shadcnblocks.com/v1/blocks and the component registry at shadcnblocks.com.
Boundary markers: There are no explicit delimiters or specific instructions for the agent to ignore potentially malicious embedded content within the fetched component metadata.
Capability inventory: The skill possesses the ability to execute shell commands (npx, curl), write files to the local filesystem (component installation), and perform network requests.
Sanitization: The skill mitigates this risk by instructing the agent to evaluate candidates based on project style and accessibility (Step 3) and to manually adapt the code to project tokens and routes (Step 5) before final verification.

shadcn-blocks