The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill retrieves external design guidance packages at runtime using npx ui-skills. This involves downloading code from the npm registry based on specific slugs or categories.
[COMMAND_EXECUTION]: The protocol directs the agent to execute shell commands, specifically npx ui-skills get <slug> and npx ui-skills list --category <category>, to fetch and process design lenses.
[DYNAMIC_EXECUTION]: The skill uses npx to execute the ui-skills package dynamically. Because the package content is fetched at runtime and executed on the host system, it presents a risk of executing unverified code.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes untrusted user-supplied data, including code snippets and screenshots.
Ingestion points: The skill reads "supplied screenshot(s)" in Step 1 and "targeted code" (components, styles, tokens) in Step 2.
Boundary markers: Absent; there are no instructions to use delimiters or ignore potential commands embedded within the user-provided code or image metadata.
Capability inventory: The skill has file system write permissions (creating directories and files in meta/design-critiques/), shell command execution capabilities via npx, and the ability to invoke other skills like generate-mockups.
Sanitization: Absent; the instructions do not specify any validation, filtering, or escaping of the user-provided code before it is analyzed by the agent.

design-critique