The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the shell to execute the uspec-skills CLI tool via node and npx at multiple stages, including preparation, validation, and rendering of component artifacts.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to invoke npx uspec-skills, which potentially triggers the download of the package from the external NPM registry if the tool is not already available in the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a _base.json file (produced by an external Figma plugin) and optionalContext provided by the user. These inputs are passed to parallel subagents and integrated into the final generated documentation, which can be exploited to influence agent behavior.
Ingestion points: The _base.json file (read from disk) and the optionalContext string parameter.
Boundary markers: There are no explicit instructions or delimiters defined to isolate the untrusted data from the execution instructions in the subagent prompts.
Capability inventory: The skill has the capability to execute shell commands, write files to the workspace, and dispatch subagents with dynamic prompts.
Sanitization: Although the skill includes a schema validation step for the cache files, it lacks specific sanitization of the natural language content to prevent injection attacks.

create-component-md