The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Step 2 of the skill executes npx ui-craft-detect. This triggers the download and execution of a package from the public NPM registry without a specified version or integrity hash, posing a supply chain risk through unverifiable remote code execution.
[COMMAND_EXECUTION]: The skill uses npx to spawn a subprocess for running the detection tool, which interacts with the project files and user-defined targets.
[EXTERNAL_DOWNLOADS]: The use of npx necessitates fetching code from the external NPM registry at runtime.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in Step 1 and Step 4.
Ingestion points: It reads the contents of .ui-craft/brief.md to extract principles (overrides).
Boundary markers: There are no boundary markers or instructions to treat the ingested data as untrusted.
Capability inventory: The skill possesses the ability to execute shell commands (npx) and read various project files, providing a path for malicious instructions in the brief to influence agent operations.
Sanitization: The extracted principles are used directly to override findings without validation or escaping, allowing potentially malicious instructions in the brief to suppress legitimate security or quality warnings.

finalize