The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from both local project files and external web pages to perform its critique, creating a vulnerability surface where malicious instructions embedded in a design could influence the agent's behavior.
Ingestion points: The skill uses the WebFetch tool to retrieve content from external URLs and the Read, Glob, and Grep tools to access local source code and design component files.
Boundary markers: The instructions do not define clear delimiters or provide "ignore embedded instructions" warnings to help the agent distinguish between data to be critiqued and potential malicious prompts within that data.
Capability inventory: The skill possesses significant capabilities, including reading local files and performing network requests to arbitrary domains.
Sanitization: No content validation, escaping, or sanitization routines are implemented for the external or local data processed by the skill.
[DATA_EXFILTRATION]: The skill is configured to use network-enabled tools such as WebFetch and WebSearch to interact with domains outside of the standard whitelist. While this is necessary for its stated purpose of auditing live websites, it establishes a communication channel to non-whitelisted external servers that could be misused.

design-critic