The Agent Skills Directory

[NO_CODE]: The skill consists entirely of instructional markdown and metadata. It does not include any scripts, executables, or external software dependencies that could execute malicious logic.
[DATA_EXPOSURE_&_EXFILTRATION]: The skill reads from a local directory (clients/<client>/personas/personas.json). This is a documented, structured use of local data intended for the skill's primary function. It does not attempt to access sensitive system files (e.g., SSH keys, credentials, or environment files).
[EXTERNAL_DOWNLOADS]: The skill utilizes WebFetch and WebSearch to gather information from target URLs provided by the user. These operations are performed at the agent level using standard platform tools, which is the intended purpose of a website review tool.
[INDIRECT_PROMPT_INJECTION]: The skill processes external website content, which represents an ingestion surface for untrusted data. However, the instructions focus on analytical tasks (scoring, focus group reactions) rather than executing commands based on that content, and it lacks dangerous output capabilities (like file writing or command execution based on crawled data) that would escalate the severity.
[COMMAND_EXECUTION]: No shell commands or system-level execution patterns were identified in the instructions.

icp-website-review