The Agent Skills Directory

[SAFE]: The skill implements a transparent, multi-stage workflow (Initialization, Gap Analysis, Issue Detection, Improvements, Compilation, and Reporting) that operates on local documentation and metadata. No suspicious or hidden behaviors were identified.
[PROMPT_INJECTION]: The skill uses instructional language to maintain workflow integrity ('Follow the mandatory sequence', 'Never speculate') without attempting to override the agent's core safety guidelines or switch to unrestricted modes.
[DATA_EXFILTRATION]: There are no network operations detected. The skill reads local architecture documents and skill metadata, and writes refined output files to a user-defined output directory. It does not attempt to access sensitive system files like SSH keys or environment secrets.
[COMMAND_EXECUTION]: No hazardous shell commands or privileged operations are invoked. The skill's 'subagent delegation' for parallel file processing is a platform-specific feature for efficient analysis and does not involve arbitrary code execution.
[REMOTE_CODE_EXECUTION]: No patterns of downloading external scripts or executing unverified remote content were found. All dependencies are handled through the agent's internal context or standard project-relative paths.
[PROMPT_INJECTION]: While the skill processes untrusted user-provided architecture documents (an indirect prompt injection surface), it utilizes specific callout markers such as [!WARNING] and [!NOTE] to distinguish its findings from original content, which serves as a basic boundary mechanism. The risk is assessed as low given the skill's primary function of documentation refinement.

skf-refine-architecture