The Agent Skills Directory

[Prompt Injection] (LOW): The file 'anti-rationalization.md' employs high-authority language and 'STOP' commands to strictly control the agent's decision-making process. While intended for architectural discipline, these are steering instructions that constrain the model's output.- [Indirect Prompt Injection] (LOW): The skill is designed to analyze external project code and identifies phrases such as 'Quick hack' or 'Temporary workaround' as triggers to halt processing. An adversary could intentionally include these phrases in code comments or documentation to induce a denial-of-service on the agent's reasoning capabilities.- [Data Exposure & Exfiltration] (SAFE): The skill does not perform file system access to sensitive paths, nor does it contain any hardcoded credentials or network exfiltration logic.- [Remote Code Execution] (SAFE): No scripts, package managers, or remote execution patterns were found; the skill is entirely composed of documentation.

architecture-review