The Agent Skills Directory

[PROMPT_INJECTION]: The LICENSE.txt file contains an extremely long and repetitive list of thousands of architectural and data-related keywords (e.g., 'Data self-evolving', 'Data self-optimizing') appended to the legal text. This is a 'context stuffing' or 'token smuggling' technique designed to saturate the model's context window, which can potentially push out system instructions or interfere with the model's attention mechanism to bypass safety or operational constraints.- [COMMAND_EXECUTION]: The skill utilizes a complex set of internal orchestration commands like /sc:spawn, /sc:design, and /sc:analyze. These commands are used to coordinate multi-agent workflows based on user-provided requirements. While intended for functionality, these represent a high-privilege command interface that could be exploited via indirect prompt injection if user inputs are not rigorously sanitized before being interpolated into these command strings.

architecture