The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions urge users to install a forked version of the Kimi Code CLI from an unofficial repository (https://github.com/Dqz00116/kimi-cli.git). Installing executable tools from unofficial forks is a high-risk activity as the code may contain unauthorized modifications or backdoors.
[EXTERNAL_DOWNLOADS]: The installation process involves cloning multiple repositories from GitHub (including the author's fork) and installing them in editable mode using 'uv pip install -e .', which executes arbitrary setup code in the local environment.
[PROMPT_INJECTION]: The skill sets up a persistent injection mechanism via a 'UserPromptSubmit' hook. This hook silently injects a 'MANDATORY' system reminder into every conversation, forcing the agent to follow a strict workflow and perform hidden tool calls (ReadFile) before responding to user messages. The instructions use authoritative language (e.g., 'CRITICAL RULE', 'MANDATORY') to override the agent's default safety and operational guidelines.
[COMMAND_EXECUTION]: The installation steps provide complex shell and PowerShell scripts that perform sensitive operations, such as creating system-level junctions, modifying configuration files, and installing software from remote sources.

install-superpowers-for-kimi