The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several internal bash scripts (detect-mode.sh, check-status.sh, write-alias.sh, verify-providers.sh) to inspect environment states, safely modify shell configuration files, and verify the validity of provider credentials.
[EXTERNAL_DOWNLOADS]: Connectivity tests and metadata retrieval are performed using curl against official API endpoints of well-known AI services, including DeepSeek, ZhipuAI (Z.ai), Alibaba Cloud (DashScope), MiniMax, and OpenRouter.
[CREDENTIALS_UNSAFE]: The skill is designed to manage and store API keys for alternative LLM providers within the user's ~/.zshrc file. Security risks are mitigated by ensuring keys are passed via standard input (preventing exposure in shell history or process lists), creating restricted configuration backups (chmod 600), and providing explicit instructions to the agent not to commit these changes to version control.
[INDIRECT_PROMPT_INJECTION]: The maintenance mode spaws subagents that research provider documentation from the web. This attack surface is mitigated by strict instruction sets that limit the subagents to a 'research and report' role and explicitly prohibit them from modifying any files on the system.

provider-switch