The Agent Skills Directory

[COMMAND_EXECUTION]: The anycap connect feishu command starts a background daemon that executes local sub-agents (Codex, Claude Code, or Cursor Agent). These agents are then controlled via remote messages from the Feishu (Lark) IM platform.
[COMMAND_EXECUTION]: The instructions recommend using high-privilege flags such as --codex-exec-mode danger-full-access, --claude-permission-mode bypassPermissions, and --cursor-force. These flags are designed to bypass tool-call confirmations and security sandboxes, allowing the remote chat platform to execute arbitrary shell commands on the host machine without human oversight.
[EXTERNAL_DOWNLOADS]: The installation process suggests downloading and executing a binary install script from the vendor's domain (https://anycap.ai/install.txt). Additionally, the CLI performs auto-updates from GitHub Releases at runtime.
[DATA_EXFILTRATION]: The skill manages sensitive Feishu credentials (App ID and App Secret). While it includes warnings against pasting these into chat, it stores them in a local configuration file (mode 0600) to facilitate the daemon's access.
[COMMAND_EXECUTION]: The installation guide describes modifying shell profile files (.bashrc, .zshrc) to persist changes to the system PATH variable.
[COMMAND_EXECUTION]: The 'Persist to Project Memory' section instructs the agent to modify project-level instruction files (e.g., CLAUDE.md, AGENTS.md) to ensure the AnyCap skill is prioritized and rediscovered in future sessions.

anycap-cli