bankai-cli
v0.6.16
Published
Launch coding agent CLIs with approval-bypass flags
Maintainers
Readme
bankai
▗▄▄▄▖
████▖▄▄▄▄ ▄▄▄
▗█▙▄▄▄▄▄▄▄▄▄▄▄▄▄ ▗▄▄ ▗▟██████████▄▝██████████████▙▄
▝▜█████████████▀▘ ▝████▖ ▗▟██▛▘ ▗███▀▘ ▀▀▜██▛▘ ▐███▀▀▘
▜██▛ ▐███ ▄▟██▛▘ ▗███▘ ▐██▌ ▐███
▐██▌ ▐███ ▀▀▀▘ ▗███▙▄▄▄▖ ▗██▛▗▄ ▗███▌
▐██▌ ▐███ ▗▟█████████▀▀█████▖▗██▘ ▝██████
▐██▌ ███▌ ▝███▌ ██▌ ████▀▄▟█ ▜███▀▀▀
▗▄▄▄▄▄▄▄▄▄▄▄▟██▙▄▄▄▄▄▄█████ ▐██▙▄▄██▙▄▄███▌ ▐██▙▄▄███▙▄▄██▙▖
█████████▀▀▀▜██▛▀▀▀▀▀▀▀▀▜██ ▝███▀▀██▛▀▀███ ▟██▀▀▀▜██▛▀▀▀▀▀▘
███▌ ▐██▌ ▐██▌ ██▌ ███ ▟██▘ ▐██▌
███▌ ▐██▌ ▐██▙▄▄████▙███▗██▄▄▄▄▄▟█████████▄
███▌ ▐██▌ ██▛▀▀▀▀ ▐██▌▀███▛▀▀▜██▛▀▀▀▀▀▀▀
███▌ ▐██▌ ▗██▘ ▐██▌ ▐██▌
▗███▘ ▗███▙▄▄▄▄▄▄▄▄▄▄ ▟█▛ ▗▄▄▖ ███▌ ▝██▌
▐███ ▜███████████████▙ ▗██▘ ▀███████▌ ██▌
▀▀ ▀▀ ▀▀▘ ▝▀████▘ ██▘
CLI tool that launches coding agent CLIs with approval-bypass flags.
Requirements
- Node.js >= 20.12.0
Install
npm install -g bankai-cli
# or: pnpm add -g bankai-cli
# or: bun install -g bankai-cliUsage
# Launch a specific agent with bypass flags
bankai claude
# Pass args directly to the target agent
bankai codex -C /path/to/project
# Interactive agent picker
bankai
# List all supported agents
bankai agents
# List only agents installed on your system
bankai agents --installedSupported Agents
CLI Agents (flag output)
| Agent | Command | Docs |
| ------------------ | -------------------------------------------------- | --------------------------------------------------------------------------------------------------------------------- |
| Claude Code | claude --dangerously-skip-permissions | Settings - Claude Code Docs |
| Codex CLI | codex --dangerously-bypass-approvals-and-sandbox | CLI Reference - OpenAI Codex |
| GitHub Copilot CLI | copilot --allow-all-tools | Copilot CLI Docs |
| Gemini CLI | gemini --yolo | Configuration - Gemini CLI |
| OpenHands | openhands --always-approve | CLI Mode - OpenHands Docs |
| Aider | aider --yes-always | Options Reference - aider |
| Qwen Code | qwen-code --yolo | Approval Mode - Qwen Code Docs |
| Kimi Code | kimi --yolo | Interaction Guide - Kimi Code Docs |
Settings Agents (config file / DB modification)
| Agent | Target | Description |
| ---------------- | ------------------------------------------------ | --------------------------------------------------------------------- |
| Claude Code (GLM)| ~/.claude/settings.json | bankai claude-glm — Claude Code on Z.AI's GLM models (see Env Vars) |
| Cursor Agent CLI | .cursor/cli.json / ~/.cursor/cli-config.json | Writes permission allow-list for Cursor Agent CLI |
| Cursor IDE | SQLite DB (state.vscdb) | Applies settings below, then launches Cursor |
Cursor IDE (bankai cursor)
Unlike CLI agents that pass a flag, Cursor IDE stores its settings in a SQLite DB. bankai cursor modifies the DB directly to apply the following, then opens Cursor:
| Setting | Effect | | -------------------------------------------- | --------------------------------------------------- | | Auto-Run Mode → Run Everything (Unsandboxed) | Agent runs all commands without sandboxing | | Browser Protection → OFF | Agent can run browser tools automatically | | MCP Tools Protection → OFF | Agent can run MCP tools automatically | | File-Deletion Protection → OFF | Agent can delete files automatically | | External-File Protection → OFF | Agent can create/modify files outside the workspace | | Dot-files Protection → OFF | Agent can modify dotfiles (.env, etc.) |
Cursor must be restarted after the first apply for changes to take effect.
Custom Agents
Register agents not in the built-in list:
# Non-interactive
bankai add --cmd opencode --line "opencode --yolo"
# Interactive
bankai add
# Edit an existing custom agent
bankai edit opencode
# Remove a custom agent
bankai remove opencodeCustom agents are stored in ~/.config/bankai/agents.json (XDG-compliant, varies by OS).
Env Vars
Any agent (built-in or custom) may define an env map that is injected into the
agent process at launch — useful for pointing a tool at an alternate, API-compatible
backend without editing its config files or maintaining shell aliases.
Values support ${VAR} interpolation against your current environment, so secrets
never need to be written into the agent definition. A ${VAR} that is unset
resolves to an empty string and prints a warning. Literal values pass through unchanged.
Claude Code on GLM (Z.AI)
bankai claude-glm launches Claude Code against Z.AI's Anthropic-compatible
endpoint, mapping the Claude tiers to GLM models:
{
"ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
"ANTHROPIC_AUTH_TOKEN": "${ZAI_API_KEY}", // read from your shell, not stored here
"API_TIMEOUT_MS": "3000000",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-4.5-air",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5-turbo",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.1"
}Set your key once, then run:
export ZAI_API_KEY="your-z.ai-api-key" # or source it from a secret manager
bankai claude-glmTo use different models, register a custom agent that overrides claude-glm with your
own env map. The same pattern works for any Anthropic-compatible provider (Kimi,
DeepSeek, OpenRouter, a local proxy, …) — just change the base URL, token, and models.
Development
bun install
bun run dev -- claude # Run from source
bun run build # Build to dist/
bun run test # Run tests