# code > Execute Every Code CLI (just-every/code, a fork of codex-cli) for multi-agent code generation, consensus planning, browser automation, and parallel AI orchestration. Use when user explicitly requests Every Code CLI operations, multi-agent consensus (/plan, /solve, /code), or browser automation via Chrome DevTools. - Author: Tyce Herrman - Repository: ulama-tech/skill-codex - Version: 20251224094456 - Stars: 0 - Forks: 0 - Last Updated: 2026-02-07 - Source: https://github.com/ulama-tech/skill-codex - Web: https://mule.run/skillshub/@@ulama-tech/skill-codex~code:20251224094456 --- --- name: code description: Execute Every Code CLI (just-every/code, a fork of codex-cli) for multi-agent code generation, consensus planning, browser automation, and parallel AI orchestration. Use when user explicitly requests Every Code CLI operations, multi-agent consensus (/plan, /solve, /code), or browser automation via Chrome DevTools. allowed-tools: - Bash - Read - Grep - Glob --- # Every Code CLI Integration Skill This skill enables Claude Code to effectively orchestrate Every Code CLI (`code`) - the `just-every/code` package, a fork of OpenAI's Codex CLI with multi-agent consensus, browser integration, and enhanced automation features. ## Agent Handoff Spawn the dedicated agent for CLI execution: ``` Task(subagent_type: "code", prompt: "[task]") ``` This keeps CLI output isolated and returns concise summaries. ## When to Use This Skill ### Ideal Use Cases 1. **Multi-Agent Consensus** - `/plan` - Get Claude, Gemini, and GPT to collaboratively plan code changes - `/solve` - Race multiple AIs to solve complex problems (fastest wins) - `/code` - Generate code with multi-agent consensus 2. **Second AI Perspective** - Code review from a different AI's viewpoint - Bug detection Claude might have missed - Alternative implementation approaches 3. **Browser Integration** - `/chrome` - Connect to external Chrome via CDP - `/browser` - Use internal headless browser - Visual testing and screenshot comparisons 4. **Auto-Drive Mode** - `/auto` - Hands-off multi-step task coordination - Background task execution - Automated workflows 5. **Parallel Processing** - Offload tasks while continuing other work - Run multiple code generations simultaneously ### When NOT to Use - Simple, quick tasks (overhead not worth it) - Tasks requiring immediate response - When context is already loaded and understood - Interactive refinement requiring conversation ## Core Instructions ### 1. Verify Installation ```bash command -v code || which code ``` ### 2. Basic Command Patterns ```bash # Interactive mode with prompt code "explain this codebase" # Full-auto mode (no approval prompts) code --full-auto "create a todo app" # Non-interactive execution code exec "fix all TypeScript errors" # Pipe prompt via stdin echo "analyze security" | code exec ``` ### 3. Key CLI Flags | Flag | Description | | ------------------ | ---------------------------------------------------- | | `--full-auto` | Automatic execution, no approval prompts | | `--sandbox ` | `read-only`, `workspace-write`, `danger-full-access` | | `--model ` | Override default model (e.g., `gpt-5.2-codex`) | | `--config key=val` | Override config values | | `--read-only` | Prevent file modifications | | `-C, --cd ` | Change working directory | ### 4. Sandbox Modes | Mode | Use Case | | -------------------- | ------------------------------------ | | `read-only` | Analysis, review, research (default) | | `workspace-write` | Apply local edits | | `danger-full-access` | Network access, broad permissions | ### 5. Multi-Agent Commands These slash commands only work **inside an interactive `code` session**: ```bash # First start an interactive session: code # Then use these commands within the session: /plan "implement user authentication with OAuth" # Consensus planning /solve "why does deleting one user drop the database?" # Racing solve /code "add rate limiting to the API" # Multi-agent code generation /auto "refactor the authentication module and add tests" # Auto-drive ``` ### 6. Browser Integration (Interactive Session) These commands work inside an interactive `code` session: ```bash /chrome # Connect to external Chrome via CDP /browser # Use internal headless browser Ctrl+B # ASCII preview in terminal ``` ### 7. Session Management ```bash # Resume last session echo "continue with the refactor" | code exec resume --last # Resume with additional context echo "now add tests" | code exec resume --last ``` ## Quick Reference Commands ### Analysis (Read-Only) ```bash code --sandbox read-only "review this codebase for security issues" ``` ### Code Generation ```bash code --full-auto --sandbox workspace-write "create user authentication module" ``` ### Multi-Agent Planning (Interactive Session) ```bash # Inside a `code` session: /plan "design a microservices architecture for this monolith" ``` ### Problem Solving (Interactive Session) ```bash # Inside a `code` session: /solve "identify the root cause of the memory leak" ``` ### Browser Testing (Interactive Session) ```bash # Inside a `code` session: /browser # Then: "take a screenshot of the login page and check for accessibility issues" ``` ## Configuration ### Config File Location `~/.code/config.toml` ### Example Configuration ```toml model = "gpt-5.2-codex" approval_policy = "on-request" sandbox_mode = "workspace-write" [profiles.full_auto] approval_policy = "on-request" sandbox_mode = "workspace-write" [profiles.readonly] approval_policy = "never" sandbox_mode = "read-only" ``` ### Subagent Configuration ```toml [[subagents.commands]] name = "plan" read_only = true agents = ["claude-opus-4.5", "gemini", "gpt-5.2-codex"] [[subagents.commands]] name = "solve" read_only = true agents = ["claude-opus-4.5", "gemini", "gpt-5.2-codex"] ``` ## Error Handling ### Command Failures - Stop and report failures when `code` exits non-zero - Request direction before retrying ### High-Impact Operations Before using these flags, ask user for permission: - `--full-auto` - `--sandbox danger-full-access` ### Rate Limits - CLI auto-retries with backoff - Use appropriate models for task complexity ## Integration Workflow ### Standard Generate-Review Cycle ```bash # 1. Generate with multi-agent consensus (in interactive session) code # Start session /code "create REST API for user management" # 2. Review the generated code (new session or standalone) code --sandbox read-only "review the generated code for bugs and security" # 3. Fix identified issues code --full-auto --sandbox workspace-write "fix the issues identified" ``` ### Cross-Validation with Claude ```bash # 1. Claude generates code (using normal Claude Code tools) # 2. Every Code reviews code --sandbox read-only "review this code for issues: [paste code]" ``` ## Unique Capabilities These features are available only through Every Code: 1. **Multi-Agent Consensus** - Orchestrate Claude, Gemini, GPT together 2. **Auto-Drive Mode** - Automated multi-step workflows 3. **Browser Integration** - Chrome DevTools via CDP 4. **Auto-Review** - Background ghost-commit watcher 5. **Theme System** - Visual customization ## See Also - `reference.md` - Complete command and flag reference - `templates.md` - Prompt templates for common operations - `patterns.md` - Advanced integration patterns - `tools.md` - Built-in tools and subagent commands