Add OpenCode integration and CLI-only mode support

[benjaminwestern]

Summary

I worked with KimiK2.5 and Opus4.6 to co-develop this PR.

This commit adds OpenCode as a supported AI provider alongside Anthropic and OpenAI, enabling users to use TinyClaw without requiring Discord, Telegram, or WhatsApp.

Changes

Key Features:

• Add OpenCode provider with full agent and skill support
• Implement CLI-only mode with --no-tmux flag for headless operation
• Add --cli-only quick setup flag for streamlined onboarding without channels
• Fix tmux window targeting to support user configs with base-index 1
• Add send, receive, and chat commands for CLI-based interaction
• Create .opencode directory structure with skill symlinks per agent
• Support 6 pre-configured agent roles (build, plan, coder, reviewer, writer, assistant)

Technical Changes:

• lib/common.sh: Support empty channels array for CLI-only configs
• lib/daemon.sh: Add no-tmux mode and dynamic window targeting
• lib/messaging.sh: Add send, receive, and chat functions
• lib/setup-wizard.sh: Add --cli-only flag for quick setup
• src/lib/agent.ts: Create .opencode structure and per-agent config
• src/lib/invoke.ts: Add OpenCode CLI invocation with --agent flag
• tinyclaw.sh: Add command routing for new flags
• opencode.json: Add root agent configuration

Documentation:

• Add docs/OPENCODE.md: OpenCode integration guide
• Add docs/5-AGENT-SQUAD.md: Multi-agent team setup guide
• Add docs/CLI.md: Complete CLI usage documentation
• Update AGENTS.md: Add OpenCode-compatible team mention syntax
• Update .gitignore: Exclude .opencode runtime directory

---

Type of Change

• Bug fix
• [ X] New feature
• Enhancement to existing feature
• Breaking change
• Documentation update
• Refactor / code cleanup

Testing

Tested locally using OpenCode + CLI-only mode:

• Verified CLI-only setup (setup --cli-only)
• Tested no-tmux daemon operation (start --no-tmux)
• Confirmed message sending/receiving via API
• Validated OpenCode provider integration (sonnet model)
• Tested file creation and code generation
• Verified response queue functionality
  Note: Multi-agent teams, tmux mode, and skill invocation not fully tested.

I think there is a Low chance of regression as changes primarily add new code rather than modifying existing functionality. Tmux targeting change improves handling of non-default base-index configs.

---

Checklist

• [ X] I have tested these changes locally
• [X ] My changes don't introduce new warnings or errors ( I have attempted to regression test in a number of different ways, but I apologise for regression if there is any)
• [ X] I have updated documentation if needed