# Implementation Plan: OpenCode Claude Sync Enhancements

## Overview

Transpose Claude Code agent/skill setup to OpenCode in parallel, per decisions from brainstorming session (`enumerated-giggling-scone.md`).

## Key Decisions (from brainstorming)

| Decision | Value |
|----------|-------|
| Primary agent | Use built-in `build` (don't port PA) |
| Agents to skip | `personal-assistant`, `master-orchestrator` |
| Other agents | All become `mode: subagent` |
| Model inheritance | Use `model: inherit` for subagents |
| State sharing | Reference via `instructions`, don't copy |
| Source of truth | Claude Code (`~/.claude/`) |

## Files to Modify

1. `~/.config/opencode/scripts/claude_sync.py` - Main sync script
2. `~/.config/opencode/opencode.json` - Config file

## Files to Create

1. `~/.config/opencode/README.md` - Documentation

## Implementation Steps

### Step 1: Backup (DONE)

Created backups:
- `~/.config/opencode-backup-20260107_120135.tar.gz`
- `~/opencode-home-backup-20260107_120136.tar.gz`

### Step 2: Enhance `claude_sync.py`

**Location**: `~/.config/opencode/scripts/claude_sync.py`

**Modifications**:

1. Add constants near top of file:
```python
SKIP_AGENTS = {"personal-assistant", "master-orchestrator"}

MODEL_MAP = {
    "opus": "anthropic/claude-opus-4",
    "sonnet": "anthropic/claude-sonnet-4-5",
    "haiku": "anthropic/claude-haiku-4-5",
}
```

2. Modify `transform_frontmatter()` for agents:
   - Check if agent name in `SKIP_AGENTS`, return `None` to signal skip
   - Add `frontmatter["mode"] = "subagent"`
   - Set `frontmatter["model"] = "inherit"`
   - Map explicit models using `MODEL_MAP`

3. Modify `sync_tree()` to handle `None` return from transform (skip file)

4. Update `expected_dest_paths_for_tree()` to exclude skipped agents

### Step 3: Run Sync

```bash
python3 ~/.config/opencode/scripts/claude_sync.py --dry-run
python3 ~/.config/opencode/scripts/claude_sync.py
python3 ~/.config/opencode/scripts/claude_sync.py --clean --apply
```

### Step 4: Update `opencode.json`

Add to existing config:
```json
{
  "model": "anthropic/claude-sonnet-4-5",
  "small_model": "anthropic/claude-haiku-4-5",
  "instructions": [
    "~/.claude/CLAUDE.md",
    "~/.claude/state/kb.json",
    "~/.claude/state/personal-assistant/memory/facts.json",
    "~/.claude/state/personal-assistant/memory/preferences.json"
  ],
  "permission": {
    "edit": "ask",
    "bash": {
      "*": "ask",
      "pacman -Q*": "allow",
      "systemctl status*": "allow",
      "kubectl get*": "allow"
    }
  }
}
```

### Step 5: Test

- Run `opencode` and verify skill discovery
- Test `@linux-sysadmin` subagent invocation
- Verify permissions work

### Step 6: Create README.md

Document:
- Architecture (Claude Code as source of truth)
- Sync workflow
- Agent mapping table
- How to invoke subagents

### Step 7: Add Future Consideration

Add entry to `~/.claude/state/future-considerations.json` about JSON minification for large instruction files.

## Estimated Time

~2 hours total (Step 1 already done)