Execute - Multi-Model Collaborative Execution

Prerequisites

Requires codeagent-wrapper binary. Verify:

ls ~/.claude/bin/codeagent-wrapper 2>/dev/null && echo "OK" || echo "MISSING"

If codeagent-wrapper is MISSING: stop immediately — use /plan + /tdd instead. See codeagent installation to install.

When to Use This vs /tdd

Use /multi-execute when	Use /tdd instead when
codeagent-wrapper is installed and configured	You don't have codeagent-wrapper
You have a /multi-plan output with SESSION_IDs	You're starting fresh without a prior plan
You want Codex/Gemini prototypes refactored by Claude	Native TDD is sufficient for the task
Multi-model audit (Codex + Gemini review) is desired	Single reviewer or speed matters more

Multi-model collaborative execution — get prototype from plan → Claude refactors and implements → multi-model audit and delivery.

$ARGUMENTS

---

Phases at a Glance

Phase	Name	What Happens
0	Read Plan	Parse plan file, confirm SESSION_ID, route by task type
1	Context Retrieval	MCP tool fetches relevant code context
3	Prototype Acquisition	Codex (backend) / Gemini (frontend) returns Unified Diff Patch
4	Code Implementation	Claude refactors dirty prototype → production-grade code
5	Audit & Delivery	Parallel Codex+Gemini review → integrate fixes → delivery report

Phase 2 is intentionally reserved. See Reference for call syntax and model parameters.

---

Execution Workflow

Phase 0: Read Plan

Parse the plan file (or direct task description), extract SESSION_ID and key files, and route to the correct model based on task type: Frontend → Gemini, Backend → Codex, Fullstack → both in parallel.

Phase 1: Context Retrieval

Call mcp__ace-tool__search_context with a semantic query built from the plan's key files — never explore the filesystem manually with find/ls.

Phase 3: Prototype Acquisition

Invoke Codex (backend) or Gemini (frontend) via codeagent-wrapper, reusing SESSION_ID from the plan where available; each model returns a Unified Diff Patch only — no direct file writes.

Phase 4: Code Implementation

Claude applies the diff as code sovereign: simulate in a mental sandbox, refactor the dirty prototype to production-grade code, apply changes with Edit/Write tools, then run lint/typecheck/tests to catch regressions before proceeding.

Phase 5: Audit and Delivery

Parallel Codex + Gemini code review (run_in_background: true); synthesize feedback weighted by domain trust (Codex for backend, Gemini for frontend); apply fixes; repeat until risk is acceptable; report a change summary + audit results to the user.

---

Core Protocols

• Language Protocol: Use English when interacting with tools/models, communicate with user in their language
• Code Sovereignty: External models have zero filesystem write access, all modifications by Claude
• Dirty Prototype Refactoring: Treat Codex/Gemini Unified Diff as "dirty prototype", must refactor to production-grade code
• Stop-Loss Mechanism: Do not proceed to next phase until current phase output is validated
• Prerequisite: Only execute after user explicitly replies "Y" to /ccg:plan output (if missing, must confirm first)

---

Key Rules

1. Code Sovereignty – All file modifications by Claude, external models have zero write access
2. Dirty Prototype Refactoring – Codex/Gemini output treated as draft, must refactor
3. Trust Rules – Backend follows Codex, Frontend follows Gemini
4. Minimal Changes – Only modify necessary code, no side effects
5. Mandatory Audit – Must perform multi-model Code Review after changes

---

Usage

# Execute plan file
/ccg:execute .claude/plan/feature-name.md

# Execute task directly (for plans already discussed in context)
/ccg:execute implement user authentication based on previous plan

---

Relationship with /ccg:plan

1. /ccg:plan generates plan + SESSION_ID
2. User confirms with "Y"
3. /ccg:execute reads plan, reuses SESSION_ID, executes implementation

After This

• /code-review — verify the refactored implementation quality
• /tdd — add missing test coverage for the delivered changes
• /verify — run full build + type-check + tests to confirm clean delivery

---

Reference

Technical call syntax — consult when debugging or customising model invocations.

Multi-Model Call Syntax

Implementation prototype (run_in_background: true):

Bash({
  command: "~/.claude/bin/codeagent-wrapper {{LITE_MODE_FLAG}}--backend <codex|gemini> {{GEMINI_MODEL_FLAG}}resume <SESSION_ID> - \"$PWD\" <<'EOF'
ROLE_FILE: <role prompt path>
<TASK>
Requirement: <description>
Context: <plan + key files>
</TASK>
OUTPUT: Unified Diff Patch ONLY. Strictly prohibit any actual modifications.
EOF",
  run_in_background: true, timeout: 3600000
})

Audit call (same shape, different ROLE_FILE + Scope in TASK body).

Model flags: {{GEMINI_MODEL_FLAG}} = --gemini-model gemini-3-pro-preview (trailing space) for Gemini; empty for Codex.

Wait for result: TaskOutput({ task_id: "<id>", block: true, timeout: 600000 }) — never kill; poll if needed.

Role prompts:

Phase	Codex	Gemini
Implementation	~/.claude/.ccg/prompts/codex/architect.md	~/.claude/.ccg/prompts/gemini/frontend.md
Review	~/.claude/.ccg/prompts/codex/reviewer.md	~/.claude/.ccg/prompts/gemini/reviewer.md

Delivery Report Template

## Execution Complete

### Change Summary
| File | Operation | Description |
|------|-----------|-------------|
| path/to/file.ts | Modified | Description |

### Audit Results
- Codex: <Passed/Found N issues>
- Gemini: <Passed/Found N issues>

### Recommendations
1. [ ] <Suggested test steps>
2. [ ] <Suggested verification steps>