Plan

Overview

Use CLI subagents to create a multi-perspective implementation plan, then synthesize into a single plan file under plans/.

Inputs

• Feature description
• Optional constraints

Workflow

1. Ensure plans/ and plans/context/ exist.
2. Run the planning subagent commands below.
3. Synthesize the best elements into one plan.
4. Save to plans/<kebab-case-name>.md.

Planning Subagent Commands

Codex (detailed plan)

codex exec -m gpt-5.2 -c model_reasoning_effort="high" -c thinking="enabled" --skip-git-repo-check \
  "Create a detailed implementation plan for: {FEATURE_DESC}. Include steps, files, and tests."

Gemini (architecture and risks)

CLOUDSDK_CORE_PROJECT="" GOOGLE_CLOUD_PROJECT="" GCLOUD_PROJECT="" GEMINI_API_KEY=${GEMINI_API_KEY} \
  gemini -m gemini-3-pro-preview -o text "Create a detailed plan for: {FEATURE_DESC}. Include architecture, steps, and risks."

Claude (tradeoffs)

claude --model opus -p "Create a detailed plan for: {FEATURE_DESC}. Focus on tradeoffs, risks, and testing."

Plan Template

# {Feature Name}

## Overview

## Requirements

## Architecture

## Implementation Steps

## Testing Strategy

## Risks and Considerations

Output

• Plan file path and a short synthesis summary