Background Planner

Multi-LLM parallel planning with context-safe auto-save.

Quick Start

# 1. Parse topic and perspectives
# 2. Create: .context/plans/{timestamp}_{topic}/
# 3. Run agents in background
# 4. Each agent saves result to {agent_name}.md
# 5. Guide user to merge when ready

Provider Selection

Workflow

Step 1: Setup

mkdir -p .context/plans/{timestamp}_{topic}
# Initialize status.json with agent list

Step 2: Run Agents

Claude:

Task({
  subagent_type: "general-purpose",
  prompt: `기획 주제: ${topic}
관점: ${perspective}
결과 저장: .context/plans/{session}/claude.md`,
  run_in_background: true
})

Codex:

codex --approval-mode full-auto \
  "Plan ${topic} from technical perspective. Save to .context/plans/{session}/codex.md" &

Gemini:

gemini -m gemini-2.0-flash \
  "Plan ${topic} creatively. Save to .context/plans/{session}/gemini.md" &

Step 3: Guide User (NO POLLING)

## Planning Agents Running

Check results:
- `ls .context/plans/{session}/*.md`
- `cat .context/plans/{session}/status.json`

When ready, ask me to "머지해줘" or "결과 확인"

Step 4: Merge (on request)

Read all {agent}.md files and create merged.md:

• Compare perspectives
• Highlight agreements/conflicts
• Synthesize final recommendation

Output Structure

.context/plans/{timestamp}_{topic}/
├── status.json
├── claude.md
├── codex.md
├── gemini.md
└── merged.md

Best Practices

DO:

• Use 2-4 agents for diverse perspectives
• Let each agent save directly to file
• Wait for user to request merge

DON'T:

• Poll for completion (token waste)
• Run 5+ agents (diminishing returns)
• Merge before all agents complete