CEO Setup

You are running the CEO plugin setup wizard. Walk the user through configuring their preferences and write the result to ${CLAUDE_PLUGIN_ROOT}/settings.json.

Step 1: Welcome

Say:

Welcome to CEO Plugin Setup

I'll walk you through a few preferences that control how the CEO orchestrates your agents. This takes about 1 minute. You can re-run /ceo:setup anytime to change these.

Check if ${CLAUDE_PLUGIN_ROOT}/settings.json already exists. If it does, read it and say:

"You already have a settings file. I'll show your current choices — just confirm or change them."

Step 2: Model Tier

Use AskUserQuestion:

{
  "questions": [
    {
      "question": "How should the CEO assign models to agents? This is the biggest cost/quality tradeoff.",
      "header": "Model Tier",
      "options": [
        {
          "label": "Balanced (Recommended)",
          "description": "Opus for architects, reviewers, strategists. Sonnet for developers, writers, operators. Best tradeoff of quality and speed."
        },
        {
          "label": "Max Quality",
          "description": "ALL agents run on Opus. Best reasoning across the board, but slower and more expensive."
        },
        {
          "label": "Max Speed",
          "description": "ALL agents run on Sonnet. Fastest and cheapest. May miss nuance on complex architecture or review tasks."
        }
      ],
      "multiSelect": false
    }
  ]
}

Map the answer:

• "Balanced" → "balanced"
• "Max Quality" → "max_quality"
• "Max Speed" → "max_speed"

Step 3: Verification & Teams

Use AskUserQuestion with two questions:

{
  "questions": [
    {
      "question": "Should every implementation task be reviewed by a read-only agent before marking complete? (Catches bugs but adds time per task)",
      "header": "Verify Loop",
      "options": [
        {
          "label": "Yes, always verify (Recommended)",
          "description": "A reviewer checks every deliverable. Max 3 retry cycles before escalating to you."
        },
        {
          "label": "Yes, but max 1 retry",
          "description": "Verify once. If it fails, escalate to you immediately instead of looping."
        },
        {
          "label": "No, skip verification",
          "description": "Trust agent output without review. Faster but riskier. Not recommended for production work."
        }
      ],
      "multiSelect": false
    },
    {
      "question": "How should the CEO coordinate agents that need to work together (e.g., frontend + backend on the same API)?",
      "header": "Team Mode",
      "options": [
        {
          "label": "Auto (Recommended)",
          "description": "Use team messaging for coupled workstreams, standalone dispatch for independent tasks."
        },
        {
          "label": "Always use teams",
          "description": "Force all multi-agent workstreams into team mode with SendMessage coordination."
        },
        {
          "label": "Never use teams",
          "description": "Always use standalone Agent() calls with handoff documents. Simpler but agents can't talk to each other."
        }
      ],
      "multiSelect": false
    }
  ]
}

Map answers:

• Verify: "Yes, always" → { enabled: true, max_retries: 3, reviewer_model: "opus" }. "Yes, max 1" → { enabled: true, max_retries: 1, reviewer_model: "opus" }. "No" → { enabled: false, max_retries: 3, reviewer_model: "opus" }.
• Team: "Auto" → "auto", "Always" → "always", "Never" → "never".

If model_tier is "max_quality", set reviewer_model: "opus". If "max_speed", set reviewer_model: "sonnet". If "balanced", set reviewer_model: "opus" (reviewers should reason deeply).

Step 4: Autonomy Level

Use AskUserQuestion:

{
  "questions": [
    {
      "question": "How often should the CEO pause to check in with you?",
      "header": "Checkpoints",
      "options": [
        {
          "label": "After each workstream (Recommended)",
          "description": "CEO reports progress and asks approval before starting the next workstream. Good balance of control and flow."
        },
        {
          "label": "After each task",
          "description": "CEO pauses after every single task. Maximum oversight but slower."
        },
        {
          "label": "Only at phase boundaries",
          "description": "CEO runs autonomously within phases, only pausing between Discovery/Planning/Pre-flight/Execution. Fastest, least control."
        }
      ],
      "multiSelect": false
    },
    {
      "question": "How many agents should review requirements during pre-flight? (More = catches more ambiguity, less = faster)",
      "header": "Pre-flight",
      "options": [
        {
          "label": "3 agents (Recommended)",
          "description": "Good coverage without slowing things down."
        },
        {
          "label": "1 agent (Lightweight)",
          "description": "Fastest pre-flight. Only the most critical agent reviews."
        },
        {
          "label": "5 agents (Thorough)",
          "description": "Maximum ambiguity detection. Best for high-stakes or large projects."
        }
      ],
      "multiSelect": false
    }
  ]
}

Map:

• Checkpoint: "After each workstream" → "workstream", "After each task" → "task", "Only at phase boundaries" → "phase".
• Pre-flight: "3" → 3, "1" → 1, "5" → 5.

Step 5: Write settings.json

Assemble the settings object and write to ${CLAUDE_PLUGIN_ROOT}/settings.json:

{
  "model_tier": "<selected>",
  "verify_fix": {
    "enabled": true,
    "max_retries": 3,
    "reviewer_model": "opus"
  },
  "team_mode": "<selected>",
  "checkpoint": "<selected>",
  "preflight_agents": 3,
  "project_dir": "./ceo-projects",
  "shared_context": true
}

Step 6: Confirm

Show the user their final settings in a readable summary table:

Setting              | Value
---------------------|------------------
Model tier           | Balanced (Opus + Sonnet)
Verify-fix loop      | Enabled, max 3 retries
Reviewer model       | Opus
Team mode            | Auto
Checkpoints          | After each workstream
Pre-flight agents    | 3
Project directory    | ./ceo-projects
Shared context       | Enabled

Then say:

Setup complete. Your preferences are saved to settings.json. Run /ceo to start a project, or /ceo:setup anytime to reconfigure.

Notes

• shared_context and project_dir use sensible defaults and are not asked during setup to keep it fast. Users can edit settings.json manually to change them.
• The schema file at ${CLAUDE_PLUGIN_ROOT}/settings.schema.json documents every field with descriptions and valid values.