/research

command: research
disable-model-invocation: true
description: Deep research with multi-source synthesis and comprehensive analysis

Research - Deep Multi-AI Research

**Your first output line MUST be:** `🐙 Octopus Research`

🤖 INSTRUCTIONS FOR CLAUDE

MANDATORY COMPLIANCE — DO NOT SKIP

**When the user explicitly invokes `/octo:research`, you MUST execute the structured research workflow below.** You are PROHIBITED from answering directly, skipping the multi-provider research, or deciding the topic is "too simple" for deep research. The user chose this command deliberately — respect that choice.

EXECUTION MECHANISM — NON-NEGOTIABLE

**You MUST execute this command by reading and following the dedicated research source. You are PROHIBITED from:**

• ❌ Using the Agent tool to research/implement yourself instead of invoking the skill
• ❌ Using WebFetch/Read/Grep as a substitute for multi-provider dispatch
• ❌ Skipping `orchestrate.sh` calls because "I can do this faster directly"
• ❌ Implementing the task using only Claude-native tools (Agent, Write, Edit)

**Multi-LLM orchestration is the purpose of this command.** If you execute using only Claude, you've violated the command's contract.

---

When the user invokes this command (e.g., `/octo:research <arguments>`):

Step 1: Resolve Research Breadth/Intensity

First parse explicit flags from the user's arguments:

• `--breadth=light|standard|exhaustive`
• `--intensity=quick|standard|deep`

Map breadth to intensity when intensity is absent:

• `light` -> `quick`
• `standard` -> `standard`
• `exhaustive` -> `deep`

If neither flag is present, use the AskUserQuestion tool to select intensity:

AskUserQuestion({
  questions: [
    {
      question: "How thorough should the research be?",
      header: "Research Intensity",
      multiSelect: false,
      options: [
        {label: "Quick (1-2 min)", description: "2 agents — fast problem space scan"},
        {label: "Standard (2-4 min)", description: "4-5 agents — balanced multi-perspective coverage (recommended)"},
        {label: "Deep (3-6 min)", description: "6-7 agents — exhaustive analysis with web search"}
      ]
    }
  ]
})

Map the answer to an intensity value:

• "Quick" → `quick`
• "Standard" → `standard`
• "Deep" → `deep`

Step 2: Load Research Workflow with Intensity

**✓ CORRECT - Read the explicit workflow source:**

Read `${CLAUDE_PLUGIN_ROOT:-${CODEX_PLUGIN_ROOT:-${HOME}/.claude-octopus/plugin}}/.claude/skills/skill-deep-research/SKILL.md`, then execute it with `[breadth=light|standard|exhaustive] [intensity=quick|standard|deep] <user's arguments without routing flags>`.

Examples:

• Read the source above and execute with `[breadth=standard] [intensity=standard] OAuth 2.0 authentication patterns`.
• `/octo:research --breadth=exhaustive current agent orchestration patterns` -> execute the source with `[breadth=exhaustive] [intensity=deep] current agent orchestration patterns`.

**✗ INCORRECT - Do NOT use these:**

Using the octo:discover skill       ❌ Wrong here! Research has a dedicated skill contract
Skill(skill: "flow-discover", ...)   ❌ Wrong! Internal skill name, not resolvable by Skill tool
Skill(skill: "discover", ...)        ❌ Wrong! Must use full namespaced name
Task(subagent_type: "octo:discover", ...)  ❌ Wrong! This is a skill, not an agent type

---

**Loads the dedicated research workflow only after this explicit command.**

Quick Usage

Just use natural language:

"Research OAuth 2.0 authentication patterns"
"Deep research on microservices architecture best practices"
"Research the trade-offs between Redis and Memcached"

What Is Research?

A dedicated research workflow aligned with the **Discover** phase of the Double Diamond methodology:

• Multi-AI research (Claude + Antigravity + Codex)
• Comprehensive analysis of options
• Trade-off evaluation
• Best practice identification

Report Format (MANDATORY)

All research output MUST follow this structured template:

1. Executive Summary

2-3 sentences summarizing the key finding. What does the reader need to know?

2. Key Themes

Group findings into 3-5 themes. Each theme gets a heading, a summary paragraph, and supporting evidence.

3. Key Takeaways

Numbered list of actionable insights. Each takeaway should be specific enough to act on.

4. Sources & Attribution

Every factual claim MUST cite its source. Claims without sources should be explicitly marked as **inference** or **opinion**. Format:

• `[Source: <name/URL>]` for verified facts
• `[Inference]` for conclusions drawn from evidence
• `[Opinion: <provider>]` for provider-specific perspectives

5. Methodology

Brief note on what was researched, which providers contributed, and any gaps or limitations:

• Providers used and their roles
• Search queries or exploration paths taken
• Areas not covered or needing deeper investigation
• Cross-references checked and gaps acknowledged

Quality Rules

• **No unsourced claims** — every assertion needs either a source or an explicit [Inference] tag
• **Acknowledge gaps** — if a topic wasn't fully explored, say so
• **Cross-reference** — when providers disagree, note the disagreement and which evidence is stronger

Natural Language Examples

"Research GraphQL vs REST API design patterns"
"I need deep research on Kubernetes security best practices"
"Research authentication strategies for microservices"

For a bounded run, add --intensity=quick, --intensity=standard, or --intensity=deep (or the equivalent --breadth=light|standard|exhaustive). Research records a local run manifest, source ledger, and verification report so an interrupted run can be resumed rather than repeated. See docs/RESEARCH-EVIDENCE.md for the run files, network boundaries, and research-verify recovery command.