Skip to content

Ultra-Think Command

Input Validation

@.claude/context/AGENT_INPUT_VALIDATION.md

Usage

/ultrathink <TASK_DESCRIPTION>

Purpose

Deep analysis mode for complex tasks. Invokes workflow skills (default-workflow or investigation-workflow) based on task type, with automatic fallback to markdown workflows if skills are not yet available. You MUST use one of the workflow skills - either default-workflow or investigation-workflow or both or its possible the user could pass in additional workflow skills like cascade or debate.

EXECUTION INSTRUCTIONS FOR CLAUDE

When this command is invoked, you MUST: Execute this exact sequence for the task: {TASK_DESCRIPTION}

  1. First, detect task type - Check if task is investigation or development
  2. Investigation keywords: investigate, explain, understand, how does, why does, analyze, research, explore, examine, study
  3. Development keywords: implement, build, create, add feature, fix, refactor, deploy
  4. If both types detected: Use hybrid workflow (investigation first, then development)
  5. If only investigation keywords found: Use @.claude/workflow/INVESTIGATION_WORKFLOW.md
  6. If only development keywords found: Use @.claude/workflow/DEFAULT_WORKFLOW.md
  7. Mandatory - not doing this wil require rework Invoke the appropriate workflow skill using the Skill tool:
  8. Investigation: Skill(skill="investigation-workflow")`
  9. Development: Skill(skill="default-workflow")
  10. FALLBACK: If skill invocation fails (skill not found), fall back to reading markdown workflows:
    • Investigation: @.claude/workflow/INVESTIGATION_WORKFLOW.md
    • Development: @.claude/workflow/DEFAULT_WORKFLOW.md
  11. ALWAYS Create a comprehensive todo list using TodoWrite tool that includes all workflow steps/phases
  12. ALWAYS Execute each step systematically, marking todos as in_progress and completed

THERE IS NO VALUE in SKIPPING STEPS - DO NOT TAKE SHORTCUTS.

  • For Each Workflow Step:
  • Mark step as in_progress in TodoWrite
  • Break down the step into smaller tasks if needed
  • Read the step requirements from workflow
  • Invoke specified agents via Task tool
  • Log decisions made
  • Mark step as completed
  • No steps are optional - all steps must be followed in sequence.
  • Agent Invocation Pattern:
For step requiring "**Use** architect agent":
→ Invoke Task(subagent_type="architect", prompt="[step requirements + task context]")

For step requiring multiple agents:
→ Invoke multiple Task calls in parallel

Agent Orchestration

When to Use Sequential

  • Architecture → Implementation → Review
  • Each step depends on previous
  • Building progressive context

When to Use Parallel

  • Multiple independent analyses
  • Different perspectives needed

  • Gathering diverse solutions

  • Decision Logging:

After each major decision, append to DECISIONS.md: - What was decided - Why this approach - Alternatives considered

  • Mandatory Cleanup: Always end with Task(subagent_type="cleanup")

  • Use the specified agents for each step (marked with "Use" or "Always use")

  • **MANDATORY: Enforce all steps.
  • Track decisions by creating and writing important decisions to .claude/runtime/logs/<session_timestamp>/DECISIONS.md
  • End with cleanup agent (development) or knowledge capture (investigation)

Task Management

Always use TodoWrite to:

  • Break down complex tasks
  • Track progress
  • Coordinate agents
  • Document decisions
  • Track workflow checklist completion

Example Flow

Development Task Example

User: "/ultrathink implement JWT authentication"

1. Detect: Development task (contains "implement")
2. Select: default-workflow skill
3. Try: Skill(skill="default-workflow")
4. Fallback if needed: Read `.claude/workflow/DEFAULT_WORKFLOW.md`
5. Begin executing workflow steps with deep analysis
6. Orchestrate multiple agents where complexity requires
7. Follow all workflow steps as defined
8. Adapt to any user customizations automatically
9. MANDATORY: Invoke cleanup agent at task completion

Investigation Task Example

User: "/ultrathink investigate how the reflection system works"

1. Detect: Investigation task (contains "investigate")
2. Select: investigation-workflow skill (6 phases)
3. Inform user: "Detected investigation task. Using investigation-workflow skill"
4. Try: Skill(skill="investigation-workflow")
5. Fallback if needed: Read `.claude/workflow/INVESTIGATION_WORKFLOW.md`
6. Execute Phase 1: Scope Definition
7. Execute Phase 2: Exploration Strategy
8. Execute Phase 3: Parallel Deep Dives (multiple agents simultaneously)
9. Execute Phase 4: Verification & Testing
10. Execute Phase 5: Synthesis
11. Execute Phase 6: Knowledge Capture
12. MANDATORY: Update DISCOVERIES.md with findings

Hybrid Workflow Example (Investigation → Development)

User: "/ultrathink investigate how authentication works, then add OAuth support"

Phase 1: Investigation
1. Detect: Investigation keywords present ("investigate")
2. Select: investigation-workflow skill (6 phases)
3. Try skill invocation (fallback to markdown if needed)
4. Execute full investigation workflow
5. Document findings in DISCOVERIES.md

Phase 2: Transition to Development
6. Detect: Development work needed ("add OAuth support")
7. Transition to default-workflow skill
8. Try skill invocation (fallback to markdown if needed)
9. Resume at Step 4 (Research and Design) using investigation insights
10. Continue through Step 15 (implementation → testing → PR)
11. MANDATORY: Invoke cleanup agent at completion

When Investigation Leads to Development:

Some development tasks require investigation first (Step 4 of DEFAULT_WORKFLOW.md):

  • Unfamiliar codebase areas
  • Complex subsystems requiring understanding
  • Unclear architecture or integration points
  • Need to understand existing patterns before designing new ones

In these cases, pause development workflow at Step 4, run full INVESTIGATION_WORKFLOW.md, then resume development with the knowledge gained.

ALWAYS PICK A WORKFLOW OR FOLLOW THE ONE THE USER TOLD YOU TO USE

YOU MAY NOT SKIP STEPS in the workflow. UltraThink enhances the workflow with deep multi-agent analysis while respecting user customizations.

Remember: Ultra-thinking means thorough analysis before action, followed by ruthless cleanup.