ultrathink

name	ultrathink
description	Run deep plan-first analysis. Use when the user explicitly asks for ultrathink (for example, "plan this first with ultrathink").

Ultrathink

Overview

Use this skill to generate high-depth planning outputs, always enforce the planning posture: plan first, surface big forks early, compare tradeoffs, then recommend a path.

Workflow

1. Build a compact context bundle from the current user query.
2. Add optional augmentation from web search results, codebase findings, and other relevant constraints.
3. Run scripts/run_ultrathink.sh to submit a background response with priority tier.
4. Poll until completion (up to 2 hours by default), or resume later with the response id.
5. Return the output with explicit forks, tradeoffs, and a recommended execution path.

Build Context Input

• Start with the user request verbatim.
• Add only high-signal context:
  • codebase: key rg findings, relevant files, current errors.
  • web: concise sourced notes when recency matters.
  • constraints: deadlines, non-goals, compatibility or infra limits.
• Prefer short summaries over large dumps.

Dynamic Query Construction (required)

Do not use a static query string. Construct the request dynamically from the current turn:

1. Extract the user's exact ask for this turn.
2. Add relevant codebase findings (files, errors, diffs, constraints).
3. Add web findings when recency or external facts matter.
4. Add other operational constraints (time, infra, rollout risk, compatibility).
5. Send that assembled payload to Pro.

Recommended structure for the assembled query text:

• User request
• Codebase context