home/productivity/decompose

Decompose

GPTClaudeDeepSeek··313 copies·updated 2026-07-14
decompose.prompt
You are a research planner. Decompose the question below into sub-questions, then route each one to the appropriate search vendor.

Question: **{{clarified_question}}**

Two search vendors are available:

- **Claude Code (WebSearch)** — best for academic / technical / general-knowledge questions where you want considered, well-cited sources. Use for: definitions, methodologies, established trade-offs, library/framework comparisons, well-documented technologies.
- **Gemini (Google Search grounding)** — best for current-events / recent / news / market-data questions where freshness matters more than depth. Use for: anything that changed in the last 12 months, vendor news, prices, deprecations.

Produce a plan via `structured_output`:

- `research_plan`:
  - `rationale`: one-sentence explanation of how you split the question.
  - `parallelism_strategy`: one-sentence reason for sending some sub-questions to each vendor.
- `claude_subqs`: array of 2–4 sub-questions for Claude Code. Each has `{id: "c1"|"c2"|..., text: "...", rationale: "..."}`.
- `gemini_subqs`: array of 1–3 sub-questions for Gemini. Each has `{id: "g1"|"g2"|..., text: "...", rationale: "..."}`.

Sub-question text must be self-contained — assume the searching agent won't see the parent question. Total sub-questions should be 3–6.

If the question doesn't need both vendors (e.g. pure historical research → no Gemini), one of the arrays may be empty `[]`.

fill the variables

This prompt has 3 variables. Pro fills them into a ready-to-paste prompt for you — no manual find-and-replace.

{{clarified_question}{id: "c1"|"c2"|..., text: "...", rationale: "..."}{id: "g1"|"g2"|..., text: "...", rationale: "..."}
Unlock with Pro →

when to use it

Community prompt sourced from the open-source GitHub repo xingchengxu/OpenExpertise (MIT). A "Decompose" style prompt — adapt the placeholders and specifics to your task. Imported as-is and not independently retested here, so check the output before relying on it.

tags

productivitycommunitydeveloper

source

xingchengxu/OpenExpertise · MIT