Renderizado do repositório de origem, preservando títulos, exemplos, código, tabelas, links e imagens.
Invoke with: <ticket file, issue URL, or task description>
Question — Decompose the Task
Transform a task description into 3-7 specific, neutral research questions. These questions drive the next phase (Research) which runs in a separate context with no knowledge of what is being built.
Input
The user provides a task description, ticket file path, or issue reference.
Process
- Read any provided files fully before doing anything else.
- Light codebase exploration: Spawn a codebase-locator agent to find which areas of the codebase relate to the task. You need to know what exists to write good questions.
- Decompose into 3-7 research questions:
- Each question should cause a researcher to explore a different relevant area of the codebase
- Questions must be neutral — they ask what exists and how it works, never how to build something
- Prefer "trace the flow" questions that reveal architecture over yes/no questions
Good: "How does the middleware chain handle request authentication, and where are auth policies defined?" Bad: "What's the best way to add a new authenticated endpoint?"
Good: "What patterns exist for database migrations, and how are they tested?" Bad: "How should we add a new migration for the users table?"
- Determine the artifact directory:
- With ticket number:
thoughts/qrspi/PROJ-1234-brief-description/(use the project's ticket prefix) - Without ticket:
thoughts/qrspi/YYYY-MM-DD-brief-description/
- Create the artifact directory (relative to the working directory) if it doesn't exist (e.g.,
mkdir -p thoughts/qrspi/<id>/).
- Write `task.md` — a clean 2-3 sentence description of what's being built and why. This file persists the task context for later phases so the user doesn't have to re-explain it.
- Write `questions.md` to the artifact directory:
# Research Questions
## Context
[2-3 sentences describing which areas of the codebase to focus on.
Do NOT mention what is being built or why.]
## Questions
1. [Neutral, fact-seeking question]
2. [Neutral, fact-seeking question]
...- Present questions to the user and wait for approval or edits before finalizing.
Output
- Directory created:
thoughts/qrspi/<id>/ - Files written:
thoughts/qrspi/<id>/task.mdandthoughts/qrspi/<id>/questions.md - Tell the user: "Next: invoke the
qrspi-researchskill withthoughts/qrspi/<id>/"
Rules
questions.mdmust NOT contain the task description, goals, or desired behaviortask.mdis a brief, honest description of the goal — it will be read by later phases but NOT by Research- The researcher who reads these questions should have no idea what feature is being built
- Each question should target a different area or concern
- If the task is too simple for 3 questions, tell the user — QRSPI is for complex tasks

