菜单
Use when starting a new deep-work task. Decomposes a task description into objective research questions that avoid premature solutioning. Phase 1 of the deep-work pipeline.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when starting the RPI flow and you want to deeply understand a question, system, or feature by reading the codebase before planning.
Use when starting the RPI flow and you want to deeply understand a question, system, or feature by reading the codebase before planning.
Use when starting a new deep-work task. Decomposes a task description into objective research questions that avoid premature solutioning. Phase 1 of the deep-work pipeline.
Use when you have research questions from deep-work Phase 1. Objectively investigates the codebase to answer pasted questions without access to the original task description.
Use when deep-work Phase 2 research is complete. Combines research findings with the original task to explore design options, evaluate tradeoffs, and make decisions interactively.
Use when deep-work Phase 3 design decisions are complete. Maps chosen decisions to concrete file changes organized into implementable phases.
| name | dw-01-research-questions |
| description | Use when starting a new deep-work task. Decomposes a task description into objective research questions that avoid premature solutioning. Phase 1 of the deep-work pipeline. |
Decompose the user's prompt into objective, investigative questions answerable by reading the codebase. Questions must NOT assume any particular solution.
Announce at start: "Starting deep-work Phase 1: Research Questions."
$ARGUMENTS:
<topic-slug> and the rest as the task description<topic-slug><topic-slug> (lowercase, hyphens, no special chars)"$SKILL_BASE_DIR/setup.sh" "<topic-slug>" and parse stdout for REPO and ARTIFACT_DIR (script also creates the directory). $SKILL_BASE_DIR is the "Base directory for this skill" path shown at the top of this prompt.00-ticket.md to the artifact directory:
---
phase: ticket
date: <today>
topic: <topic-slug>
repo: <repo>
git_sha: <HEAD>
status: complete
---
## Ticket
<user's prompt or file contents>
Identify the key nouns, systems, and actions mentioned in the prompt. These are the seeds for your research questions. Focus on concrete components, data flows, and interactions — avoid abstract goals or desired outcomes.
Gather lightweight structural context (NOT deep implementation details):
Generate 5-20 questions. EVERY question must be:
Distribute across categories:
| Category | Pattern | Example |
|---|---|---|
| Subsystem Understanding | "How does [component] work?" | "How does auth middleware chain requests?" |
| Code Tracing | "What is the [data] flow from [A] to [B]?" | "Request lifecycle from handler to DB?" |
| Pattern Discovery | "What patterns exist for [action]?" | "Patterns for adding API endpoints?" |
| Dependency Mapping | "What does [module] depend on?" | "What does handlers package import?" |
| Boundary Identification | "Where do [A] and [B] integrate?" | "Where do HTTP and storage connect?" |
| Constraint Discovery | "What invariants does [system] enforce?" | "What do tests enforce for handlers?" |
FORBIDDEN question patterns:
Write 01-research-questions.md to the artifact directory:
---
phase: research-questions
date: <today>
topic: <topic-slug>
repo: <repo>
git_sha: <HEAD>
status: complete
---
## Original Prompt
<full prompt — stored for traceability, NOT passed to Phase 2>
## Research Questions
### Subsystem Understanding
1. <question>
### Code Tracing
2. <question>
### Pattern Discovery
3. <question>
...
digraph handoff {
rankdir=LR;
node [shape=box, style=rounded, fontname="Helvetica"];
p1_out [label="01-research-questions.md\n(contains prompt + questions)"];
user [label="User reviews &\nedits questions" shape=diamond style=filled fillcolor="#ffffcc"];
script [label="extract-research-questions.sh\n(auto-run by Phase 2)" shape=octagon style=filled fillcolor="#ff4444" fontcolor=white];
p2_in [label="Questions only\n(prompt never exposed)"];
p1_out -> user [label="review & edit"];
user -> script [label="fresh conversation\n/dw-02-research <slug>"];
script -> p2_in [label="sed extraction"];
}
.state.json in the artifact directory:
{
"topic": "<topic-slug>",
"repo": "<repo>",
"current_phase": 1,
"completed_phases": [1],
"last_updated": "<ISO timestamp>"
}
01-research-questions.md as needed.
When ready, run /dw-02-research <topic-slug> in a fresh conversation.
The research skill automatically extracts only the questions section via
extract-research-questions.sh — the original prompt is never exposed to Phase 2."