-
Notifications
You must be signed in to change notification settings - Fork 95
Deep Discuss
Deep Discuss is the companion skill for structured technical discussion. Use it when the goal is not yet ready for implementation and the user needs analysis, debugging direction, architecture trade-offs, brainstorming, or decision support.
Spec-Driven Develop is for executing a large project. Deep Discuss is for understanding what should be done before execution starts.
Use Deep Discuss when the user says things like:
- "Let's discuss this."
- "Help me analyze this."
- "I have a problem."
- "What do you think?"
- "I'm torn between two approaches."
- "讨论一下."
- "帮我分析."
- "我遇到一个问题."
- "我在纠结."
Do not use it for simple fact lookup or direct implementation requests.
Phase 1 Receive Information
Phase 2 Problem Audit
Phase 3 Deep Analysis
Phase 4 Solution Design
Phase 5 Self-Review
Phase 6 Final Review
Phase 7 Execution, only when explicitly requested
The agent listens and restates the key context. It should not rush into diagnosis. If the description is too vague, it asks one or two critical clarification questions.
This is the quality gate.
The agent checks:
- Does the reported issue actually constitute a problem?
- Is the user's suspected cause supported by evidence?
- Is the available information enough for analysis?
- What important information is missing?
- Are there hidden problems behind the visible symptom?
If information is insufficient, the workflow pauses instead of proceeding on weak assumptions.
The agent performs multi-angle analysis and explicitly marks uncertainty. The goal is root-cause reasoning, not a fast answer.
Good Deep Discuss analysis should:
- Consider multiple explanations.
- Separate evidence from inference.
- Identify assumptions.
- Mark confidence levels.
- Explain what would change the conclusion.
The agent proposes two or three solution options when the problem space supports multiple approaches.
Each option should include:
- What it does.
- Why it works.
- Trade-offs.
- Risks.
- When to choose it.
The agent can recommend a path, but the user remains the decision-maker.
The agent reviews its own proposed solution before asking the user to accept it.
It checks for:
- Missing scenarios.
- Weak assumptions.
- Underestimated implementation complexity.
- Overly broad or unnecessary work.
- Whether the solution covers hidden issues from Phase 2.
After the user confirms direction, the agent reviews completeness, risks, and validation.
This phase turns a plausible solution into an executable one.
Execution only happens when the user explicitly asks for it.
At that point, the agent follows the confirmed plan, reports meaningful progress, and returns to discussion mode if new facts invalidate the plan.
| Need | Use |
|---|---|
| Understand a problem | Deep Discuss |
| Compare technical options | Deep Discuss |
| Debug a confusing symptom | Deep Discuss |
| Decide project direction | Deep Discuss |
| Plan and execute a large code change | Spec-Driven Develop |
| Create Issues, milestones, and PRs | Spec-Driven Develop |
| Track cross-session implementation progress | Spec-Driven Develop |