Chapter 8: 🎯 Chapter 8: Exploring Ideas with /opsx:explore
📖 Learning Objectives
Learn to "think before you act" – use explore mode to clarify vague ideas into actionable decisions.
💡 What is explore mode?
It's not "code writing mode," it's "thinking partner mode." Rules:
✅ Can read existing code
✅ Can ask clarifying questions
✅ Can suggest solutions
✅ Can make comparisons
❌ Cannot write code
❌ Cannot create changes (unless explicitly requested)
→ It's like discussing "how to implement this feature" with a colleague in front of a whiteboard.
⏱️ When to use explore
flowchart TD
Start["Have an idea"] --> Q{Clarity?}
Q -->|Completely clear
know every step| P["Directly /opsx:propose"]
Q -->|Have a general direction
but details unclear| E["First /opsx:explore"]
Q -->|Completely vague
don't even know what to do| B["Chat first
(any brainstorm)"]
E --> Clear["After clarification
opsx:propose"]
B --> E
style E fill:#fff9c4
style P fill:#c8e6c9💻 Hands-on: Exploring doc2video
Our conversation flow at the time:
Round 1 - Me:
"I want to create a project: given a tutorial document, such as the installation of Claude Code, it can explain and install Claude Code simultaneously."
Round 1 - explore agent asks back:
- Who is the user? (Learner / Document author)
- Real execution or simulation?
- Does the desktop need a GUI?
- How strict is the document format?
Round 2 - My answer:
"1 run on a real machine 2 real execution 3 markdown 4 audio-video sync is a bit difficult"
Round 2 - explore continues:
- Presents 3 tension points (sandbox boundaries, synchronization strategy, document protocol)
- 2~3 options for each tension point
Round 3~N: One question, one choice per round, until all my major decisions were finalized.
📝 Fixed Pattern for explore Conversations
flowchart TD
Idea["You propose an idea
(vague)"] --> Q1["explore asks clarifying questions
(one at a time)"]
Q1 --> Ans["You answer / Choose from options"]
Ans --> Surface["explore surfaces tension points
(2~5 trade-offs)"]
Surface --> Pick["You make the decision"]
Pick --> More{Are there unclarified points?}
More -->|Yes| Q1
More -->|No| Crystal["All major decisions formed"]
Crystal --> Trigger["You say: /opsx:propose"]
Trigger --> Done["Solidify decisions into the 4-piece set"]
style Idea fill:#fff9c4
style Done fill:#c8e6c9→ Pacing: Back and forth, focusing on a single point, don't ask 5 questions at once.
📦 What is explore's "output"?
It's not a file, it's a list of clarified decisions. When we feel an idea is mature, I say "/opsx:propose", and all previous decisions are packaged into a change artifact.
🚫 explore Anti-patterns
❌ Let explore write code directly
→ "Just discuss this with me, don't start coding"
❌ Let explore make decisions for you
→ It should offer options + recommendations, you make the final call
❌ explore hasn't converged after 30 minutes
→ Indicates the problem is too big, break it into smaller problems first
❌ explore leaves no trace after completion
→ Key decisions should ideally be written into a future design.md
📊 A Comparison Technique We've Used
Have explore create a "decision matrix":
You: For voiceover, should we use macOS say, edge-tts, or OpenAI TTS?
explore:
| Option | Cost | Quality | Speed | Recommendation |
|---|---|---|---|---|
| macOS say | Free | Robotic | Fast | MVP |
| edge-tts | Free | Good | Faster | Recommended |
| OpenAI TTS | $$ | Excellent | Fast | Large-scale scenarios |
I suggest edge-tts...
You: edge-tts
→ A single sentence to convey the decision, saving a lot of back-and-forth.
✅ What You Can Do Now
- Know when to use explore instead of directly proposing
- Be able to drive progress with the "question → options → recommendation → decision" rhythm
- Identify signs that explore mode is being misused
The next chapter will thoroughly explain "decisions" – a core skill for specification-driven development.