| name | explore |
| description | Explore mode for codebase investigation and requirements clarification before committing to a change package. Use when requirements are unclear, a codebase investigation is needed before planning, options are being compared, or a risk scan is needed before a commit. Triggers on: "explore", "investigate", "look around", "understand the codebase", "scan for risks", "what's in here", "before I commit to anything". Stays in read-only observe mode — no files created — until an explicit artifact commit signal is received.
|
Explore Mode
Read-only investigation mode. Observe and analyze without creating artifacts until explicitly signaled to proceed.
When to Enter Explore Mode
Enter explore mode when any of the following apply:
- Requirements are not yet clear; need to ask clarifying questions or read the codebase first
- A codebase investigation is needed before a plan can be written
- Option comparison is in progress (weighing approaches before deciding)
- Risk scan needed before committing to a change package scope
While in Explore Mode
No files are created or modified until an explicit artifact commit signal is received.
Permitted actions in explore mode:
- Read files (
read / view tools)
- Search codebase (
search / grep tools)
- Ask the user clarifying questions
- Build an internal understanding of the system
Prohibited actions until signaled:
- Creating or modifying
changes/ files
- Creating or modifying any source or documentation file
- Running
sync-dotgithub.ps1
Explicit Artifact Commit Triggers
Explore mode ends and artifact creation begins only when one of the following explicit signals is received from the user:
/proceed"create change package""start brainstorm""I want to formalize this"
Any other phrasing that sounds like "let's start writing" should be confirmed with the user before exiting explore mode: "Do you want me to create a change package now?"
Explore Mode Summary Output
Before transitioning out of explore mode, produce a brief summary:
## Explore Summary
- **Scope**: [what was investigated]
- **Key Findings**: [2–5 bullet points]
- **Recommended Next Step**: [brainstorm / fast-path plan / other]
- **Risk Signal**: [Low / Med / High — brief rationale]
This summary becomes input to the brainstorm or plan step.
Common Rationalizations
在進行 explore 探索過程中，AI 可能以下列藉口略過關鍵步驟：
| 常見藉口 | 反制說明 |
|---|
| "看一看就夠了，不需要正式 explore" | ⛔ 非正式「看一看」沒有輸出物——Explore 必須產出 Findings 文件，才能作為 brainstorm/plan 的可靠輸入 |
| "程式碼庫我已經很熟了，可以直接進入計畫" | 熟悉感不等於對當前狀態的準確掌握——任何超過兩週未接觸的程式碼庫，或涉及 brownfield 改動，均應重新 explore |
| "Explore 會拖慢速度，直接開始實作更有效率" | 未探索就實作等於在未知地圖上規劃路線——explore 的時間投入遠低於因誤解現況而導致的返工成本 |
Verification
在完成 explore 並準備移交產出物前，逐項確認（Gate = 交付前閘門；Verification = 自我完成確認）：
