메뉴
Use when receiving ambiguous instructions, preparing for state-changing operations, or needing explicit user confirmation
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Use when the current Agent LLM cannot process images directly and visual analysis is needed — bridges images through KimiCode CLI print mode to a multimodal Kimi model for text description
Use when building HUDs, menus, inventory screens, settings panels, or any widget-based interface in Unreal Engine 5. Also use when connecting C++ logic to UMG Blueprint visuals, handling gamepad or keyboard focus navigation, managing UI state, creating widget animations, or troubleshooting UMG performance issues like frame drops, hitches, or widget memory leaks.
Use when working in a DevFlow project with .devflow/ directory and gate-based step-by-step workflows
Use when contributing new skills to the skill-lib repository, installing skills locally, or verifying skill compliance with repository standards
Use when analyzing unfamiliar code modules, understanding system architecture, or preparing for refactoring
Use when implementing new modules from design documents, adding features to existing code, or generating structured implementations
| name | requirement-clarification |
| description | Use when receiving ambiguous instructions, preparing for state-changing operations, or needing explicit user confirmation |
| version | 1.1 |
| author | agent |
Intercept and clarify ambiguous user instructions before execution.
Use this skill when:
Trigger Threshold: Score >= 30 or any mandatory trigger condition
Calculate ambiguity score (0-100):
| Factor | Score |
|---|---|
| Missing critical parameter | +25 each |
| Vague reference | +15 each |
| Non-affirmative statement | +20 |
| Scope ambiguity | +10 |
| Conditional clause | +15 |
| High-risk operation | +25 |
| Non-confirmation word (准备 (Preparing)/考虑 (Considering)/研究 (Studying/Researching)/讨论 (Discussing)) | +20 |
Thresholds:
Before ANY state-changing action, ask:
"Stop. Does this change state? Yes. Is it confirmed? No. Go confirm."
Triggers mandatory pause:
Single Question Focus: Ask ONE critical question at a time.
Provide Sensible Defaults:
Target path?
- A) ./backup/ (backup directory)
- B) ./archive/ (archive directory)
- C) Other: _____
Use Concrete Examples:
Performance target?
Examples:
- "Startup time < 3 seconds"
- "Memory usage < 100MB"
- "QPS > 1000"
Standard format:
📋 Task Plan
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🎯 Goal: [Clear description of final state]
Steps:
1. [Step 1 with full parameters]
2. [Step 2 with full parameters]
...
Expected Output:
- [Output 1]
- [Output 2]
Risks:
- [Risk 1]
- [Risk 2]
Estimated Time: [X] minutes
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✅ Confirmation: Reply "批准计划 (Approve plan)" to proceed
Valid confirmations:
NOT confirmations (require clarification):
Once "批准计划 (Approve plan)" received:
PAUSE and re-confirm if:
✅ Task Complete Report
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Completed Steps:
✓ Step 1
✓ Step 2
...
Actual Output:
- [Output 1] - [description]
- [Output 2] - [description]
Deviations from Plan: [None / explanation]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
✅ Ask ONE question at a time for simple gaps ✅ Provide sensible defaults with options ✅ Use concrete examples to illustrate ✅ Acknowledge when clarification is needed ✅ Keep tone collaborative, not interrogative ✅ Respect "你决定 (You decide)" but still require final approval
❌ Ask more than 3 questions in one turn ❌ Proceed with assumptions without confirmation ❌ Treat questions as directives ❌ Ignore conditionals ("if", "when", "maybe") ❌ Pressure user for perfect precision
Even with plan approval, require separate confirmation:
git push (pushing to remote)git reset --hard (force reset)rm -rf or permanent deletionSymptom: User provides multiple answers that don't converge.
Solution:
ℹ️ After multiple clarifications, requirements still have uncertainties.
Options:
A) Split task into smaller steps
B) I provide a recommendation for your approval
Choose?
Symptom: User says "你决定 (You decide)".
Solution:
📝 Recommended Plan
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Based on common scenarios, I suggest:
[Plan details]
Reply "批准计划 (Approve plan)" to proceed, or tell me specific changes.
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Symptom: User uses "准备实施 (Preparing to implement)".
Solution:
ℹ️ Clarification
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
You used "准备" (preparing), which indicates planning phase.
Please confirm:
- A) Continue planning/discussion (no execution)
- B) Approve execution (say "批准实施 (Approve implementation)" or "确认执行 (Confirm execution)")
Current status: Waiting for explicit approval
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Symptom: User contradicts previous clarifications.
Solution:
ℹ️ Received new information. Updated plan:
[Revised plan]
Confirm execution of this updated version?
Symptom: Cannot form reliable plan.
Solution:
⚠️ Instruction too ambiguous for reliable plan.
Please rephrase in format:
Format: "对 [specific target] 执行 [specific action]，要求 [specific standard/condition]"
Example:
"对 memory/2026-02-11.md 添加今天的 git 操作记录 (Add today's Git operation records), 要求包含完整的命令输出 (require complete command output)"
Input: User instruction
│
├─ Read-only query? (查看/读取/显示 (View/Read/Display))
│ └─ YES → Skip, proceed directly
│
├─ High-risk operation? (git, file ops, config)
│ └─ YES → Trigger (+25, mandatory pause)
│
├─ Non-confirmation words? (准备 (Preparing)/考虑 (Considering)/研究 (Studying/Researching)/讨论 (Discussing))
│ └─ YES → Trigger (+20)
│
├─ Action without confirmation? ("Do it" / "Go ahead")
│ └─ YES → Trigger (intent ≠ approval)
│
├─ Vague reference? ("this", "that", "it")
│ └─ YES → Trigger
│
├─ Question format? (吗？ (? - question marker)/？)
│ └─ YES → Trigger
│
├─ Conditional? ("if", "when", "maybe")
│ └─ YES → Trigger
│
├─ Missing parameters?
│ └─ YES → Trigger
│
└─ Final Check
├─ Score < 30 AND no mandatory triggers?
│ └─ YES → Proceed
└─ Score ≥ 30 OR mandatory trigger?
└─ YES → Trigger skill
v1.1 (2026-02-11) - Integration with agent policies
v1.0 (2026-02-11) - Initial release