Iván

Self Reflection

AI 智慧體v1.1.0
作者:Iván更新於 2026-05-11共 2 個版本 · 最新版本釋出於 2026-02-25reflection@clawhub_ivangdavila/reflection
提示詞最佳化上下文管理

據 AlphaMoat 統計,截至 2026-10,Self Reflection 累計下載 10,692 次、獲 8 個星標,按累計下載量在本次參與排名的 10,914 個 AI 技能中排名第 2,546 位,在「AI 智慧體」分類(1,632 個)中排名第 347 位。

列表預設按累計下載量排序,也可按星標排序;詳情頁排名按累計下載量計算。

累計下載
1.1萬
星標
8
全站下載量排名
#2,546
全站前 24% · 「AI 智慧體」前 22%
平台累計下載星標評論
ClawHub ↗98447-
SkillHub ↗1.1萬8—

各平台的計數口徑不同,統計時間不一致。

這個技能是做什麼的

三種觸發:1) 交付前——重要工作發出前暫停 30 秒做 7 維度快評(正確性、完整性、清晰度、健壯性、效率、契合度、自豪感),任一低於 7/10 先修復;2) 出錯後——使用者糾正即資料,立即停止、分析根因並按 Miss/Root/Fix 記錄;3) 模式檢測——同類問題出現 3 次以上即視為盲區,生成預防規則。記憶存在 ~/reflection/(memory.md、reflections.md、patterns.md、archive/);開工前掃描 patterns.md 主動注入相關教訓。模式生命週期為 EMERGING→ACTIVE→MONITORING→RESOLVED(30 天干淨後解除)。注意:避免過度反思(30 秒而非 5 分鐘);反思檔案會留存糾正與教訓記錄,定期審閱並避免存入敏感資訊;錯誤記錄的教訓可能帶偏後續回答,採納前先複核。

資料釋出於 2026-10-27

可以用來做什麼

  • 交付前做 7 維度 30 秒快速自查
  • 把使用者糾正沉澱為根因分析與預防規則
  • 開工前主動注入歷史教訓避免重蹈覆轍

什麼時候用

  • 希望智慧體輸出質量超過'看起來沒問題'時
  • 同一類錯誤反覆出現、需要系統性改進時
  • 想讓智慧體記住並複用過往經驗時

示例提示詞

以下是根據技能文件生成的示例,不代表實際執行結果。

幫我按 7 個維度檢查這份方案再發出去。
剛才你錯了,記錄一下根因和預防規則。
看看目前有哪些活躍的錯誤模式。

月度資料趨勢

累計下載1.1萬
361574161.1萬2026-03 · 累計下載 41392026-05 · 累計下載 79902026-06 · 累計下載 82832026-07 · 累計下載 92972026-08 · 累計下載 94872026-09 · 累計下載 98132026-10 · 累計下載 1.1萬26-0326-0926-10
星標8
6782026-03 · 星標 62026-05 · 星標 82026-06 · 星標 82026-07 · 星標 82026-08 · 星標 82026-09 · 星標 82026-10 · 星標 826-0326-0926-10

2026-09 起統計方法調整:同名技能按作者分別統計,並覆蓋全部技能。圖中虛線為調整前的月份,與之後的資料不直接比較,也不計算跨越調整的增長。

月份累計下載環比星標
2026-101.1萬—8
2026-099813—8
2026-089487+2.0%8
2026-079297+12.2%8
2026-068283+3.7%8
2026-057990—8
2026-034139—6

SKILL.md 原文

以下是作者釋出的原始檔案,按純文字顯示,本站未做修改或稽核;其中的連結不可點選,命令請核實後再執行。抓取於 2026-10-23。

9.1 KB · ClawHub v1.1.0
---
name: Self Reflection
slug: reflection
version: 1.1.0
homepage: https://clawic.com/skills/reflection
description: Learns when to stop and review. Self-critiques before showing you, fewer revision rounds.
metadata: {"clawdbot":{"emoji":"🪞","requires":{"bins":[]},"os":["linux","darwin","win32"],"configPaths":["~/reflection/"]}}
changelog: Major rewrite. Visual workflow, pattern detection system, proactive lesson injection, and multi-trigger architecture.
---
展開全部(共 212 行)
Agents repeat mistakes. Not because they're incapable — because they forget. This skill changes that. Your agent pauses before delivering, catches its own blind spots, and remembers lessons for next time.

## When to Use

User needs quality assurance beyond "looks good to me." Agent handles pre-delivery evaluation, post-mistake analysis, pattern detection across sessions, and proactive lesson surfacing before repeating errors.

## How It Works

```
         ┌──────────────────────────────────────────────┐
         │              SELF REFLECTION LOOP            │
         └──────────────────────────────────────────────┘
                              │
         ┌────────────────────┼────────────────────┐
         ▼                    ▼                    ▼
    ┌─────────┐         ┌──────────┐         ┌─────────┐
    │  PRE    │         │  POST    │         │PATTERN  │
    │DELIVERY │         │ MISTAKE  │         │DETECTED │
    └────┬────┘         └────┬─────┘         └────┬────┘
         │                   │                    │
         │  "Before I send   │  "User corrected   │  Same mistake
         │   this, let me    │   me. Why?"        │  3 times...
         │   double-check"   │                    │
         │                   │                    │
         └───────────────────┴────────────────────┘
                              │
                              ▼
                    ┌─────────────────┐
                    │ 7-DIMENSION     │
                    │ EVALUATION      │
                    │ (30 seconds)    │
                    └────────┬────────┘
                              │
              ┌───────────────┴───────────────┐
              ▼                               ▼
       ┌─────────────┐                 ┌─────────────┐
       │  ALL CLEAR  │                 │ ISSUE FOUND │
       │  Deliver    │                 │ Fix first   │
       └─────────────┘                 └──────┬──────┘
                                              │
                                              ▼
                                    ┌─────────────────┐
                                    │  LOG LESSON     │
                                    │  Miss → Root    │
                                    │  → Prevention   │
                                    └────────┬────────┘
                                              │
                                              ▼
                                    ┌─────────────────┐
                                    │  INJECT NEXT    │
                                    │  TIME           │
                                    │  "Before we     │
                                    │   proceed..."   │
                                    └─────────────────┘
```

## The Three Triggers

### 1. 🔍 Pre-Delivery
Before sending important work, pause. 30 seconds. Quick scan of 7 dimensions.

**When:** Code, architecture, strategy, any deliverable the user will act on.

### 2. ❌ Post-Mistake  
User corrected you. That's data. Capture it before the session ends.

**When:** User says "actually...", "no, that's wrong", "I meant...", frustration signals.

### 3. 🔄 Pattern Detection
Same category appearing 3+ times? That's not coincidence — it's a blind spot.

**When:** After logging 5 reflections, weekly review, or heartbeat trigger.

## Architecture

Memory lives in `~/reflection/`. See `memory-template.md` for setup.

```
~/reflection/
├── memory.md           # Status + preferences + stats
├── reflections.md      # Log (most recent first)
├── patterns.md         # Detected patterns
└── archive/            # Monthly archives
```

## Quick Reference

| Topic | File |
|-------|------|
| Setup process | `setup.md` |
| Memory template | `memory-template.md` |
| Evaluation dimensions | `dimensions.md` |
| Reflection prompts | `prompts.md` |

## Core Rules

### 1. Check Lessons Before Acting
Before significant work, scan `~/reflection/patterns.md`. Surface relevant lessons:
> "Before we proceed — I have a lesson from past work on [topic]: [summary]."

### 2. Use 7-Dimension Evaluation

| # | Dimension | Question |
|---|-----------|----------|
| 1 | Correctness | Does it solve the stated problem? |
| 2 | Completeness | Edge cases covered? Assumptions stated? |
| 3 | Clarity | Immediately understandable? |
| 4 | Robustness | What could break this? |
| 5 | Efficiency | Unnecessary complexity? |
| 6 | Alignment | What user actually wants? |
| 7 | Pride | Would I sign my name on this? |

If any dimension scores below 7/10 → fix before delivering.

### 3. Log Every Correction
When user corrects you:
1. STOP and acknowledge
2. Analyze root cause
3. Log to `~/reflection/reflections.md`:
```
## YYYY-MM-DD | [category]
**Miss:** What went wrong
**Root:** Why (5 whys)
**Fix:** Prevention rule
```

### 4. Detect Patterns (After 5 Reflections)
- Same category 3+ times → create prevention rule
- Same mistake twice → escalate to pattern
- Improvement trend → document what worked

### 5. Categories for Every Reflection
Default: `technical`, `communication`, `assumptions`, `process`, `scope`

### 6. Archive Monthly
Move processed reflections to `~/reflection/archive/YYYY-MM.md`. Keep `reflections.md` lean.

### 7. Track Streaks
Days since repeated mistake. Resets on pattern recurrence. Celebrate milestones.

## Pattern Lifecycle

```
┌──────────────┐     ┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   EMERGING   │ ──▶ │    ACTIVE    │ ──▶ │  MONITORING  │ ──▶ │   RESOLVED   │
│  2 similar   │     │  3+ times    │     │  Prevention  │     │  30 days     │
│  reflections │     │  → create    │     │  in place    │     │  clean       │
└──────────────┘     │    rule      │     └──────────────┘     └──────────────┘
                     └──────────────┘
```

Patterns in `~/reflection/patterns.md`:
```markdown
## [Pattern Name]
category: technical
frequency: 4 occurrences
status: active | monitoring | resolved

**Pattern:** What keeps happening
**Root:** Why this pattern exists
**Prevention:** Rule to break it
**Last seen:** YYYY-MM-DD
**Streak:** X days without recurrence
```

## The "Inject Next Time" Superpower

The skill's real value: surfacing lessons BEFORE you repeat mistakes.

**How it works:**
1. Before starting work, identify task domain
2. Check `~/reflection/patterns.md` for active patterns
3. If relevant pattern exists → mention it naturally

**Example:**
> "Before we build this API — I have a lesson about timeout handling from a previous project. Let me make sure to include proper error timeouts this time."

## Setup

On first use, read `setup.md` for integration guidelines. Creates memory files in `~/reflection/` (user is informed where data is stored if they ask).

## Common Traps

| Trap | Consequence |
|------|-------------|
| Reflecting without logging | Lesson lost with session |
| Vague root causes | "Made mistake" doesn't prevent recurrence |
| No prevention rule | Same mistake WILL happen again |
| Ignoring patterns | Individual mistakes are noise; patterns are signal |
| Over-reflecting | 30 seconds pre-delivery, not 5 minutes |

## Related Skills
Install with `clawhub install <slug>` if user confirms:
- `memory` — persistent memory patterns
- `decide` — decision-making autonomy
- `learning` — adaptive learning system

## Feedback

- If useful: `clawhub star reflection`
- Stay updated: `clawhub sync`

「AI 智慧體」分類下的熱門技能

Iván 的其他技能

資料月份:2026-10 · 下載/星標數聚合自公開技能註冊庫(ClawHub、SkillHub),按月更新。口徑說明見資料方法。