Learning Feedback Skill

Global learning skill that captures user corrections, feedback, and refined processing logic. Learning records are automatically bound to relevant skills for future application.

Features

Correction Learning

• Captures user corrections ("不对", "错了", "应该是")
• Records original vs corrected approach
• Binds learning to affected skills

Refinement Learning

• Captures optimization suggestions ("优化一下", "改进")
• Records processing logic refinements
• Applies to future similar tasks

Scene-Skill Mapping

• Records user-specified handling preferences
• Auto-applies on matching scenes
• Supports complex multi-skill workflows

Knowledge Application

• Pre-execution learning check
• Automatic correction application
• Learning effectiveness tracking

Trigger Patterns

"不对", "错了", "应该是", "正确的是"
"不是这样", "重新处理", "修改为"
"你应该...", "需要...", "改成..."
"更准确的做法是", "更好的方式是"
"优化一下", "改进", "调整"
"流程不对", "顺序错了", "逻辑有问题"
"先...再...", "应该先..."
"记住这个方式", "下次直接这样"
"按照上次的方式处理"

Output Format

┌─────────────────────────────────────────────────────┐
│ ✅ Learning Record Saved                            │
├─────────────────────────────────────────────────────┤
│ Record ID: {uuid}                                   │
│ Skill: {affected-skill}                             │
│ Type: {correction-type}                             │
├─────────────────────────────────────────────────────┤
│ Original: {brief-summary}                           │
│ Corrected: {brief-summary}                          │
├─────────────────────────────────────────────────────┤
│ This learning will be auto-applied in future tasks  │
└─────────────────────────────────────────────────────┘

Learning Types

Scene Mapping

Scene Detection

User Input → Keyword Extraction → Pattern Matching → Scene Match
                                                    ↓
                                            [Match] → Apply Recorded Workflow
                                            [No Match] → Normal Flow → Record if Confirmed

Scene Mapping Examples

场景: 查询TraceId链路
关键词: traceId, 链路, 调用链
技能: sls-log-audit
偏好: 顺序列表格式

→ 自动应用: 使用 sls-log-audit，输出顺序列表

Learning Storage

~/.flowmind/learning/
├── records/               # Learning records
│   ├── _global/          # Global learning
│   ├── code-review/      # Per-skill learning
│   └── log-audit/
├── skill-bindings.json   # Skill binding mapping
├── scene-mappings.json   # Scene-skill mappings
└── stats.json            # Learning statistics

Learning Integration

This skill is a global skill that enhances all other skills:

• Before executing any skill, check for applicable learning
• Apply learned corrections automatically
• Track learning effectiveness over time

Examples

Example 1: Output Format Correction

User: 查询 traceId abc123 的日志
FlowMind: [Returns tree format]
User: 不对，应该用顺序列表格式

FlowMind:
┌─────────────────────────────────────────────────────┐
│ ✅ Learning Record Saved                            │
├─────────────────────────────────────────────────────┤
│ Skill: log-audit                                    │
│ Type: output_format                                 │
│ Original: Tree format                               │
│ Corrected: Sequential list format                   │
├─────────────────────────────────────────────────────┤
│ Future log queries will use sequential list format  │
└─────────────────────────────────────────────────────┘

Example 2: Scene Mapping

User: 排查线上问题先查错误日志，再查链路，最后查代码

FlowMind:
┌─────────────────────────────────────────────────────┐
│ ✅ Scene Mapping Saved                              │
├─────────────────────────────────────────────────────┤
│ Scene: 线上问题排查                                  │
│ Keywords: 排查, 线上问题, 故障排查                    │
│ Skill Sequence:                                     │
│  1. log-audit (错误日志)                            │
│  2. log-audit (链路分析)                            │
│  3. code-review (代码审查)                          │
├─────────────────────────────────────────────────────┤
│ Next time will auto-apply this workflow             │
└─────────────────────────────────────────────────────┘

Configuration

{
  "learning-feedback": {
    "autoApply": true,
    "confidenceThreshold": 0.7,
    "maxRecordsPerSkill": 100,
    "archiveAfterDays": 90
  }
}