learned-patterns

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Learned Patterns Management

已学习模式管理

Manage the auto-learning pattern library. View, list, archive, and adjust confidence.

管理自动学习模式库。查看、列出、归档并调整置信度。

Summary

概述

Goal: View, list, archive, boost, or penalize learned patterns in the auto-learning pattern library.

List patterns:

node scripts/list-patterns.cjs [category] [--low|--high]

View details:
```
node scripts/view-pattern.cjs <id>
```

Adjust confidence:

node scripts/adjust-confidence.cjs <id> boost|penalize

(+10% / -15%)

Archive:

node scripts/archive-pattern.cjs <id> [reason]

Stats:
```
node scripts/pattern-stats.cjs
```

Key Principles:

Patterns auto-decay after 30 days unused; below 20% confidence = auto-archived
Lifecycle: Detection (40-80%) -> Confirmation -> Injection -> Feedback Loop -> Decay
Max 5 patterns injected per context (~400 tokens)

目标： 查看、列出、归档、提升或降低自动学习模式库中的已学习模式。

列出模式：

node scripts/list-patterns.cjs [category] [--low|--high]

查看详情：
```
node scripts/view-pattern.cjs <id>
```

调整置信度：

node scripts/adjust-confidence.cjs <id> boost|penalize

（+10% / -15%）

归档：

node scripts/archive-pattern.cjs <id> [reason]

统计信息：
```
node scripts/pattern-stats.cjs
```

核心原则：

未使用的模式会在30天后自动衰减；置信度低于20%时会自动归档
生命周期：检测（40-80%）→ 确认 → 注入 → 反馈循环 → 衰减
每个上下文最多注入5个模式（约400个tokens）

Quick Commands

快速命令

/learned-patterns                    # List all active patterns
/learned-patterns list [category]    # List patterns (optionally filtered)
/learned-patterns list --low|--high  # Filter by confidence
/learned-patterns view <id>          # View pattern details
/learned-patterns archive <id>       # Archive a pattern
/learned-patterns boost <id>         # Increase confidence (+10%)
/learned-patterns penalize <id>      # Decrease confidence (-15%)
/learned-patterns stats              # Show pattern statistics

/learned-patterns                    # 列出所有活跃模式
/learned-patterns list [category]    # 列出模式（可选择过滤）
/learned-patterns list --low|--high  # 按置信度过滤
/learned-patterns view <id>          # 查看模式详情
/learned-patterns archive <id>       # 归档模式
/learned-patterns boost <id>         # 提升置信度（+10%）
/learned-patterns penalize <id>      # 降低置信度（-15%）
/learned-patterns stats              # 显示模式统计信息

Script Execution

脚本执行

Action	Script
List	`node .claude/skills/learned-patterns/scripts/list-patterns.cjs [category] [--low\|--high]`
View	`node .claude/skills/learned-patterns/scripts/view-pattern.cjs <id>`
Archive	`node .claude/skills/learned-patterns/scripts/archive-pattern.cjs <id> [reason]`
Boost/Penalize	`node .claude/skills/learned-patterns/scripts/adjust-confidence.cjs <id> boost\|penalize`
Stats	`node .claude/skills/learned-patterns/scripts/pattern-stats.cjs`

操作	脚本
列出	`node .claude/skills/learned-patterns/scripts/list-patterns.cjs [category] [--low\|--high]`
查看	`node .claude/skills/learned-patterns/scripts/view-pattern.cjs <id>`
归档	`node .claude/skills/learned-patterns/scripts/archive-pattern.cjs <id> [reason]`
提升/降低置信度	`node .claude/skills/learned-patterns/scripts/adjust-confidence.cjs <id> boost\|penalize`
统计信息	`node .claude/skills/learned-patterns/scripts/pattern-stats.cjs`

Pattern Lifecycle

模式生命周期

DETECTION (UserPromptSubmit)
  └── Confidence: 40% (implicit) or 80% (explicit /learn)

CONFIRMATION
  └── User confirms → saved | User rejects → discarded

INJECTION (SessionStart/PreToolUse)
  └── Relevant patterns injected (max 5, ~400 tokens)

FEEDBACK LOOP
  └── Followed → +5% | Ignored → -10%

DECAY & PRUNING
  └── 30 days unused → decay | Below 20% → auto-archived

检测（UserPromptSubmit）
  └── 置信度：40%（隐式）或80%（显式/learn）

确认
  └── 用户确认 → 保存 | 用户拒绝 → 丢弃

注入（SessionStart/PreToolUse）
  └── 注入相关模式（最多5个，约400个tokens）

反馈循环
  └── 被遵循 → +5% | 被忽略 → -10%

衰减与清理
  └── 30天未使用 → 衰减 | 低于20% → 自动归档

Confidence Thresholds

置信度阈值

Range	Behavior
80-100%	Always injected when relevant
50-79%	Injected with context match
30-49%	Injected only on strong match
20-29%	Candidate for review
< 20%	Auto-archived

范围	行为
80-100%	相关时始终注入
50-79%	匹配上下文时注入
30-49%	仅强匹配时注入
20-29%	待审核候选
< 20%	自动归档

Pattern Schema

模式 Schema

yaml

id: pat_abc123
category: backend
type: anti-pattern | best-practice | preference | convention
trigger:
  keywords: [validation, exception]
  file_patterns: ["*CommandHandler.cs"]
content:
  wrong: "throw new ValidationException()"
  right: "return PlatformValidationResult.Invalid()"
  rationale: "Framework uses result pattern, not exceptions"
metadata:
  source: explicit-teaching | user-correction
  confidence: 0.75
  first_seen: 2025-01-10
  occurrences: 3
  confirmations: 2
  conflicts: 0