When to Use
- Need to check learning stage of a pattern or correction
- Want to identify emerging patterns from repeated corrections
- Need to coordinate promotion/demotion of patterns across memory tiers
- Integrating with correction‑logger and preference‑tracker for learning workflows
Architecture
NeverOnce 增强功能
- ✅ 有效性反馈集成：从增强correction-logger获取有效性分数，跟踪修正使用历史
- ✅ 动态阶段转换算法：基于有效性的自动阶段提升/降级
- 高有效性模式 → 加速确认
- 低有效性模式 → 自动降级或标记
- ✅ 反馈循环监控：跟踪模式有效性趋势，识别高/低效学习模式
- ✅ 学习速度计算：基于有效性和反馈趋势的学习速度评估
- ✅ 增强报告生成：模式有效性报告、反馈循环统计、学习进度跟踪
- ✅ 自动调整规则：基于置信度的自动阶段调整，减少人工干预
增强算法
- 阶段置信度计算：
confidence = (repetition_count * 0.4) + (effectiveness_score * 0.4) + (time_factor * 0.2)
- 学习速度评估：
learning_speed = (help_ratio * 0.6) + (effectiveness_trend * 0.4)
- 自动调整阈值：
- 自动提升: confidence ≥ 0.8
- 自动降级: effectiveness ≤ 0.2
集成说明
- 依赖: 增强correction-logger v2.0.0+（可选，但推荐）
- 数据源: 从纠正记录器获取有效性分数和反馈历史
- 兼容性: 原有API完全兼容，新增增强方法可选使用
The plugin provides a LearningCoordinator class that:
- Monitors learning signals – watches corrections and preferences via their respective adapters.
- Manages learning stages – tracks patterns through stages: tentative, emerging, pending, confirmed, archived.
- Coordinates promotion/demotion – applies rules for when to move patterns between stages and tiers.
- Exposes learning statistics – reports on learning progress and pattern evolution.
The plugin does not store its own data; it relies on existing adapters (correction‑logger, preference‑tracker) and the learning‑rules file (learning.md).
Installation
clawhub install learning-coordinator
Or manually copy the plugin directory to your workspace skills folder.
Configuration
Default configuration loads the learning rules file and references other adapters:
learning_rules_file: ~/self-improving/learning.md
correction_adapter: "correction_logger"
preference_adapter: "preference_tracker"
auto_create: true
API Reference
LearningCoordinator Class
from learning_coordinator import LearningCoordinator
coordinator = LearningCoordinator(config=None)
# Get learning statistics
stats = coordinator.get_learning_stats()
# Check emerging patterns
emerging = coordinator.get_emerging_patterns(threshold=2)
# Promote a pattern (after user confirmation)
result = coordinator.promote_pattern(correction_ids=[1, 2, 3], new_status="confirmed")
# Get stage counts
stage_counts = coordinator.get_stage_counts()
# Health check
health = coordinator.health_check()
Adapter Interface
The plugin includes a LearningCoordinatorAdapter that conforms to the star‑architecture MemoryAdapter base class, providing:
health_check() – reports availability of required adapters and rule fileget_stats() – returns learning statistics (stage counts, promotion rates, etc.)search(query, limit=10) – searches across learning rules and pattern descriptionssync() – ensures coordinator state is in sync (no‑op for this plugin)get_learning_stats(), get_emerging_patterns(), promote_pattern() – convenience methods
Integration with Star Architecture
Once installed and its adapter is registered in the star‑architecture registry, other plugins can query learning coordination via the adapter factory:
from integration.adapter_factory import AdapterFactory
factory = AdapterFactory()
coordinator_adapter = factory.get_adapter("learning_coordinator")
if coordinator_adapter:
stats = coordinator_adapter.get_learning_stats()
emerging = coordinator_adapter.get_emerging_patterns(threshold=2)
Learning Rules
The plugin reads the learning.md file (see SIPA skill) to obtain:
- Trigger definitions – what counts as a learning signal
- Confirmation flow – how and when to ask for user confirmation
- Stage evolution – rules for moving between stages
- Anti‑patterns – what not to learn
The file is treated as read‑only; modifications must be made manually.
Troubleshooting
Missing adapters – If correction‑logger or preference‑tracker adapters are unavailable, the coordinator will operate with limited functionality.
Rule file not found – If learning.md does not exist, the plugin will create a minimal version based on the SIPA skill's default content.
Permission errors – Ensure the process has read access to the learning rules file.
Related Plugins
- correction‑logger – logs user corrections and system improvements
- preference‑tracker – manages user preferences and patterns
- heartbeat‑manager – manages heartbeat state and logs
- reflection‑logger – logs self‑reflection entries
Version History
- v0.1.0 – Initial split from SIPA skill, basic coordination, star‑architecture adapter.
错误码
| 错误码 | 描述 | 解决方案 |
|---|
| E001 | 未知错误 | 检查日志，联系开发者 |
| E002 | 配置错误 | 验证配置文件格式 |
| E003 | 依赖缺失 | 安装所需依赖包 |
