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

• - ✅ 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

1. 1. 阶段置信度计算：

   confidence = (repetition_count * 0.4) + (effectiveness_score * 0.4) + (time_factor * 0.2)
   

1. 2. 学习速度评估：

   learning_speed = (help_ratio * 0.6) + (effectiveness_trend * 0.4)
   

1. 3. 自动调整阈值：

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:

1. 1. Monitors learning signals – watches corrections and preferences via their respective adapters.
2. Manages learning stages – tracks patterns through stages: tentative, emerging, pending, confirmed, archived.
3. Coordinates promotion/demotion – applies rules for when to move patterns between stages and tiers.
4. 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

CODEBLOCK2

Or manually copy the plugin directory to your workspace skills folder.

Configuration

Default configuration loads the learning rules file and references other adapters:

CODEBLOCK3

API Reference

LearningCoordinator Class

CODEBLOCK4

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 file
• INLINECODE5 – returns learning statistics (stage counts, promotion rates, etc.)
• INLINECODE6 – searches across learning rules and pattern descriptions
• INLINECODE7 – ensures coordinator state is in sync (no‑op for this plugin)
• INLINECODE8, 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:

CODEBLOCK5

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