Back to Details

episode-complete

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Episode Complete

完成Episode

Complete and score a learning episode to extract patterns and update heuristics.
完成学习episode并为其评分，以提取模式并更新启发式规则。

Purpose

目的

Finalize an episode with outcome scoring, reflection generation, and pattern extraction for future retrieval.
通过结果评分、反思生成和模式提取来收尾episode，以便未来检索使用。

Steps

步骤

  1. Gather outcome data:
    • Final verdict (success, partial_success, failure)
    • Total time spent
    • Total tokens used (if applicable)
    • Key artifacts produced
    • Errors encountered
  2. Create TaskOutcome:
    rust
    let outcome = TaskOutcome {
    verdict: Verdict::Success,
    time_ms: total_time,
    tokens: total_tokens,
    artifacts: vec![/* paths to created/modified files */],
    errors: vec![/* any errors encountered */],
};
  3. Call complete_episode:
    rust
    memory.complete_episode(episode_id, outcome).await?;
  4. System processes:
    • Computes RewardScore based on:
      • Success/failure
      • Time efficiency
      • Code quality
    • Generates Reflection:
      • What worked well
      • What could be improved
      • Key learnings
    • Extracts Patterns:
      • Tool sequences
      • Decision points
      • Common pitfalls
  5. Update storage:
    • Store in Turso (permanent record)
    • Update redb cache
    • Index by task_type and timestamp
    • Update related patterns and heuristics
  6. Validation:
    • Verify episode was scored
    • Check patterns were extracted
    • Ensure heuristics were updated
  1. 收集结果数据:
    • 最终结论（success、partial_success、failure）
    • 总耗时
    • 总使用token数（如适用）
    • 生成的关键工件
    • 遇到的错误
  2. 创建TaskOutcome:
    rust
    let outcome = TaskOutcome {
    verdict: Verdict::Success,
    time_ms: total_time,
    tokens: total_tokens,
    artifacts: vec![/* paths to created/modified files */],
    errors: vec![/* any errors encountered */],
};
  3. 调用complete_episode:
    rust
    memory.complete_episode(episode_id, outcome).await?;
  4. 系统处理流程:
    • 基于以下指标计算RewardScore:
      • 成功/失败状态
      • 时间效率
      • 代码质量
    • 生成反思内容:
      • 哪些部分表现良好
      • 哪些部分可以改进
      • 关键学习点
    • 提取模式:
      • 工具使用序列
      • 决策节点
      • 常见陷阱
  5. 更新存储:
    • 存储到Turso（永久记录）
    • 更新redb缓存
    • 按task_type和时间戳建立索引
    • 更新相关模式与启发式规则
  6. 验证:
    • 确认episode已完成评分
    • 检查模式是否已提取
    • 确保启发式规则已更新

Pattern Types Extracted

提取的模式类型

  • ToolSequence: Common tool usage patterns
  • DecisionPoint: Key decision moments and outcomes
  • ErrorPattern: Common errors and resolutions
  • PerformancePattern: Optimization opportunities
  • ToolSequence: 常见工具使用模式
  • DecisionPoint: 关键决策时刻及结果
  • ErrorPattern: 常见错误与解决方案
  • PerformancePattern: 优化机会

Scoring Rubric

评分标准

  • Success: Task completed, tests pass, meets requirements
  • Partial Success: Task mostly complete, minor issues
  • Failure: Task incomplete, major issues, tests failing
  • Success: 任务完成，测试通过，满足需求
  • Partial Success: 任务基本完成，存在小问题
  • Failure: 任务未完成，存在重大问题，测试失败

Example

示例

rust
let outcome = TaskOutcome {
    verdict: Verdict::Success,
    time_ms: 45000,
    tokens: 12000,
    artifacts: vec![
        "src/storage/batch.rs".to_string(),
        "tests/integration/batch_test.rs".to_string(),
    ],
    errors: vec![],
};

memory.complete_episode(episode_id, outcome).await?;
rust
let outcome = TaskOutcome {
    verdict: Verdict::Success,
    time_ms: 45000,
    tokens: 12000,
    artifacts: vec![
        "src/storage/batch.rs".to_string(),
        "tests/integration/batch_test.rs".to_string(),
    ],
    errors: vec![],
};

memory.complete_episode(episode_id, outcome).await?;

Post-Completion

完成后操作

  • Patterns are now available for future retrieval
  • Heuristics updated for similar tasks
  • Episode stored for long-term learning
  • Embeddings computed (if service configured)
  • 提取的模式可用于未来检索
  • 针对相似任务更新启发式规则
  • Episode被存储用于长期学习
  • 计算嵌入向量（若服务已配置）