knowledge-distillation

# Knowledge Distillation ## 概述 Analyze iteration data to extract reusable lessons, categorize them into the most relevant skills, and perform idempotent updates to Skill.md files. ## 能力维度 ### Classification Determine which specific skill (e.g., `python_engineering`, `schema_design`) a lesson most accurately belongs to. ### Deduplication Analyze existing "Lessons Learned" sections in target skills to ensure new entries provide unique value and don't replicate existing knowledge. ### Skill Injection Precisely inject or merge structured lessons into the target Skill.md files while preserving the overall file structure. ## 工具支持 - **markdown**: Parse and generate formatted Markdown for Skill sections. - **file_read**: Load existing Skill.md files for context and deduplication checks. - **file_write**: Apply targeted updates to Skill.md files. - **code_search**: Cross-reference lessons with actual code implementation for validation. ## 约束条件 - must_identify_target_skill_accurately - must_check_existing_lessons_to_prevent_duplicates - must_maintain_consistent_lesson_format - must_only_modify_lessons_learned_section ## 能力等级 ### Level 1 Identifies obvious lessons and adds them to a generic "General" skill or log. ### Level 2 Correctly categorizes lessons into specific functional skills and avoids exact duplicates. ### Level 3 Performs semantic deduplication, merges related lessons into more powerful generalized rules, and maintains a highly optimized Skill library.