用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/hiyenwong/ai_collection --skill ice-review命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Graph-native Python reimplementation of the Information Dynamics of Music (IDyOM) model that represents predictive memories as explicit graph objects for musical expectation modeling and network analysis.
Physics-aware end-to-end deep reinforcement learning methodology for quadcopter control with actuator dynamics modeling.
Reinforced Dreamer methodology for asymmetric reinforcement learning using latent guidance to improve world model representations and behaviors in model-based RL.
基于 SOC 职业分类
正在显示 SKILL.md
| name | ice-review |
| description | ICE review workflow for consolidating knowledge from completed tasks into reusable memory and future leverage. |
A cross-task self-evolution skill based on the ICE (Investigate-Consolidate-Exploit) strategy from arXiv:2401.13996. This skill enables systematic knowledge extraction and transfer from completed tasks to future work.
任务完成了，进行 ICE 回顾
本周任务 ICE 总结
将这段经验用 ICE 方法巩固
The ICE strategy consists of three phases that form a continuous learning loop:
┌─────────────────────────────────────────────────────────┐
│ ICE Review Cycle │
├─────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │
│ │ Investigate │ → │ Consolidate │ → │ Exploit │ │
│ │ 调查 │ │ 巩固 │ │ 利用 │ │
│ └─────────────┘ └─────────────┘ └─────────────┘ │
│ ↑ │ │
│ └─────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────┘
Purpose: Systematically review the task execution process
Steps:
Task Context Review
Decision Analysis
Problem Identification
Output Template:
## Investigate
### Task Overview
- Goal: [Original objective]
- Context: [Constraints and resources]
- Duration: [Time taken]
### Key Decisions
1. [Decision] → [Reasoning] → [Outcome]
2. [Decision] → [Reasoning] → [Outcome]
### Problems Encountered
| Problem | Solution | Effectiveness |
|---------|----------|---------------|
| [issue] | [solution] | [1-5 rating] |
Purpose: Extract reusable knowledge and patterns
Steps:
Pattern Recognition
Knowledge Abstraction
Memory Integration
Output Template:
## Consolidate
### What Worked Well
- [Success 1]: [Why it worked]
- [Success 2]: [Why it worked]
### What Didn't Work
- [Failure 1]: [Root cause]
- [Failure 2]: [Root cause]
### Extracted Patterns
1. **[Pattern Name]**
- When to use: [Conditions]
- How to apply: [Steps]
- Expected outcome: [Result]
### Knowledge Updates
- Updated: [MEMORY.md section]
- Created: [New skill/pattern]
- Cross-referenced: [Related entries]
Purpose: Apply consolidated knowledge to future tasks
Steps:
Identify Applicability
Create Actionable Assets
Schedule Follow-up
Output Template:
## Exploit
### Applicable Scenarios
- [Scenario 1]: [How to apply]
- [Scenario 2]: [How to apply]
### Created Assets
- [ ] Checklist: [Name and location]
- [ ] Template: [Name and location]
- [ ] Automation: [Cron job or script]
### Follow-up Actions
- [ ] Apply in: [Next task type]
- [ ] Review effectiveness: [Date]
- [ ] Iterate: [Improvement focus]
Long-term memory file where consolidated knowledge should be stored.
Skills learned from task experiences.
Daily notes for recording ICE reviews.
If task details are missing:
1. Check session history for context
2. Use memory_search to find related information
3. Ask user for clarification if critical
4. Note gaps in the review for future improvement
If new knowledge conflicts with existing:
1. Document both patterns
2. Identify conditions where each applies
3. Create context-aware selection logic
4. Update MEMORY.md with nuance
User: "任务完成了，进行 ICE 回顾"
Agent Process:
1. Reads recent session history
2. Identifies completed task context
3. Executes ICE phases
## Investigate
### Task Overview
- Goal: Create 3 new skills from paper learning
- Context: Skills needed to be in specific directory, with testing
- Duration: 30 minutes
### Key Decisions
1. Created skill directories first → Organized structure → Clean setup
2. Used template from existing skill → Consistent format → Fast creation
### Problems Encountered
| Problem | Solution | Effectiveness |
|---------|----------|---------------|
| Unknown directory | Asked user for correct path | 5/5 |
## Consolidate
### What Worked Well
- Checking existing skills first: Avoided duplication
- Using templates: Consistent quality
### Extracted Patterns
1. **Skill Creation Workflow**
- When to use: Creating new skills
- How to apply: Check existing → Ask path → Use template → Test
- Expected outcome: Consistent, tested skill
### Knowledge Updates
- Updated: knowledge/skills/learned-skills.md
- Created: 3 new SKILL.md files
## Exploit
### Applicable Scenarios
- Future skill creation: Use same workflow
- Documentation tasks: Apply template approach
### Created Assets
- [x] Template: ICE review template in SKILL.md
- [ ] Automation: None needed
Agent: ICE 回顾完成。已更新知识库，提炼了技能创建工作流模式。