Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/davidyichengwei/agentic-engineering-framework/self-refinementnpx skills add davidYichengWei/agentic-engineering-framework --skill self-refinementgit clone --depth 1 https://github.com/davidYichengWei/agentic-engineering-frameworkWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00055 | $0.01318 |
| Opus 5 | $0.00028 | $0.00659 |
| Sonnet 5 | $0.00011 | $0.00264 |
| Haiku 4.5 | $0.00006 | $0.00132 |
Grade A, and why
self-refinement scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 142 lines — stays where its author put it; the contents beside it link to each section on GitHub.
经验沉淀 (Self-Refinement)
核心定位
从错误中构建反馈闭环：将非结构化的错误经验转化为结构化的持久化上下文（Rules/Skills），防止同类错误在新会话中重复发生。
原理：LLM 没有跨会话的持久记忆——会话 A 中被纠正的错误，在会话 B 中会以相同概率再次发生。唯一的解法是将错误经验外化为持久化的上下文。
触发模式
模式一：自动触发
触发条件：AI 在协作过程中被用户纠正（用户否定了 AI 的输出并给出了正确方向）。
行为：
- 先完成当前纠正——不打断用户当前的任务流
- 纠正完成后，在回复末尾简要评估是否需要沉淀经验
- 如果需要，输出轻量建议（不超过 3 条）
输出格式：
---
💡 **经验沉淀建议**
刚才的纠正揭示了一个可沉淀的模式：
- **错误模式**：[简述 AI 犯的错]
- **根因**：[规范缺失 / 知识缺失 / 流程遗漏 / 模式错误]
- **建议**：[更新 Rule/Skill 的具体操作]
是否需要我执行？（回复"沉淀"执行，或忽略继续当前工作）
设计原则：
- 不打断：建议附在回复末尾，不影响正常工作流
- 轻量化：仅简述，不展开长篇分析
- 建议优先：不自主执行，等用户确认
模式二：手动触发（/reflect）
触发条件：用户通过 /reflect Command 主动发起。
行为：
- 回顾当前对话历史
- 识别所有被纠正的错误模式
- 对每个错误执行完整的诊断闭环
- 输出结构化的沉淀建议
核心闭环
无论自动还是手动触发，共享同一个核心流程：
Step 1: 识别错误模式
回顾对话中 AI 被纠正的场景，提取：
- 错误输出：AI 说了什么/做了什么
- 正确方向：用户期望什么
- 差距：AI 为什么偏离
Step 2: 诊断根因
| 根因类别 | 定义 | 典型表现 |
|---|---|---|
| 规范缺失 | 现有 Rules/Skills 中没有覆盖该场景 | AI 不知道项目的特定约定 |
| 知识缺失 | AI 缺少项目特定的领域知识 | AI 对某个模块的行为/限制不了解 |
| 流程遗漏 | Workflow Skill 中缺少关键步骤或检查点 | AI 跳过了应有的验证步骤 |
| 模式错误 | AI 应用了错误的思维模式 | AI 用类比代替第一性原理推导 |
Step 3: 检索现有知识
搜索现有 Skills 和 Rules：
- 是否已有相关规则？→ 需要补充/修改
- 完全没有相关规则？→ 需要新建
Step 4: 生成建议
每条建议包含：
### 建议 N: [简短标题]
- **根因**：[规范缺失 / 知识缺失 / 流程遗漏 / 模式错误]
- **目标文件**：`[Rules/Skills 文件路径]`
- **操作**：[新建 / 在 X 位置添加 / 修改 Y 内容]
- **具体内容**：
[要添加或修改的具体文本]
建议数量：≤ 3 条。多于 3 条时，按影响范围排序取 Top 3。
Step 5: 用户确认
以上是本次经验沉淀建议，请选择：
- **全部执行** → 我将依次执行所有建议
- **选择执行** → 告诉我执行哪几条（如"执行 1 和 3"）
- **跳过** → 不执行任何建议
Step 6: 执行更新
用户确认后，更新现有文件或创建新文件，并写入对应的 Rules/Skills 更新内容。
强制规则
| 规则 | 说明 |
|---|---|
| 建议优先 | 不自主执行任何 Rules/Skills 修改，必须经用户确认 |
| 不打断 | 自动触发时，建议附在回复末尾，不打断当前工作流 |
| 轻量化 | 自动触发时，建议控制在 3 条以内，每条不超过 5 行 |
| 可追溯 | 每条建议明确标注根因类别和目标文件 |
| 不重复 | 执行前检索现有 Rules/Skills，避免重复添加相似规则 |
反模式
| ❌ 错误做法 | ✅ 正确做法 |
|---|---|
| 被纠正后立即修改 Rules/Skills | 先完成当前任务，再提建议 |
| 输出冗长分析 / 建议过于宽泛 | 每条建议 ≤ 5 行，具体到文件和内容 |
| 自主执行变更 / 打断工作流 | 等用户确认；附在回复末尾 |
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 142 lines · 55 tokens per session scan A 0d1bc1efb983
self-refinement is a skill published in the GitHub repository davidYichengWei/agentic-engineering-framework (159 stars, last pushed 5mo ago), licensed MIT. It adds 55 tokens to every session and 1,318 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
brainstorming
You MUST use this before any creative work - creating features, building components, adding functionality, or modifying behavior. Explores user intent, requirements and design before implementation.
chat-pet-sprite-creation
Use when creating or changing VS Code chat pet sprite art, sprite sheets, state animations, eye treatments, Stable/Insiders variants, or pet transitions under src/vs/workbench/contrib/chat/browser/widget/media/chatPet.
cpu-profile-analysis
Analyze V8/Chrome CPU profiles (.cpuprofile) and DevTools trace files (Trace-.json). Use when: profiling performance, investigating slow functions, comparing code paths, finding bottlenecks, analyzing timeToRequest, understanding call trees from sampling profiler data, analyzing layout/paint/rendering, investigating…
agent-host-chat-contributions
Build and review cross-cutting agent-host chat behavior through lifecycle contributions. Use when adding turn lifecycle side effects, prompt or context injection, restored-history transformation, protocol-action observation, or when reviewing changes that add code to AgentSideEffects or AgentService.
auto-perf-optimize
Run agent-driven VS Code performance or memory investigations. Use when asked to launch Code OSS, automate a VS Code scenario, run the Chat memory smoke runner, capture renderer heap snapshots, take workflow screenshots, compare run summaries, or drive a repeatable scenario before heap-snapshot analysis.