add-coder 0.2.6 → 0.2.8

package/README.en.md

@@ -27,7 +27,26 @@ Traditional AI development: Chat → Generate code → Dig through chat history
 | Auditing relies on developer discipline | The **MCP audit toolchain** automatically records every operation; system gateways enforce checks |
 | No traceability | Audit events are naturally linked: Plan → Spec → Task → Step → Tool Call, forming a complete evidence chain |
 
-### ② Gateway-Driven, Not Free-Form Conversation
+### ② Prompt Cache Native — ¥218/mo, 98% Savings
+
+The ADD paradigm isn't just methodology — its structured Step workflow naturally aligns with DeepSeek's Prompt Cache prefix-matching mechanism, delivering extreme token cost efficiency. **Real-world billing validation**:
+
+| Metric | Value |
+|--------|-------|
+| July actual DeepSeek bill | **¥218.35** |
+| Theoretical cost without cache | ¥11,100 |
+| Cache hit rate | **99.31%** |
+| Cache hit vs miss price gap | **120x** (¥0.025/M vs ¥3/M) |
+| Total cost savings | **98.1%** |
+
+```
+Traditional IDE free chat:  cache hit rate 85–91%, ~5,000 MISS tokens/req
+ADD paradigm + Qoder:       cache hit rate 99.31%, only 2,426 MISS tokens/req
+```
+
+> 📊 [Full analysis report](./docs/ADD范式缓存命中分析报告.md) — 4 Mermaid diagrams, 17-day daily data, cross-IDE comparison, and cost modeling.
+
+### ③ Gateway-Driven, Not Free-Form Conversation
 
 Traditional AI coding is "you say, I do" — quality depends entirely on the LLM's state that day. add-coder embeds **dual quality gateways** into the architecture:
 
@@ -38,7 +57,7 @@ RAHS (Runtime Architecture Health Score) — Runtime architecture health, < 90%
 
 These are not "suggestions" — they are **architectural blocks**. A Step cannot advance without passing its gateway.
 
-### ③ Cross-Session Memory, Not Per-Session Amnesia
+### ④ Cross-Session Memory, Not Per-Session Amnesia
 
 The fatal flaw of AI conversations: architectural decisions from last session, bugs fixed, agreements reached — all forgotten in the next conversation. add-coder solves this at the architecture level:
 
@@ -46,7 +65,7 @@ The fatal flaw of AI conversations: architectural decisions from last session, b
 - **Plan Index** — All Plans are centrally indexed via `index.md`, supporting fuzzy-match quick lookup
 - **DevLog Timeline** — Every operation is written to the `{YYYY-MM}/{DD}/` timeline, enabling full historical state traceability
 
-### ④ Policy-Update-Loop: Self-Evolving Governance (the scaffold itself does not include this architectural capability; a DEMO repo will be provided next to better illustrate the Policy-Update-Loop and Report system)
+### ⑤ Policy-Update-Loop: Self-Evolving Governance (the scaffold itself does not include this architectural capability; a DEMO repo will be provided next to better illustrate the Policy-Update-Loop and Report system)
 
 Not a static template, but a **closed-loop adaptive system**:
 
@@ -56,7 +75,7 @@ Execute → Audit → Boundary Report → Rule Adjustment → Next Execution
 
 Runtime-generated Reports feed back into governance rules, enabling continuous evolution of governance strategies.
 
-### ⑤ Multi-IDE Hooks as the Governance Layer
+### ⑥ Multi-IDE Hooks as the Governance Layer
 
 Hooks are not "notification push" — they are the **IDE runtime interception layer**:
 

package/README.md

@@ -27,7 +27,26 @@ npx add-coder init
 | 审计靠开发者自觉记录 | **MCP 审计工具链** 自动记录每次操作，系统闸门强制检查 |
 | 无关联性 | 审计事件天然关联 Plan → Spec → Task → Step → Tool Call，形成完整证据链 |
 
-### ② 门禁驱动，而非自由对话
+### ② Prompt Cache 原生友好 — 月费 ¥218，节省 98%
+
+ADD 范式不仅是方法论——它的结构化 Step 流程天然适配 DeepSeek Prompt Cache 的前缀匹配机制，带来极致的 Token 成本效率。**实测数据验证**：
+
+| 指标 | 数值 |
+|------|------|
+| DeepSeek 7月实际账单 | **¥218.35** |
+| 若无 Prompt Cache 理论费用 | ¥11,100 |
+| Cache 命中率 | **99.31%** |
+| 缓存命中 vs 未命中价差 | **120 倍**（¥0.025/M vs ¥3/M） |
+| 总费用节省 | **98.1%** |
+
+```
+传统 IDE 自由对话:  cache 命中率 85-91%, 每次请求 MISS 5,000 tokens
+ADD 范式 + Qoder:     cache 命中率 99.31%, 每次请求 MISS 仅 2,426 tokens
+```
+
+> 📊 [完整分析报告](./docs/ADD范式缓存命中分析报告.md) — 含 4 张 Mermaid 图表、17 天逐日数据、跨 IDE 对比与成本建模。
+
+### ③ 门禁驱动，而非自由对话
 
 传统 AI coding 是「你说我做」，质量完全依赖 LLM 当天状态。add-coder 在架构中嵌入了 **双质量闸门**：
 
@@ -38,7 +57,7 @@ RAHS (Runtime Architecture Health Score) — 运行时架构健康度，< 90% BL
 
 这不是「建议」，是**架构阻断** — 不通过闸门的 Step 无法推进到下一步。
 
-### ③ 跨轮记忆，而非每轮失忆
+### ④ 跨轮记忆，而非每轮失忆
 
 AI 对话的致命缺陷：上次讨论的架构决策、已修复的 Bug、达成的约定，下轮对话全部遗忘。add-coder 在架构层面解决：
 
@@ -46,7 +65,7 @@ AI 对话的致命缺陷：上次讨论的架构决策、已修复的 Bug、达
 - **Plan 索引** — 所有 Plan 通过 `index.md` 集中索引，支持模糊匹配快速定位
 - **DevLog 时序记录** — 每一步操作写入 `{YYYY-MM}/{DD}/` 时间轴，可回溯任意历史状态
 
-### ④ Policy-Update-Loop：治理自我进化(脚手架不包含此架构能力,接下来会给到DEMO仓库让大家更好理解Policy-Update-Loop和Report体系)
+### ⑤ Policy-Update-Loop：治理自我进化(脚手架不包含此架构能力,接下来会给到DEMO仓库让大家更好理解Policy-Update-Loop和Report体系)
 
 不是静态模板，而是**闭环自适应系统**：
 
@@ -56,7 +75,7 @@ AI 对话的致命缺陷：上次讨论的架构决策、已修复的 Bug、达
 
 运行时产生的 Report 会反过来更新 governance rules，实现治理策略的持续进化。
 
-### ⑤ 多 IDE 的 Hook 即治理层
+### ⑥ 多 IDE 的 Hook 即治理层
 
 hook 不是「通知推送」，而是 **ADD 范式在 IDE agent 生命周期中的 17 个确定性治理卡位**。每个 IDE（Claude Code / Qoder CN / VS Code Copilot / Trae / Codex）有各自的 hook 机制，但治理逻辑统一——架构一致，适配层不同。
 
@@ -134,7 +153,21 @@ npx add-coder init
 
 ## MCP 审计工具链
 
-`init` 自动部署 MCP 服务器 (`mcp-server.ts`) 到项目中，IDE 通过 `mcp.json` 加载。提供以下审计与治理工具：
+`init` 自动部署 MCP 服务器 (`mcp-server.ts`) 到项目中，IDE 通过 `mcp.json` 加载。基于 **MCP 协议六大能力**（四大原语 + 两个横切）：
+
+```
+MCP 能力            方向              当前状态     说明
+─────────────────────────────────────────────────────────────
+Tools              Client→Server     ✅ 已实现     17 个审计与治理工具（pull 模式）
+Resources+Sub      Client←Server     🔜 本轮       Plan/Review/Route/Task 状态实时推送
+Notifications      Server→Client     🔜 本轮       HITL 就绪通知 / Hook 结果推送
+Sampling           Server→Client     🔜 本轮       服务端回调 AI 生成 Review
+── 横切 ──
+Elicitation        Server→Client     🔜 本轮       向用户请求 HITL 确认/风险输入
+Tasks (实验性)     双向              🔜 本轮       长任务持久化 + 状态追踪
+```
+
+**当前已实现的 17 个 Tools**：
 
 | 工具 | 用途 | 触发场景 |
 |------|------|---------|
@@ -148,7 +181,7 @@ npx add-coder init
 | `check_spec_sync` | Spec 文档勾选状态与代码一致性 | Spec 执行后 |
 | `find_related_docs` | 检索相关架构/规范文档 | 语境理解 |
 
-> 完整工具列表见 [MCP 工具链规范](https://github.com/xiaomingming92/codein2027/blob/main/docs/大田精准耕播智能决策系统/knowledge/02-规范/%E3%80%8A%E5%BC%80%E5%8F%91%E6%93%8D%E4%BD%9C%E5%AE%A1%E8%AE%A1%E5%AD%98%E6%A1%A3%E8%A7%84%E8%8C%83%E3%80%8B.md)。
+> 完整六能力架构设计见 [MCP 重构 Plan](.qoder/plans/2026-07/23/add-coder-mcp-restructure-plan-v1.md)。
 
 ## 架构全景
 
@@ -249,7 +282,26 @@ Traditional AI development: Chat → Generate code → Dig through chat history
 | Auditing relies on developer discipline | The **MCP audit toolchain** automatically records every operation; system gateways enforce checks |
 | No traceability | Audit events are naturally linked: Plan → Spec → Task → Step → Tool Call, forming a complete evidence chain |
 
-### ② Gateway-Driven, Not Free-Form Conversation
+### ② Prompt Cache Native — ¥218/mo, 98% Savings
+
+The ADD paradigm isn't just methodology — its structured Step workflow naturally aligns with DeepSeek's Prompt Cache prefix-matching mechanism, delivering extreme token cost efficiency. **Real-world billing validation**:
+
+| Metric | Value |
+|--------|-------|
+| July actual DeepSeek bill | **¥218.35** |
+| Theoretical cost without cache | ¥11,100 |
+| Cache hit rate | **99.31%** |
+| Cache hit vs miss price gap | **120x** (¥0.025/M vs ¥3/M) |
+| Total cost savings | **98.1%** |
+
+```
+Traditional IDE free chat:  cache hit rate 85–91%, ~5,000 MISS tokens/req
+ADD paradigm + Qoder:       cache hit rate 99.31%, only 2,426 MISS tokens/req
+```
+
+> 📊 [Full analysis report](./docs/ADD范式缓存命中分析报告.md) — 4 Mermaid diagrams, 17-day daily data, cross-IDE comparison, and cost modeling.
+
+### ③ Gateway-Driven, Not Free-Form Conversation
 
 Traditional AI coding is "you say, I do" — quality depends entirely on the LLM's state that day. add-coder embeds **dual quality gateways** into the architecture:
 
@@ -260,7 +312,7 @@ RAHS (Runtime Architecture Health Score) — Runtime architecture health, < 90%
 
 These are not "suggestions" — they are **architectural blocks**. A Step cannot advance without passing its gateway.
 
-### ③ Cross-Session Memory, Not Per-Session Amnesia
+### ④ Cross-Session Memory, Not Per-Session Amnesia
 
 The fatal flaw of AI conversations: architectural decisions from last session, bugs fixed, agreements reached — all forgotten in the next conversation. add-coder solves this at the architecture level:
 
@@ -268,7 +320,7 @@ The fatal flaw of AI conversations: architectural decisions from last session, b
 - **Plan Index** — All Plans are centrally indexed via `index.md`, supporting fuzzy-match quick lookup
 - **DevLog Timeline** — Every operation is written to the `{YYYY-MM}/{DD}/` timeline, enabling full historical state traceability
 
-### ④ Policy-Update-Loop: Self-Evolving Governance (the scaffold itself does not include this architectural capability; a DEMO repo will be provided next to better illustrate the Policy-Update-Loop and Report system)
+### ⑤ Policy-Update-Loop: Self-Evolving Governance (the scaffold itself does not include this architectural capability; a DEMO repo will be provided next to better illustrate the Policy-Update-Loop and Report system)
 
 Not a static template, but a **closed-loop adaptive system**:
 
@@ -278,7 +330,7 @@ Execute → Audit → Boundary Report → Rule Adjustment → Next Execution
 
 Runtime-generated Reports feed back into governance rules, enabling continuous evolution of governance strategies.
 
-### ⑤ Multi-IDE Hooks as the Governance Layer
+### ⑥ Multi-IDE Hooks as the Governance Layer
 
 Hooks are not "notification push" — they are the **IDE runtime interception layer**:
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "add-coder",
-  "version": "0.2.6",
+  "version": "0.2.8",
   "private": false,
   "description": "[codein2027](https://github.com/xiaomingming92/codein2027) - A complete scaffolding for building the ADD programming paradigm, the implementation layer for AI code governance. Core principle: Audit as Infrastructure. Breaks the black-box programming process and cross-session amnesia, evolving programming paradigms into an auditable, traceable, and convergent new era. npx-ready.\n\n[codein2027](https://github.com/xiaomingming92/codein2027) 快速构建 ADD 编程范式的完整脚手架——AI 代码治理的落地方案。以「审计即基础设施」为核心，彻底打破编程过程黑盒与跨轮失忆，让编程范式进化为可审计、可追溯、可收敛的新时代。npx 即用，人人可体验。",
   "repository": {

package/templates/core/scripts/mcp-server.ts

@@ -2847,11 +2847,11 @@ server.registerTool(
       let specDirName = ""
       let specContent = ""
       let tasksContent = ""
-      // 从 Plan 绑定或 §7 提取
-      const specRef = planContent.match(/Spec:\s*\.(qoder|claude|add|vscode)\/specs\/([^/\s]+)/)
+      // 从 Plan 绑定或 §7 提取（兼容 Spec: / Spec | / `path` 三种格式）
+      const specRef = planContent.match(/Spec[:|\s`]+\.?(qoder|claude|add|vscode)\/specs\/([^/`\s]+)/)
       if (specRef) specDirName = specRef[2]
       if (!specDirName) {
-        const taskRef = planContent.match(/Tasks:\s*\.(qoder|claude|add|vscode)\/specs\/([^/\s]+)/)
+        const taskRef = planContent.match(/Tasks[:|\s`]+\.?(qoder|claude|add|vscode)\/specs\/([^/`\s]+)/)
         if (taskRef) specDirName = taskRef[2]
       }
       if (!specDirName) {