@hongmaple0820/scale-engine 0.11.1 → 0.12.0

package/README.en.md

@@ -1,14 +1,14 @@
 <p align="center">
-  <img src="https://img.shields.io/badge/version-0.10.1-orange?style=flat-square" alt="version" />
+  <img src="https://img.shields.io/badge/version-0.11.2-orange?style=flat-square" alt="version" />
   <img src="https://img.shields.io/badge/platforms-11-blue?style=flat-square" alt="platforms" />
   <img src="https://img.shields.io/badge/agents-12-blue?style=flat-square" alt="agents" />
   <img src="https://img.shields.io/badge/workflows-10-green?style=flat-square" alt="workflows" />
-  <img src="https://img.shields.io/badge/detectors-9-red?style=flat-square" alt="detectors" />
-  <img src="https://img.shields.io/badge/tests-461-passing-brightgreen?style=flat-square" alt="tests" />
-  <img src="https://img.shields.io/badge/npm-0.10.1-cb3837?style=flat-square&logo=npm" alt="npm" />
+  <img src="https://img.shields.io/badge/detectors-19-red?style=flat-square" alt="detectors" />
+  <img src="https://img.shields.io/badge/tests-499-passing-brightgreen?style=flat-square" alt="tests" />
+  <img src="https://img.shields.io/badge/npm-0.11.2-cb3837?style=flat-square&logo=npm" alt="npm" />
 </p>
 
-# SCALE Engine v0.10.1
+# SCALE Engine v0.11.2
 
 SCALE Engine is an AI engineering workflow runtime for agentic coding tools. It turns prompt-level engineering rules into stateful workflow gates, persisted evidence, review records, and release checks.
 
@@ -28,19 +28,42 @@ Prompt instructions are advisory. Production engineering needs mechanisms:
 
 ## Current Release
 
-v0.10.1 hardens the phase-aligned delivery workflow:
+v0.11.1 introduces four priority improvements:
+
+### Phase Commands FSM Blocking
+- `canTransition` + `process.exit(1)` ensures FSM guard failures block execution, not continue
+- define/plan/build/verify phases add clear blocking prompts
+
+### OWASP Top 10 Detector
+- New `OWASPDetector` covers SQL injection, XSS, path traversal, SSRF, Auth Bypass, weak crypto, CORS misconfiguration, CSRF, file upload, sensitive data exposure
+- 19 security detection patterns, auto-recognizes regex definitions to avoid false positives
+
+### Browser QA Capability
+- `BrowserQACapability` wraps Playwright MCP tools
+- Supports navigation, click, screenshot, console check, E2E test flows
+
+### L6 Evolution Self-Improve Loop
+- `LessonExtractor` extracts reusable lessons from session Defect events
+- `SelfImproveEngine` implements `Defect → Lesson → Rule → Hook` promotion pipeline
+- New CLI commands: `scale evolution extract/improve/report/hooks`
+
+---
+
+**Complete phase-aligned delivery workflow**:
 
 - `define -> plan -> build -> verify -> review -> ship`
-- FSM-backed artifacts for specs, plans, and tasks
-- persisted gate evidence and review records
-- deterministic review scanner blocks empty `catch`, `@ts-ignore`, focused tests, dangerous shell/Git commands, and security-sensitive changes without G7 evidence
-- built-in G7 security scanning records explainable file/line evidence, blocks CRITICAL by default, and can block HIGH findings in strict mode
-- optional strict TDD evidence gate with `--tdd-evidence` and `--tdd-strict`
+- FSM-backed artifacts with blocking on guard failures
+- Persisted gate evidence and review records
+- Deterministic review scanner blocks empty `catch`, `@ts-ignore`, focused tests, dangerous shell/Git commands, and security-sensitive changes without G7 evidence
+- OWASP Top 10 security detector extends coverage
+- Built-in G7 security scanning records explainable file/line evidence, blocks CRITICAL by default, can block HIGH in strict mode
+- Optional strict TDD evidence gate with `--tdd-evidence` and `--tdd-strict`
 - `ship --no-commit` delivery reports
-- review-gated release commits
-- 11 platform adapters and 12 professional agent profiles
-- Exa/search capability integration
-- 461 Vitest tests passing after scoped ship, TDD evidence, review scanner, and G7 security scan hardening
+- Review-gated release commits
+- 11 platform adapters, 12 professional agent profiles
+- Browser QA Capability (Playwright MCP)
+- Evolution self-improve loop
+- 499 Vitest tests passing
 
 ## Installation
 
@@ -75,6 +98,29 @@ scale verify <task-id> --tdd-strict --tdd-evidence .scale/tdd/<task-id>.json
 
 The TDD evidence JSON must include `red`, `green`, `refactor`, and `testFirst` set to `true`.
 
+## Evolution Self-Improve Loop
+
+Extract lessons from session defects and promote to rules and hooks:
+
+```bash
+# Extract Lessons from session
+scale evolution extract <session-id>
+
+# Run self-improve loop: Defect → Lesson → Rule → Hook
+scale evolution improve <session-id>
+
+# Show self-improve report
+scale evolution report <session-id>
+
+# View generated Hooks config
+scale evolution hooks <session-id> --json
+```
+
+Thresholds:
+- Lesson → Rule: requires 3 verifications
+- Rule → Active: requires 10 hits
+- Rule → Hook: requires 20 hits
+
 ## Safety Model
 
 SCALE Engine uses multiple enforcement layers:
@@ -117,14 +163,19 @@ It also includes 12 professional agent profiles:
 ```text
 src/api/cli.ts                 CLI entrypoint
 src/cli/phaseCommands.ts       DEFINE/PLAN/BUILD/VERIFY/REVIEW/SHIP
+src/cli/evolutionCommands.ts   L6 Evolution CLI commands
 src/workflow/gates/            Quality gates and persisted evidence
 src/workflow/ReviewAnalyzer.ts Deterministic review analysis
 src/workflow/ReviewStore.ts    Review record persistence
 src/workflow/EvidenceStore.ts  Gate evidence persistence
+src/workflow/evolution/        LessonExtractor + SelfImproveEngine
+src/workflow/qa/               BrowserQA + E2ETestRunner
 src/artifact/                  Artifact store and FSM definitions
 src/guardrails/                Detector and gateway logic
+src/guardrails/OWASPDetector.ts OWASP Top 10 security detection
+src/capabilities/BrowserQACapability.ts Playwright MCP wrapper
 src/evolution/                 Defect/Lesson/Rule/Hook evolution layer
-tests/                         Vitest test suites
+tests/                         Vitest test suites (499 tests)
 ```
 
 ## Development
@@ -145,6 +196,16 @@ npx vitest run tests/workflow/reviewAnalyzer.test.ts tests/workflow/reviewStore.
 
 ## Release Notes
 
+### v0.11.1
+
+- Phase Commands FSM blocking: `canTransition` + `process.exit(1)` for guard failures
+- OWASP Top 10 Detector: 19 security detection patterns
+- Browser QA Capability: Playwright MCP wrapper for E2E testing
+- L6 Evolution: `Defect → Lesson → Rule → Hook` self-improve loop
+- Evolution CLI: `scale evolution extract/improve/report/hooks`
+- ReviewAnalyzer regex fix: avoid false positives on pattern definitions
+- 499 tests passing
+
 ### v0.10.1
 
 - Hardened `ship` so release commits stage only files covered by passing review records.

package/README.md

@@ -1,168 +1,229 @@
-<p align="center">
-  <img src="https://img.shields.io/badge/version-0.10.1-orange?style=flat-square" alt="version" />
-  <img src="https://img.shields.io/badge/platforms-11-blue?style=flat-square" alt="platforms" />
-  <img src="https://img.shields.io/badge/agents-12-blue?style=flat-square" alt="agents" />
-  <img src="https://img.shields.io/badge/workflows-10-green?style=flat-square" alt="workflows" />
-  <img src="https://img.shields.io/badge/detectors-9-red?style=flat-square" alt="detectors" />
-  <img src="https://img.shields.io/badge/tests-461-passing-brightgreen?style=flat-square" alt="tests" />
-  <img src="https://img.shields.io/badge/npm-0.10.1-cb3837?style=flat-square&logo=npm" alt="npm" />
-</p>
-
-# SCALE Engine v0.10.1
-
-SCALE Engine 是一个面向 AI 编码 Agent 的工程化工作流运行时。它把提示词里的工程纪律，下沉为状态机、质量门禁、持久化证据、确定性 review 记录和发布检查。
-
-源码仓库：https://github.com/hongmaple0820/scale-engine
-国内镜像：https://gitee.com/hongmaple/scale-engine
-npm：https://www.npmjs.com/package/@hongmaple0820/scale-engine
-语言：[中文](README.md) | [English](README.en.md)
-
-## 为什么需要它
-
-提示词是建议，工程交付需要机制：
-
-- Agent 可以声称测试通过，SCALE 会保存真实验证证据。
-- Agent 可以跳过 review，SCALE 会在缺少 review 记录时阻断 `ship`。
-- Agent 可以误提交无关文件，SCALE 只暂存已通过 review 覆盖的文件。
-- Agent 可以丢失阶段状态，SCALE 会把 artifact 和 FSM 状态保存在 `.scale`。
-
-## 当前版本
-
-v0.10.1 聚焦阶段化交付链路加固：
-
-- `define -> plan -> build -> verify -> review -> ship`
-- Spec、Plan、Task artifact 接入 FSM
-- 验证门禁证据持久化
-- 代码 review 记录持久化
-- 确定性 review scanner 会阻断空 `catch`、`@ts-ignore`、focused test、危险 shell/git 命令和缺 G7 证据的安全敏感变更
-- G7 内置安全扫描会记录可解释的文件/行号证据，默认阻断 CRITICAL，严格模式可阻断 HIGH
-- 可选严格 TDD evidence 门禁：`--tdd-evidence` 和 `--tdd-strict`
-- `ship --no-commit` 交付报告
-- `ship` 发布前强制验证 review evidence
-- 11 个平台适配器，12 个专业 Agent Profile
-- Exa/search capability 集成
-- 本轮 scoped ship、TDD evidence、review scanner 与 G7 security scan 加固后，461 个 Vitest 测试通过
-
-## 安装
-
-```bash
-npm install -g @hongmaple0820/scale-engine
-scale --version
-```
-
-需要 Node.js 20 或更高版本。
-
-## 阶段工作流
-
-```bash
-scale define "Scoped release workflow" \
-  --description "Implement a TypeScript CLI workflow with verification evidence, review records, rollback constraints, and release safety checks." \
-  --success-criteria "verify evidence is persisted,review evidence is persisted,ship blocks unreviewed files"
-
-scale plan <spec-id> --rollback "Revert the release commit and remove generated artifacts"
-scale build <plan-id> --description "Implement scoped release workflow"
-scale verify <task-id>
-scale review <task-id>
-scale ship <task-id> --message "feat(workflow): add scoped release workflow"
-```
-
-如果只需要生成交付报告，不创建 Git commit：
-
-```bash
-scale ship <task-id> --no-commit
-```
-
-需要严格校验 TDD 证据时：
-
-```bash
-scale verify <task-id> --tdd-strict --tdd-evidence .scale/tdd/<task-id>.json
-```
-
-TDD evidence JSON 需要包含 `red`、`green`、`refactor`、`testFirst` 且值都为 `true`。
-
-## 安全模型
-
-| 层级 | 作用 |
-| --- | --- |
-| FSM | 阻止非法 artifact 状态流转 |
-| GateSystem | 执行 build、lint、test、coverage、security 门禁 |
-| EvidenceStore | 持久化验证证据，用于审计和发布门禁 |
-| ReviewStore | 持久化确定性 review 记录 |
-| ReviewAnalyzer | 扫描 diff 中的高风险代码、流程债和缺失安全证据 |
-| Detectors | 检测暴力重试、过早完成、甩锅、忙碌假象等失败模式 |
-| Ship gate | 发布前必须验证通过，并且 review evidence 必须存在且通过 |
-
-`ship` 不再执行 `git add .`。它只会暂存已通过 review 记录覆盖的文件；如果 review 后出现新的可 review 变更，`ship` 会阻断并要求重新 review。
-
-G7 `SecurityGate` 内置轻量安全扫描，覆盖硬编码密钥、私钥、TLS 校验关闭、`eval`/`Function`、原始 HTML 注入、危险 shell 命令、shell 执行和空 `catch` 等模式。兼容模式只阻断 CRITICAL；严格模式会同时阻断 HIGH。
-
-## 支持的平台与角色
-
-SCALE Engine 内置 11 个平台适配器，包括 Claude Code、Codex CLI、OpenCode、Cursor、Gemini CLI、OpenClaw、Hermes、Trae、WorkBuddy、VS Code Copilot CLI、QCoder。
-
-内置 12 个专业 Agent Profile：
-
-- frontend
-- backend
-- testing
-- UI design
-- operations
-- product
-- code review
-- security
-- database
-- performance
-- documentation
-- architecture
-
-## 项目结构
-
-```text
-src/api/cli.ts                 CLI 入口
-src/cli/phaseCommands.ts       DEFINE/PLAN/BUILD/VERIFY/REVIEW/SHIP
-src/workflow/gates/            质量门禁与验证证据
-src/workflow/ReviewAnalyzer.ts 确定性 review 分析
-src/workflow/ReviewStore.ts    review 记录持久化
-src/workflow/EvidenceStore.ts  gate evidence 持久化
-src/artifact/                  artifact 存储与 FSM 定义
-src/guardrails/                detector 与 gateway
-src/evolution/                 Defect/Lesson/Rule/Hook 自进化层
-tests/                         Vitest 测试
-```
-
-## 开发与验证
-
-```bash
-npm install
-npm run build
-npx vitest run
-npm pack --dry-run
-```
-
-工作流相关定向测试：
-
-```bash
-npx vitest run tests/workflow/phaseCli.test.ts
-npx vitest run tests/workflow/reviewAnalyzer.test.ts tests/workflow/reviewStore.test.ts tests/workflow/gateSystem.test.ts
-```
-
-## v0.10.1 更新
-
-- 新增 `ship --no-commit`。
-- 强化 `ship`：发布提交只暂存已 review 文件。
-- 新增可选严格 TDD evidence 校验。
-- 增强命令证据元数据：工作目录、时间戳、stdout/stderr 尾部和输出 hash。
-- 强化 deterministic review scanner：阻断空 `catch`、`@ts-ignore`、focused test、危险 shell/git 命令与缺 G7 证据的安全敏感变更。
-- 强化 G7 内置安全扫描：输出文件/行号证据，默认阻断 CRITICAL，严格模式阻断 HIGH。
-- 新增 `review -> ship`、未 review 文件阻断和安全扫描误报边界的 CLI/单元回归测试。
-- 发版前已验证 `npm run build`、完整 Vitest 测试和 `npm pack --dry-run`。
-
-## v0.10.0 更新
-
-- 新增阶段化 CLI 工作流，并接入 FSM。
-- 新增验证证据和 review 记录持久化。
-- 发布 `@hongmaple0820/scale-engine@0.10.0`。
-
-## License
-
-MIT
+<p align="center">
+  <img src="https://img.shields.io/badge/version-0.12.0-orange?style=flat-square" alt="version" />
+  <img src="https://img.shields.io/badge/platforms-13-blue?style=flat-square" alt="platforms" />
+  <img src="https://img.shields.io/badge/agents-12-blue?style=flat-square" alt="agents" />
+  <img src="https://img.shields.io/badge/workflows-10-green?style=flat-square" alt="workflows" />
+  <img src="https://img.shields.io/badge/detectors-19-red?style=flat-square" alt="detectors" />
+  <img src="https://img.shields.io/badge/tests-562-passing-brightgreen?style=flat-square" alt="tests" />
+  <img src="https://img.shields.io/badge/npm-0.12.0-cb3837?style=flat-square&logo=npm" alt="npm" />
+</p>
+
+# SCALE Engine v0.12.0
+
+SCALE Engine 是一个面向 AI 编码 Agent 的工程化工作流运行时。它把提示词里的工程纪律，下沉为状态机、质量门禁、持久化证据、确定性 review 记录和发布检查。
+
+源码仓库：https://github.com/hongmaple0820/scale-engine
+国内镜像：https://gitee.com/hongmaple/scale-engine
+npm：https://www.npmjs.com/package/@hongmaple0820/scale-engine
+语言：[中文](README.md) | [English](README.en.md)
+
+## 为什么需要它
+
+提示词是建议，工程交付需要机制：
+
+- Agent 可以声称测试通过，SCALE 会保存真实验证证据。
+- Agent 可以跳过 review，SCALE 会在缺少 review 记录时阻断 `ship`。
+- Agent 可以误提交无关文件，SCALE 只暂存已通过 review 覆盖的文件。
+- Agent 可以丢失阶段状态，SCALE 会把 artifact 和 FSM 状态保存在 `.scale`。
+
+## 当前版本
+
+v0.11.1 新增四大优先级改进：
+
+### Phase Commands FSM 阻断
+- `canTransition` + `process.exit(1)` 确保 FSM guard 失败时阻塞流程，而非继续执行
+- define/plan/build/verify 各阶段添加明确的阻断提示
+
+### OWASP Top 10 检测器
+- 新增 `OWASPDetector` 覆盖 SQL 注入、XSS、路径遍历、SSRF、Auth Bypass、弱加密、CORS 错误配置、CSRF、文件上传、敏感数据泄露
+- 19 类安全检测模式，自动识别 regex 定义避免误报
+
+### Browser QA Capability
+- `BrowserQACapability` 封装 Playwright MCP 工具
+- 支持导航、点击、截图、console 检查、E2E 测试流程
+
+### L6 Evolution 自改进闭环
+- `LessonExtractor` 从会话 Defect 事件提取可复用教训
+- `SelfImproveEngine` 实现 `Defect → Lesson → Rule → Hook` 晋升流水线
+- 新增 CLI 命令：`scale evolution extract/improve/report/hooks`
+
+---
+
+**完整阶段化交付链路**：
+
+- `define -> plan -> build -> verify -> review -> ship`
+- Spec、Plan、Task artifact 接入 FSM，guard 失败时阻断而非继续
+- 验证门禁证据持久化
+- 代码 review 记录持久化
+- 确定性 review scanner 会阻断空 `catch`、`@ts-ignore`、focused test、危险 shell/git 命令和缺 G7 证据的安全敏感变更
+- OWASP Top 10 安全检测器扩展安全覆盖
+- G7 内置安全扫描会记录可解释的文件/行号证据，默认阻断 CRITICAL，严格模式可阻断 HIGH
+- 可选严格 TDD evidence 门禁：`--tdd-evidence` 和 `--tdd-strict`
+- `ship --no-commit` 交付报告
+- `ship` 发布前强制验证 review evidence
+- 11 个平台适配器，12 个专业 Agent Profile
+- Browser QA Capability (Playwright MCP)
+- Evolution 自改进闭环
+- 本轮加固后，499 个 Vitest 测试通过
+
+## 安装
+
+```bash
+npm install -g @hongmaple0820/scale-engine
+scale --version
+```
+
+需要 Node.js 20 或更高版本。
+
+## 阶段工作流
+
+```bash
+scale define "Scoped release workflow" \
+  --description "Implement a TypeScript CLI workflow with verification evidence, review records, rollback constraints, and release safety checks." \
+  --success-criteria "verify evidence is persisted,review evidence is persisted,ship blocks unreviewed files"
+
+scale plan <spec-id> --rollback "Revert the release commit and remove generated artifacts"
+scale build <plan-id> --description "Implement scoped release workflow"
+scale verify <task-id>
+scale review <task-id>
+scale ship <task-id> --message "feat(workflow): add scoped release workflow"
+```
+
+如果只需要生成交付报告，不创建 Git commit：
+
+```bash
+scale ship <task-id> --no-commit
+```
+
+需要严格校验 TDD 证据时：
+
+```bash
+scale verify <task-id> --tdd-strict --tdd-evidence .scale/tdd/<task-id>.json
+```
+
+TDD evidence JSON 需要包含 `red`、`green`、`refactor`、`testFirst` 且值都为 `true`。
+
+## Evolution 自改进闭环
+
+从会话缺陷中提取教训，晋升为规则和 Hook：
+
+```bash
+# 从会话提取 Lessons
+scale evolution extract <session-id>
+
+# 运行自改进闭环：Defect → Lesson → Rule → Hook
+scale evolution improve <session-id>
+
+# 显示自改进报告
+scale evolution report <session-id>
+
+# 查看生成的 Hooks 配置
+scale evolution hooks <session-id> --json
+```
+
+阈值配置：
+- Lesson → Rule：需验证 3 次
+- Rule → Active：需触发 10 次
+- Rule → Hook：需触发 20 次
+
+## 安全模型
+
+| 层级 | 作用 |
+| --- | --- |
+| FSM | 阻止非法 artifact 状态流转 |
+| GateSystem | 执行 build、lint、test、coverage、security 门禁 |
+| EvidenceStore | 持久化验证证据，用于审计和发布门禁 |
+| ReviewStore | 持久化确定性 review 记录 |
+| ReviewAnalyzer | 扫描 diff 中的高风险代码、流程债和缺失安全证据 |
+| Detectors | 检测暴力重试、过早完成、甩锅、忙碌假象等失败模式 |
+| Ship gate | 发布前必须验证通过，并且 review evidence 必须存在且通过 |
+
+`ship` 不再执行 `git add .`。它只会暂存已通过 review 记录覆盖的文件；如果 review 后出现新的可 review 变更，`ship` 会阻断并要求重新 review。
+
+G7 `SecurityGate` 内置轻量安全扫描，覆盖硬编码密钥、私钥、TLS 校验关闭、`eval`/`Function`、原始 HTML 注入、危险 shell 命令、shell 执行和空 `catch` 等模式。兼容模式只阻断 CRITICAL；严格模式会同时阻断 HIGH。
+
+## 支持的平台与角色
+
+SCALE Engine 内置 11 个平台适配器，包括 Claude Code、Codex CLI、OpenCode、Cursor、Gemini CLI、OpenClaw、Hermes、Trae、WorkBuddy、VS Code Copilot CLI、QCoder。
+
+内置 12 个专业 Agent Profile：
+
+- frontend
+- backend
+- testing
+- UI design
+- operations
+- product
+- code review
+- security
+- database
+- performance
+- documentation
+- architecture
+
+## 项目结构
+
+```text
+src/api/cli.ts                 CLI 入口
+src/cli/phaseCommands.ts       DEFINE/PLAN/BUILD/VERIFY/REVIEW/SHIP
+src/cli/evolutionCommands.ts   L6 Evolution CLI 命令
+src/workflow/gates/            质量门禁与验证证据
+src/workflow/ReviewAnalyzer.ts 确定性 review 分析
+src/workflow/ReviewStore.ts    review 记录持久化
+src/workflow/EvidenceStore.ts  gate evidence 持久化
+src/workflow/evolution/        LessonExtractor + SelfImproveEngine
+src/workflow/qa/               BrowserQA + E2ETestRunner
+src/artifact/                  artifact 存储与 FSM 定义
+src/guardrails/                detector 与 gateway
+src/guardrails/OWASPDetector.ts OWASP Top 10 安全检测
+src/capabilities/BrowserQACapability.ts Playwright MCP 包装器
+src/evolution/                 Defect/Lesson/Rule/Hook 自进化层
+tests/                         Vitest 测试 (499 tests)
+```
+
+## 开发与验证
+
+```bash
+npm install
+npm run build
+npx vitest run
+npm pack --dry-run
+```
+
+工作流相关定向测试：
+
+```bash
+npx vitest run tests/workflow/phaseCli.test.ts
+npx vitest run tests/workflow/reviewAnalyzer.test.ts tests/workflow/reviewStore.test.ts tests/workflow/gateSystem.test.ts
+```
+
+## v0.11.1 更新
+
+- Phase Commands FSM 阻断：`canTransition` + `process.exit(1)` 确保 guard 失败时阻塞
+- OWASP Top 10 检测器：19 类安全检测模式
+- Browser QA Capability：Playwright MCP 包装器用于 E2E 测试
+- L6 Evolution：`Defect → Lesson → Rule → Hook` 自改进闭环
+- Evolution CLI：`scale evolution extract/improve/report/hooks`
+- ReviewAnalyzer regex 修复：避免模式定义误报
+- 499 测试通过
+
+## v0.10.1 更新
+
+- 新增 `ship --no-commit`。
+- 强化 `ship`：发布提交只暂存已 review 文件。
+- 新增可选严格 TDD evidence 校验。
+- 增强命令证据元数据：工作目录、时间戳、stdout/stderr 尾部和输出 hash。
+- 强化 deterministic review scanner：阻断空 `catch`、`@ts-ignore`、focused test、危险 shell/git 命令与缺 G7 证据的安全敏感变更。
+- 强化 G7 内置安全扫描：输出文件/行号证据，默认阻断 CRITICAL，严格模式阻断 HIGH。
+- 新增 `review -> ship`、未 review 文件阻断和安全扫描误报边界的 CLI/单元回归测试。
+- 发版前已验证 `npm run build`、完整 Vitest 测试和 `npm pack --dry-run`。
+
+## v0.10.0 更新
+
+- 新增阶段化 CLI 工作流，并接入 FSM。
+- 新增验证证据和 review 记录持久化。
+- 发布 `@hongmaple0820/scale-engine@0.10.0`。
+
+## License
+
+MIT

package/dist/adapters/DeepSeekTuiAdapter.d.ts

@@ -0,0 +1,19 @@
+import type { IAgentAdapter, AdapterConfig, InitResult, SettingsJson } from './ClaudeCodeAdapter.js';
+export declare class DeepSeekTuiAdapter implements IAgentAdapter {
+    readonly agentType = "deepseek-tui";
+    private projectDir;
+    private scaleDir;
+    getSettingsPath(): string;
+    getKnowledgeDocPath(): string;
+    getSkillsDir(): string;
+    isInstalled(): boolean;
+    generateSettings(): SettingsJson;
+    /**
+     * 生成 .deepseek/config.toml 项目覆盖配置
+     * 仅包含 #485 项目 overlay 支持的顶层字段
+     */
+    generateConfigToml(scenarioMode?: 'sandbox' | 'standard' | 'critical'): string;
+    mergeSettings(existing: SettingsJson): SettingsJson;
+    generateKnowledgeDoc(projectName: string, techStack?: string[]): string;
+    init(config: AdapterConfig): Promise<InitResult>;
+}