@jennie-shawn/starwork 0.1.0-alpha.16 → 0.1.0-alpha.18

package/README.md

@@ -5,11 +5,11 @@ StarWork 是一套面向 AI 的工作台协议和工具集，用来帮助用户
 它包含四个核心部分：
 
 - **Core**：定义工作台应该长什么样，包括目录结构、角色边界、状态文件和健康检查规则。
-- **CLI**：提供 `init`、`doctor`、`spawn`、`upgrade`、`adapt`、`pack install`、`multiagent` 等命令，用来创建、检查、升级、协作和扩展工作台。
+- **CLI**：提供 `init`、`doctor`、`knowledge`、`spawn`、`upgrade`、`adapt`、`pack install`、`multiagent` 等命令，用来创建、检查、升级、协作和扩展工作台。
 - **Packs**：场景模板包。当前 A 测阶段默认使用通用工作能力。
 - **Skills**：给 Codex 等 AI 工具使用的工作流说明，让 AI 能更可靠地帮用户设计和生成 StarWork 工作台。
 
-当前版本处于 A 测阶段，适合测试安装流程、基础命令、工作台结构和 AI skill 使用体验。
+当前版本处于 A 测阶段。npm `latest` 为 `@jennie-shawn/starwork@0.1.0-alpha.17`，适合测试安装流程、基础命令、工作台结构、AI Skill 使用体验和 Codex 多会话协作链路。
 
 ## 安装 CLI
 
@@ -33,7 +33,7 @@ npx @jennie-shawn/starwork --help
 StarWork skills 分两类管理：
 
 - 系统 Skill 通过 GitHub 仓库和 `skills` CLI 安装到 AI 工具的全局环境。
-- Kit 自带 Skill 跟着工作台走，例如 Hub Kit 自带 `starworkSpawn`，单项目 Kit 自带 `neat-freak`。
+- 工作台自带 Skill 跟着工作台走，例如项目中心自带 `starworkSpawn`，项目工作台自带 `neat-freak`。
 
 给 Codex 安装全部 StarWork 系统级 Skills：
 
@@ -45,15 +45,18 @@ npx skills add jennie-shawn/StarWork -g -a codex -y
 
 当前系统级 Skills：
 
-- `starworkInit`：帮助 AI 判断是创建 Project 还是 Hub，选择语言；用户确认创建时，会带着 `starwork init` dry-run、执行并用 `doctor` 验证。
-- `starworkDoctor`：帮助 AI 基于 `starwork doctor --json` 解释目录问题、识别旧模板或 Hub-like 旧主库；用户明确要求升级时，生成可预览、可确认的升级方案。
-- `starworkMultiagent`：帮助 AI 把“登记当前会话为常用智能体”“管理多 AI 分工”“登记共享输出”等请求转换成 `starwork multiagent` 命令组合。
+- `starworkInit`：帮助 AI 判断是创建项目工作台还是项目中心，选择语言；用户确认创建时，会带着 `starwork init` dry-run、执行并用 `doctor` 验证。
+- `starworkDoctor`：帮助 AI 基于 `starwork doctor --json` 解释目录问题、识别旧模板或类似项目中心的旧工作区；用户明确要求升级时，生成可预览、可确认的升级方案。
+- `starworkMultiagent`：帮助 AI 把“登记当前会话为常用智能体”“管理多 AI 分工”“跨会话指令”“读取其他 lane 状态”等请求转换成 `starwork multiagent` 命令组合。
+- `starworkKnowledge`：帮助 AI 开启和维护项目本地知识库；创建结构时先调用 `starwork knowledge`，整理内容时区分主题页和综合判断。
+
+说明：`starworkKnowledge` 是全局入口 Skill。项目开启知识库后，CLI 会把 `starworkKnowledgeProject` 安装到当前项目的 `.agents/skills/` 和 `.claude/skills/`，它不属于全局安装清单。
 
 Kit 随附 Skills 不走全局安装命令，它们跟着具体工作台走：
 
-- `starworkSpawn`：Hub Kit 自带 Skill，帮助已有 Hub 设计 `starwork spawn --blueprint` 工作台定制单。
-- `starworkAudit`：Hub Kit 自带 Skill，帮助 Hub 解读 `starwork audit --json`，判断旗下 Project Satellite 的健康状况，并生成保守的 `starwork repair --blueprint` 修复蓝图。
-- `neat-freak`：单项目 Kit 自带 Skill，帮助项目收尾、整理和归档。
+- `starworkSpawn`：项目中心自带 Skill，帮助已有项目中心设计 `starwork spawn --blueprint` 工作台定制单。
+- `starworkAudit`：项目中心自带 Skill，帮助项目中心解读 `starwork audit --json`，判断它管理的项目工作台健康状况，并生成保守的 `starwork repair --blueprint` 修复蓝图。
+- `neat-freak`：项目工作台自带 Skill，帮助项目收尾、整理和归档。
 
 ## 让 Agent 帮你安装
 
@@ -72,7 +75,7 @@ https://raw.githubusercontent.com/jennie-shawn/StarWork/main/docs/agent-install-
 
 ## 快速开始
 
-创建一个 Project 工作台：
+创建一个项目工作台：
 
 ```bash
 starwork init \
@@ -90,18 +93,26 @@ starwork init \
 starwork doctor --target ~/Desktop/starwork-a-test
 ```
 
-创建一个多项目中枢：
+可选：给项目开启知识库：
+
+```bash
+starwork knowledge init --target ~/Desktop/starwork-a-test --dry-run
+starwork knowledge init --target ~/Desktop/starwork-a-test --yes
+starwork knowledge status --target ~/Desktop/starwork-a-test --json
+```
+
+创建一个项目中心：
 
 ```bash
 starwork init \
   --type hub \
   --language zh \
-  --name "StarWork Hub A Test" \
+  --name "StarWork Project Center A Test" \
   --target ~/Desktop/starwork-hub-a-test \
   --yes
 ```
 
-从 Hub 生成一个项目：
+从项目中心创建一个项目工作台：
 
 ```bash
 starwork spawn \
@@ -118,7 +129,7 @@ starwork spawn \
 starwork doctor --target ~/Desktop/starwork-alpha-project
 ```
 
-从 Hub 巡检项目：
+从项目中心巡检项目：
 
 ```bash
 starwork audit --hub ~/Desktop/starwork-hub-a-test
@@ -129,6 +140,7 @@ starwork audit --hub ~/Desktop/starwork-hub-a-test
 ```text
 starwork init
 starwork doctor
+starwork knowledge
 starwork spawn
 starwork upgrade
 starwork adapt
@@ -136,15 +148,32 @@ starwork pack install
 starwork multiagent
 ```
 
+## 产品规划资料
+
+StarWork 的功能规划、版本 SPEC、讨论沉淀、参考资料和验收材料统一放在：
+
+```text
+product/planning/
+```
+
+其中独立功能按功能点归档：
+
+```text
+product/planning/features/<feature>/
+```
+
+正式用户文档和 HTML 阅读稿仍放在 `product/docs/`。
+
 当前能力：
 
-- `init`：创建普通项目工作台或多项目 Hub；交互默认推荐普通项目工作台。
-- `doctor`：检查工作台是否完整、关键文件是否还在；也会为旧模板和 Hub-like 旧主库列出目录线索，交给 `starworkDoctor` 判断。
-- `spawn`：从已有 Hub 创建并登记卫星项目，支持中文或英文卫星项目目录镜像。
-- `upgrade`：按 `starworkDoctor` skill 生成的升级蓝图，把历史模板或 Hub-like 旧主库安全接入 StarWork 工作台。
+- `init`：创建项目工作台或项目中心；交互默认推荐项目工作台。
+- `doctor`：检查工作台是否完整、关键文件是否还在；也会为旧模板和类似项目中心的旧工作区列出目录线索，交给 `starworkDoctor` 判断。
+- `knowledge`：开启、检查和按确认方案整理项目本地知识库；成功开启后安装项目内知识库助手，默认不会移动旧 `知识/` 或 `knowledge/`。
+- `spawn`：从已有项目中心创建并登记项目工作台，支持中文或英文目录镜像。
+- `upgrade`：按 `starworkDoctor` skill 生成的升级蓝图，把历史模板或类似项目中心的旧工作区安全接入 StarWork 工作台。
 - `adapt`：生成 Claude Code、Cursor 等 AI 工具的适配文件。
 - `pack install`：向兼容工作台安装支持的场景能力。
-- `multiagent`：为同一项目建立自定义 AI 职责位、绑定会话，并登记跨 lane 共享输出。
+- `multiagent`：为同一项目建立自定义 AI 职责位、绑定会话、登记跨 lane 共享输出，并支持 Codex 多会话读取、launch 和格式化指令发送；未观察到目标 turn 完成时会返回 `started_unverified`，避免把未完成交付误说成已完成。
 
 ## 仓库结构
 
@@ -158,7 +187,8 @@ starwork multiagent
 ├── schemas/    # 结构化 schema
 ├── adapters/   # AI 工具适配规则
 ├── examples/   # 示例
-└── docs/       # 产品文档和 A 测指南
+├── docs/       # 产品文档和 A 测指南
+└── planning/   # 功能规划、版本 SPEC、讨论、参考和验收
 ```
 
 ## A 测反馈重点
@@ -168,9 +198,9 @@ starwork multiagent
 - CLI 是否能顺利安装和运行。
 - `init` 创建的工作台结构是否容易理解。
 - `doctor` 的检查结果是否能指导修复问题。
-- Hub + Satellite 工作流是否自然。
+- 项目中心创建和管理项目工作台的流程是否自然。
 - 系统 Skill 是否能被 AI 正确识别和调用。
-- Hub Kit 自带的 `starworkSpawn`、单项目 Kit 自带的 `neat-freak` 是否能在对应工作台内被正确发现。
+- 项目中心自带的 `starworkSpawn`、项目工作台自带的 `neat-freak` 是否能在对应工作台内被正确发现。
 
 更完整的测试脚本见 [A 测安装指南](docs/alpha-test-guide.md)。
 

package/adapters/README.md

@@ -3,3 +3,37 @@
 这里存放不同 Agent 的适配层，如 Codex、Claude Code、Cursor、Trae 等。
 
 适配层不混入 StarWork Core。
+
+## 当前定位
+
+Adapter 是 StarWork 的全局宿主适配层，不只服务 MultiAgent。
+
+它负责把 StarWork 工作台协议翻译成不同 AI 宿主能正确读取、正确行动、正确保存上下文的方式，包括：
+
+- 规则入口
+- Skill 识别
+- 会话能力
+- transcript / memory 边界
+- 宿主 CLI / API 探测
+- 私有状态安全边界
+
+## v0.1 当前进度
+
+Host Adapter v0.1 已先落地底座：
+
+- `contract.md`：定义 adapter profile、工作台 adapter state 和宿主安全边界。
+- `codex/`、`claude-code/`、`cursor/`、`trae/`：分别维护 `profile.json`、`rules.md`、`safety.md`。
+- `starwork adapt <host|all> --capabilities`：只读输出宿主能力，不改工作台。
+- `starwork adapt <host|all>`：写入宿主规则入口、Skill 目录和 `.starwork/adapters.json`。
+- `starwork doctor --host <host|all>`：检查宿主入口、Skill 目录、Cursor / Trae frontmatter、Trae disabled config 和不安全能力声明。
+- `starwork init --adapter <host|all>`：初始化完成后继续生成对应宿主入口。
+- `starwork multiagent`：Claude Code 可用 `CLAUDE_CODE_SESSION_ID` 绑定和 `claude --resume` 继续命令；Cursor / Trae 默认走 manual handoff，不伪装后台送达。
+
+Cursor / Trae 的深度 probe、Claude Code 真实本机 transcript 路径适配和更完整用户文档仍需继续 A 测打磨。
+
+产品规划事实源见：
+
+```text
+product/planning/features/host-adapters/specs/v0.1.md
+product/planning/features/host-adapters/specs/v0.1-implementation.md
+```

package/adapters/claude-code/profile.json

@@ -0,0 +1,55 @@
+{
+  "schema": "starwork.adapter.profile.v0.1",
+  "host": "claude-code",
+  "label": "Claude Code",
+  "version": "0.1",
+  "rules": {
+    "entry_file": "CLAUDE.md",
+    "can_generate_entry": "supported",
+    "can_update_entry": "partial",
+    "supports_project_scope": "supported",
+    "supports_nested_rules": "unknown"
+  },
+  "skills": {
+    "system_install": "manual",
+    "project_mount": "manual",
+    "primary_project_dir": ".claude/skills/",
+    "project_mount_dirs": [".claude/skills/", ".agents/skills/"],
+    "host_native_dirs": [".claude/skills/"],
+    "global_dirs": ["~/.claude/skills/"],
+    "conflict_priority": [".claude/skills/", ".agents/skills/"],
+    "discovery": "manual",
+    "trigger_style": "rules-guided",
+    "adapter_rule_required": true
+  },
+  "sessions": {
+    "detect_current": "supported",
+    "rename": "partial",
+    "list": "partial",
+    "read": "supported",
+    "continue": "supported",
+    "send_message": "manual",
+    "create": "partial",
+    "pin": "unsupported",
+    "archive": "unsupported"
+  },
+  "memory": {
+    "transcript_read": "supported",
+    "transcript_write": "unsupported",
+    "summary_visible": "unsupported",
+    "project_memory": "supported",
+    "export": "partial"
+  },
+  "commands": {
+    "cli_available": "partial",
+    "api_available": "unsupported",
+    "requires_auth": "partial",
+    "can_probe": "supported"
+  },
+  "safety": {
+    "private_paths_readonly": ["~/.claude/projects/**", "~/.claude/sessions/**"],
+    "never_write_paths": ["~/.claude/projects/**/*.jsonl", "~/.claude.json"],
+    "destructive_commands": ["claude project purge"],
+    "degradation_required": ["sessions.send_message", "sessions.pin", "sessions.archive"]
+  }
+}

package/adapters/claude-code/rules.md

@@ -0,0 +1,13 @@
+# Claude Code Adapter Rules
+
+Claude Code uses `CLAUDE.md` as its project rule entry.
+
+The entry should point Claude Code back to:
+
+- `AGENTS.md`
+- `.starwork/workspace.json`
+- `.starwork/skills.json`
+- `.claude/skills/`
+- `.agents/skills/`
+
+Claude Code can bind or resume sessions, and transcripts may be read as candidate evidence. It cannot safely receive background messages to another session in StarWork v0.1.

package/adapters/claude-code/safety.md

@@ -0,0 +1,6 @@
+# Claude Code Adapter Safety
+
+- Never write Claude Code transcript JSONL files.
+- Never write `~/.claude.json`.
+- Never run destructive commands such as `claude project purge`.
+- Treat transcript content as candidate evidence only, not as StarWork formal memory.

package/adapters/codex/profile.json

@@ -0,0 +1,55 @@
+{
+  "schema": "starwork.adapter.profile.v0.1",
+  "host": "codex",
+  "label": "Codex",
+  "version": "0.1",
+  "rules": {
+    "entry_file": "AGENTS.md",
+    "can_generate_entry": "supported",
+    "can_update_entry": "partial",
+    "supports_project_scope": "supported",
+    "supports_nested_rules": "partial"
+  },
+  "skills": {
+    "system_install": "supported",
+    "project_mount": "supported",
+    "primary_project_dir": ".agents/skills/",
+    "project_mount_dirs": [".agents/skills/"],
+    "host_native_dirs": [".agents/skills/"],
+    "global_dirs": ["~/.agents/skills/"],
+    "conflict_priority": [".agents/skills/"],
+    "discovery": "partial",
+    "trigger_style": "rules-guided",
+    "adapter_rule_required": true
+  },
+  "sessions": {
+    "detect_current": "partial",
+    "rename": "supported",
+    "list": "supported",
+    "read": "supported",
+    "continue": "supported",
+    "send_message": "supported",
+    "create": "supported",
+    "pin": "partial",
+    "archive": "unsupported"
+  },
+  "memory": {
+    "transcript_read": "partial",
+    "transcript_write": "unsupported",
+    "summary_visible": "partial",
+    "project_memory": "supported",
+    "export": "partial"
+  },
+  "commands": {
+    "cli_available": "supported",
+    "api_available": "partial",
+    "requires_auth": "partial",
+    "can_probe": "supported"
+  },
+  "safety": {
+    "private_paths_readonly": [],
+    "never_write_paths": [],
+    "destructive_commands": [],
+    "degradation_required": ["sessions.archive"]
+  }
+}

package/adapters/codex/rules.md

@@ -0,0 +1,12 @@
+# Codex Adapter Rules
+
+Codex uses `AGENTS.md` as the primary project rule entry.
+
+The adapter should ensure Codex can find:
+
+- `AGENTS.md`
+- `.starwork/workspace.json`
+- `.starwork/skills.json`
+- `.agents/skills/`
+
+Codex can support automatic MultiAgent `launch`, `read`, and `instruct`, but completion must still be verified through StarWork state and target thread reads.

package/adapters/codex/safety.md

@@ -0,0 +1,5 @@
+# Codex Adapter Safety
+
+- Do not write Codex private runtime state.
+- Do not treat thread history as StarWork formal project memory.
+- If a turn is started but completion is not observed, report `started_unverified`, not completed delivery.

package/adapters/contract.md

@@ -0,0 +1,64 @@
+# StarWork Host Adapter Contract
+
+Host Adapter is the StarWork layer that translates the workspace protocol for a concrete AI host.
+
+It does not redefine Core, Kit, or Pack structure. It records what each host can read, generate, mount, inspect, continue, or only handle manually.
+
+## Profile Schema
+
+Each built-in host has:
+
+```text
+product/adapters/<host>/profile.json
+product/adapters/<host>/rules.md
+product/adapters/<host>/safety.md
+```
+
+`profile.json` must use:
+
+```json
+{
+  "schema": "starwork.adapter.profile.v0.1",
+  "host": "cursor",
+  "label": "Cursor",
+  "version": "0.1",
+  "rules": {},
+  "skills": {},
+  "sessions": {},
+  "memory": {},
+  "commands": {},
+  "safety": {}
+}
+```
+
+## Capability Levels
+
+Only these values are valid:
+
+```text
+supported | partial | manual | unsupported | unknown
+```
+
+- `supported`: stable automatic behavior.
+- `partial`: available but incomplete, environment-dependent, or not stable enough to promise.
+- `manual`: StarWork can generate instructions or handoff text, but the user must act.
+- `unsupported`: not supported or not safe to rely on.
+- `unknown`: not researched enough to make a promise.
+
+`partial`, `manual`, and `unknown` must never be described as automatic completion.
+
+## State
+
+Workspace adapter state lives in:
+
+```text
+.starwork/adapters.json
+```
+
+`.starwork/workspace.json.adapters` remains a short compatibility summary only.
+
+## Safety Rules
+
+Adapters must not write host private transcripts, private databases, encrypted stores, or global configuration unless the user explicitly asks for a documented host-level action.
+
+Host private history is never a StarWork source of truth.

package/adapters/cursor/profile.json

@@ -0,0 +1,60 @@
+{
+  "schema": "starwork.adapter.profile.v0.1",
+  "host": "cursor",
+  "label": "Cursor",
+  "version": "0.1",
+  "rules": {
+    "entry_file": ".cursor/rules/starwork.mdc",
+    "can_generate_entry": "supported",
+    "can_update_entry": "partial",
+    "supports_project_scope": "supported",
+    "supports_nested_rules": "supported"
+  },
+  "skills": {
+    "system_install": "manual",
+    "project_mount": "supported",
+    "primary_project_dir": ".cursor/skills/",
+    "project_mount_dirs": [".cursor/skills/", ".agents/skills/"],
+    "host_native_dirs": [".cursor/skills/", ".agents/skills/", ".claude/skills/", ".codex/skills/"],
+    "global_dirs": ["~/.cursor/skills/", "~/.agents/skills/", "~/.claude/skills/", "~/.codex/skills/"],
+    "conflict_priority": [".cursor/skills/", ".agents/skills/", ".claude/skills/", ".codex/skills/"],
+    "discovery": "supported",
+    "trigger_style": "native-skill-discovery",
+    "adapter_rule_required": true,
+    "frontmatter": {
+      "required": ["name", "description"],
+      "name_pattern": "^[a-z0-9-]+$",
+      "name_matches_parent": true
+    }
+  },
+  "sessions": {
+    "detect_current": "partial",
+    "rename": "partial",
+    "list": "partial",
+    "read": "partial",
+    "continue": "partial",
+    "send_message": "partial",
+    "create": "partial",
+    "pin": "unsupported",
+    "archive": "unsupported"
+  },
+  "memory": {
+    "transcript_read": "partial",
+    "transcript_write": "unsupported",
+    "summary_visible": "unknown",
+    "project_memory": "supported",
+    "export": "partial"
+  },
+  "commands": {
+    "cli_available": "partial",
+    "api_available": "unsupported",
+    "requires_auth": "partial",
+    "can_probe": "partial"
+  },
+  "safety": {
+    "private_paths_readonly": ["~/Library/Application Support/Cursor/**"],
+    "never_write_paths": ["~/Library/Application Support/Cursor/**/*.vscdb", "~/Library/Application Support/Cursor/**/*.sqlite"],
+    "destructive_commands": [],
+    "degradation_required": ["sessions.pin", "sessions.archive"]
+  }
+}

package/adapters/cursor/rules.md

@@ -0,0 +1,7 @@
+# Cursor Adapter Rules
+
+Cursor uses `.cursor/rules/starwork.mdc` as the StarWork entry.
+
+Cursor can discover project Skills from `.cursor/skills/` and `.agents/skills/`, and can also see compatible `.claude/skills/` and `.codex/skills/` directories. StarWork still treats `.starwork/skills.json` as the project Skill fact source.
+
+Cursor session automation is partial in v0.1. If automatic delivery is not verified, StarWork must generate a handoff message instead of claiming delivery.

package/adapters/cursor/safety.md

@@ -0,0 +1,5 @@
+# Cursor Adapter Safety
+
+- Do not write Cursor private databases or chat history.
+- Do not treat Cursor transcript or UI state as StarWork formal memory.
+- Do not claim background delivery unless a supported, verified route exists.

package/adapters/trae/profile.json

@@ -0,0 +1,61 @@
+{
+  "schema": "starwork.adapter.profile.v0.1",
+  "host": "trae",
+  "label": "Trae",
+  "version": "0.1",
+  "rules": {
+    "entry_file": ".trae/rules/starwork.md",
+    "can_generate_entry": "supported",
+    "can_update_entry": "partial",
+    "supports_project_scope": "supported",
+    "supports_nested_rules": "unknown"
+  },
+  "skills": {
+    "system_install": "manual",
+    "project_mount": "supported",
+    "primary_project_dir": ".trae/skills/",
+    "project_mount_dirs": [".trae/skills/", ".agents/skills/"],
+    "host_native_dirs": [".trae/skills/", ".agents/skills/"],
+    "global_dirs": ["~/.trae-cn/skills", "%userprofile%/.trae-cn/skills"],
+    "conflict_priority": [".trae/skills/", ".agents/skills/"],
+    "discovery": "supported",
+    "trigger_style": "native-skill-discovery",
+    "adapter_rule_required": true,
+    "frontmatter": {
+      "required": ["name", "description"],
+      "name_pattern": "^[a-z0-9-]+$",
+      "name_matches_parent": true
+    },
+    "disabled_config": ".trae/skill-config.json"
+  },
+  "sessions": {
+    "detect_current": "partial",
+    "rename": "unsupported",
+    "list": "partial",
+    "read": "unsupported",
+    "continue": "manual",
+    "send_message": "manual",
+    "create": "manual",
+    "pin": "unsupported",
+    "archive": "unsupported"
+  },
+  "memory": {
+    "transcript_read": "unsupported",
+    "transcript_write": "unsupported",
+    "summary_visible": "unsupported",
+    "project_memory": "supported",
+    "export": "unsupported"
+  },
+  "commands": {
+    "cli_available": "partial",
+    "api_available": "unsupported",
+    "requires_auth": "partial",
+    "can_probe": "partial"
+  },
+  "safety": {
+    "private_paths_readonly": ["~/Library/Application Support/Trae/**", "~/Library/Application Support/Trae CN/**"],
+    "never_write_paths": ["**/database.db", "**/*.vscdb", "**/*.sqlite"],
+    "destructive_commands": [],
+    "degradation_required": ["sessions.read", "sessions.send_message", "sessions.create", "sessions.continue", "sessions.pin", "sessions.archive"]
+  }
+}

package/adapters/trae/rules.md

@@ -0,0 +1,7 @@
+# Trae Adapter Rules
+
+Trae uses `.trae/rules/starwork.md` as the StarWork entry.
+
+Trae can discover project Skills from `.trae/skills/` and `.agents/skills/`. When the same Skill exists in both locations, `.trae/skills/` wins. StarWork still treats `.starwork/skills.json` as the project Skill fact source.
+
+Trae is a manual-handoff host in v0.1. It should not be described as supporting automatic background cross-session delivery.

package/adapters/trae/safety.md

@@ -0,0 +1,6 @@
+# Trae Adapter Safety
+
+- Do not read or write encrypted `database.db`.
+- Do not write Trae private state.
+- Do not rely on snapshot history as StarWork memory.
+- MultiAgent delivery should degrade to manual handoff unless a future stable API is confirmed.