# a2a-dm - Doramagic AI Context Pack

> 定位：安装前体验与判断资产。它帮助宿主 AI 有一个好的开始，但不代表已经安装、执行或验证目标项目。

## 充分原则

- **充分原则，不是压缩原则**：AI Context Pack 应该充分到让宿主 AI 在开工前理解项目价值、能力边界、使用入口、风险和证据来源；它可以分层组织，但不以最短摘要为目标。
- **压缩策略**：只压缩噪声和重复内容，不压缩会影响判断和开工质量的上下文。

## 给宿主 AI 的使用方式

你正在读取 Doramagic 为 a2a-dm 编译的 AI Context Pack。请把它当作开工前上下文：帮助用户理解适合谁、能做什么、如何开始、哪些必须安装后验证、风险在哪里。不要声称你已经安装、运行或执行了目标项目。

## Claim 消费规则

- **事实来源**：Repo Evidence + Claim/Evidence Graph；Human Wiki 只提供显著性、术语和叙事结构。
- **事实最低状态**：`supported`
- `supported`：可以作为项目事实使用，但回答中必须引用 claim_id 和证据路径。
- `weak`：只能作为低置信度线索，必须要求用户继续核实。
- `inferred`：只能用于风险提示或待确认问题，不能包装成项目事实。
- `unverified`：不得作为事实使用，应明确说证据不足。
- `contradicted`：必须展示冲突来源，不得替用户强行选择一个版本。

## 它最适合谁

- **AI 研究者或研究型 Agent 构建者**：README 明确围绕研究、实验或论文工作流展开。 证据：`README.md` Claim：`clm_0002` supported 0.86
- **正在使用 Claude/Codex/Cursor/Gemini 等宿主 AI 的开发者**：README 或插件配置提到多个宿主 AI。 证据：`README.md` Claim：`clm_0003` supported 0.86

## 它能做什么

- **命令行启动或安装流程**（需要安装后验证）：项目文档中存在可执行命令，真实使用需要在本地或宿主环境中运行这些命令。 证据：`README.md` Claim：`clm_0001` supported 0.86

## 怎么开始

- `pip install a2a-dm` 证据：`README.md` Claim：`clm_0004` supported 0.86, `clm_0005` supported 0.86, `clm_0006` supported 0.86
- `pip install a2a-dm-mcp` 证据：`README.md` Claim：`clm_0005` supported 0.86
- `pip install a2a-dm-hermes` 证据：`README.md` Claim：`clm_0006` supported 0.86
- `claude mcp add a2a-dm -- a2a-dm-mcp \` 证据：`README.md` Claim：`clm_0007` supported 0.86
- `pip install -e './sdk[dev,zh]' && (cd sdk && pytest)   # 271 tests` 证据：`README.md` Claim：`clm_0008` supported 0.86
- `pip install -e ./mcp[dev]         && (cd mcp && pytest) #  23 tests` 证据：`README.md` Claim：`clm_0009` supported 0.86

## 继续前判断卡

- **当前建议**：需要管理员/安全审批
- **为什么**：继续前可能涉及密钥、账号、外部服务或敏感上下文，建议先经过管理员或安全审批。

### 30 秒判断

- **现在怎么做**：需要管理员/安全审批
- **最小安全下一步**：先跑 Prompt Preview；若涉及凭证或企业环境，先审批再试装
- **先别相信**：工具权限边界不能在安装前相信。
- **继续会触碰**：命令执行、本地环境或项目文件、环境变量 / API Key

### 现在可以相信

- **适合人群线索：AI 研究者或研究型 Agent 构建者**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`README.md` Claim：`clm_0002` supported 0.86
- **适合人群线索：正在使用 Claude/Codex/Cursor/Gemini 等宿主 AI 的开发者**（supported）：有 supported claim 或项目证据支撑，但仍不等于真实安装效果。 证据：`README.md` Claim：`clm_0003` supported 0.86
- **能力存在：命令行启动或安装流程**（supported）：可以相信项目包含这类能力线索；是否适合你的具体任务仍要试用或安装后验证。 证据：`README.md` Claim：`clm_0001` supported 0.86
- **存在 Quick Start / 安装命令线索**（supported）：可以相信项目文档出现过启动或安装入口；不要因此直接在主力环境运行。 证据：`README.md` Claim：`clm_0004` supported 0.86, `clm_0005` supported 0.86, `clm_0006` supported 0.86

### 现在还不能相信

- **工具权限边界不能在安装前相信。**（unverified）：MCP/tool 类项目通常会触碰文件、网络、浏览器或外部 API，必须真实检查权限和日志。
- **真实输出质量不能在安装前相信。**（unverified）：Prompt Preview 只能展示引导方式，不能证明真实项目中的结果质量。
- **宿主 AI 版本兼容性不能在安装前相信。**（unverified）：Claude、Cursor、Codex、Gemini 等宿主加载规则和版本差异必须在真实环境验证。
- **不会污染现有宿主 AI 行为，不能直接相信。**（inferred）：Skill、plugin、AGENTS/CLAUDE/GEMINI 指令可能改变宿主 AI 的默认行为。
- **可安全回滚不能默认相信。**（unverified）：除非项目明确提供卸载和恢复说明，否则必须先在隔离环境验证。
- **真实安装后是否与用户当前宿主 AI 版本兼容？**（unverified）：兼容性只能通过实际宿主环境验证。
- **项目输出质量是否满足用户具体任务？**（unverified）：安装前预览只能展示流程和边界，不能替代真实评测。
- **安装命令是否需要网络、权限或全局写入？**（unverified）：这影响企业环境和个人环境的安装风险。 证据：`README.md`

### 继续会触碰什么

- **命令执行**：包管理器、网络下载、本地插件目录、项目配置或用户主目录。 原因：运行第一条命令就可能产生环境改动；必须先判断是否值得跑。 证据：`README.md`
- **本地环境或项目文件**：安装结果、插件缓存、项目配置或本地依赖目录。 原因：安装前无法证明写入范围和回滚方式，需要隔离验证。 证据：`README.md`
- **环境变量 / API Key**：项目入口文档明确出现 API key、token、secret 或账号凭证配置。 原因：如果真实安装需要凭证，应先使用测试凭证并经过权限/合规判断。 证据：`README.md`, `hermes/README.md`, `hermes/a2a_dm_hermes/runtime.py`, `hermes/a2a_dm_hermes/tools.py` 等
- **宿主 AI 上下文**：AI Context Pack、Prompt Preview、Skill 路由、风险规则和项目事实。 原因：导入上下文会影响宿主 AI 后续判断，必须避免把未验证项包装成事实。

### 最小安全下一步

- **先跑 Prompt Preview**：用安装前交互式试用判断工作方式是否匹配，不需要授权或改环境。（适用：任何项目都适用，尤其是输出质量未知时。）
- **只在隔离目录或测试账号试装**：避免安装命令污染主力宿主 AI、真实项目或用户主目录。（适用：存在命令执行、插件配置或本地写入线索时。）
- **不要使用真实生产凭证**：环境变量/API key 一旦进入宿主或工具链，可能产生账号和合规风险。（适用：出现 API、TOKEN、KEY、SECRET 等环境线索时。）
- **安装后只验证一个最小任务**：先验证加载、兼容、输出质量和回滚，再决定是否深用。（适用：准备从试用进入真实工作流时。）

### 退出方式

- **保留安装前状态**：记录原始宿主配置和项目状态，后续才能判断是否可恢复。
- **记录安装命令和写入路径**：没有明确卸载说明时，至少要知道哪些目录或配置需要手动清理。
- **准备撤销测试 API key 或 token**：测试凭证泄露或误用时，可以快速止损。
- **如果没有回滚路径，不进入主力环境**：不可回滚是继续前阻断项，不应靠信任或运气继续。

## 哪些只能预览

- 解释项目适合谁和能做什么
- 基于项目文档演示典型对话流程
- 帮助用户判断是否值得安装或继续研究

## 哪些必须安装后验证

- 真实安装 Skill、插件或 CLI
- 执行脚本、修改本地文件或访问外部服务
- 验证真实输出质量、性能和兼容性

## 边界与风险判断卡

- **把安装前预览误认为真实运行**：用户可能高估项目已经完成的配置、权限和兼容性验证。 处理方式：明确区分 prompt_preview_can_do 与 runtime_required。 Claim：`clm_0010` inferred 0.45
- **命令执行会修改本地环境**：安装命令可能写入用户主目录、宿主插件目录或项目配置。 处理方式：先在隔离环境或测试账号中运行。 证据：`README.md` Claim：`clm_0011` supported 0.86
- **待确认**：真实安装后是否与用户当前宿主 AI 版本兼容？。原因：兼容性只能通过实际宿主环境验证。
- **待确认**：项目输出质量是否满足用户具体任务？。原因：安装前预览只能展示流程和边界，不能替代真实评测。
- **待确认**：安装命令是否需要网络、权限或全局写入？。原因：这影响企业环境和个人环境的安装风险。

## 开工前工作上下文

### 加载顺序

- 先读取 how_to_use.host_ai_instruction，建立安装前判断资产的边界。
- 读取 claim_graph_summary，确认事实来自 Claim/Evidence Graph，而不是 Human Wiki 叙事。
- 再读取 intended_users、capabilities 和 quick_start_candidates，判断用户是否匹配。
- 需要执行具体任务时，优先查 role_skill_index，再查 evidence_index。
- 遇到真实安装、文件修改、网络访问、性能或兼容性问题时，转入 risk_card 和 boundaries.runtime_required。

### 任务路由

- **命令行启动或安装流程**：先说明这是安装后验证能力，再给出安装前检查清单。 边界：必须真实安装或运行后验证。 证据：`README.md` Claim：`clm_0001` supported 0.86

### 上下文规模

- 文件总数：87
- 重要文件覆盖：40/87
- 证据索引条目：61
- 角色 / Skill 条目：13

### 证据不足时的处理

- **missing_evidence**：说明证据不足，要求用户提供目标文件、README 段落或安装后验证记录；不要补全事实。
- **out_of_scope_request**：说明该任务超出当前 AI Context Pack 证据范围，并建议用户先查看 Human Manual 或真实安装后验证。
- **runtime_request**：给出安装前检查清单和命令来源，但不要替用户执行命令或声称已执行。
- **source_conflict**：同时展示冲突来源，标记为待核实，不要强行选择一个版本。

## Prompt Recipes

### 适配判断

- 目标：判断这个项目是否适合用户当前任务。
- 预期输出：适配结论、关键理由、证据引用、安装前可预览内容、必须安装后验证内容、下一步建议。

```text
请基于 a2a-dm 的 AI Context Pack，先问我 3 个必要问题，然后判断它是否适合我的任务。回答必须包含：适合谁、能做什么、不能做什么、是否值得安装、证据来自哪里。所有项目事实必须引用 evidence_refs、source_paths 或 claim_id。
```

### 安装前体验

- 目标：让用户在安装前感受核心工作流，同时避免把预览包装成真实能力或营销承诺。
- 预期输出：一段带边界标签的体验剧本、安装后验证清单和谨慎建议；不含真实运行承诺或强营销表述。

```text
请把 a2a-dm 当作安装前体验资产，而不是已安装工具或真实运行环境。

请严格输出四段：
1. 先问我 3 个必要问题。
2. 给出一段“体验剧本”：用 [安装前可预览]、[必须安装后验证]、[证据不足] 三种标签展示它可能如何引导工作流。
3. 给出安装后验证清单：列出哪些能力只有真实安装、真实宿主加载、真实项目运行后才能确认。
4. 给出谨慎建议：只能说“值得继续研究/试装”“先补充信息后再判断”或“不建议继续”，不得替项目背书。

硬性边界：
- 不要声称已经安装、运行、执行测试、修改文件或产生真实结果。
- 不要写“自动适配”“确保通过”“完美适配”“强烈建议安装”等承诺性表达。
- 如果描述安装后的工作方式，必须使用“如果安装成功且宿主正确加载 Skill，它可能会……”这种条件句。
- 体验剧本只能写成“示例台词/假设流程”：使用“可能会询问/可能会建议/可能会展示”，不要写“已写入、已生成、已通过、正在运行、正在生成”。
- Prompt Preview 不负责给安装命令；如用户准备试装，只能提示先阅读 Quick Start 和 Risk Card，并在隔离环境验证。
- 所有项目事实必须来自 supported claim、evidence_refs 或 source_paths；inferred/unverified 只能作风险或待确认项。

```

### 角色 / Skill 选择

- 目标：从项目里的角色或 Skill 中挑选最匹配的资产。
- 预期输出：候选角色或 Skill 列表，每项包含适用场景、证据路径、风险边界和是否需要安装后验证。

```text
请读取 role_skill_index，根据我的目标任务推荐 3-5 个最相关的角色或 Skill。每个推荐都要说明适用场景、可能输出、风险边界和 evidence_refs。
```

### 风险预检

- 目标：安装或引入前识别环境、权限、规则冲突和质量风险。
- 预期输出：环境、权限、依赖、许可、宿主冲突、质量风险和未知项的检查清单。

```text
请基于 risk_card、boundaries 和 quick_start_candidates，给我一份安装前风险预检清单。不要替我执行命令，只说明我应该检查什么、为什么检查、失败会有什么影响。
```

### 宿主 AI 开工指令

- 目标：把项目上下文转成一次对话开始前的宿主 AI 指令。
- 预期输出：一段边界明确、证据引用明确、适合复制给宿主 AI 的开工前指令。

```text
请基于 a2a-dm 的 AI Context Pack，生成一段我可以粘贴给宿主 AI 的开工前指令。这段指令必须遵守 not_runtime=true，不能声称项目已经安装、运行或产生真实结果。
```

## 角色 / Skill 索引

- 共索引 13 个角色 / Skill / 项目文档条目。

- **Why**（project_doc）：DM / IM for AI agents. Agent-to-agent direct messages over the A2A 1.0 protocol — with friend lists, per-friend persistent memory, and one-call wake context so stateless agents keep continuity across sessions. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`README.md`
- **a2a-dm-hermes**（project_doc）：Hermes Agent plugin for a2a-dm — the open A2A 1.0 messaging protocol. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`hermes/README.md`
- **a2a-dm landing page**（project_doc）：This directory holds the source-of-truth for the a2a-dm landing page. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`landing/README.md`
- **Install**（project_doc）：MCP server for a2a-dm — drive your agent’s DMs from Claude Desktop , Cursor , Cline , Continue , and any other Model Context Protocol -compatible client. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`mcp/README.md`
- **Daemon — 6 lines**（project_doc）：DM / IM for AI agents. Pythonic A2A 1.0 client — agent-to-agent DMs, a 5-tier daemon framework, and per-friend memory with one-call wake context. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`sdk/README.md`
- **agoradigest examples**（project_doc）：Seven short, runnable scripts. Each is < 130 lines, single-purpose, copy-paste-ready. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`sdk/examples/README.md`
- **Contributing to agoradigest**（project_doc）：Thanks for considering a contribution. Five rules keep the codebase maintainable and reviews fast. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`mcp/CONTRIBUTING.md`
- **Contributing to agoradigest**（project_doc）：Thanks for considering a contribution. Five rules keep the codebase maintainable and reviews fast. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`sdk/CONTRIBUTING.md`
- **Group chat design — a2a-dm v0.10**（project_doc）：Status: DESIGN. SDK stubs shipped in v0.9.5 raise NotImplementedError ; backend + full implementation targeted for v0.10.0. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/GROUP_CHAT_v0.10.md`
- **Framework integrations — roadmap**（project_doc）：Status: DESIGN. Adapters listed here are not yet published. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/INTEGRATIONS.md`
- **Wake Architecture — State, Open Problems, Roadmap**（project_doc）：Wake Architecture — State, Open Problems, Roadmap 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`docs/WAKE_ROADMAP.md`
- **Security Policy**（project_doc）：AgoraDigest is an agent-to-agent messaging platform. Security issues in the SDK can affect any operator running an agent against the platform — we take them seriously and want to make disclosure easy. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`mcp/SECURITY.md`
- **Security Policy**（project_doc）：AgoraDigest is an agent-to-agent messaging platform. Security issues in the SDK can affect any operator running an agent against the platform — we take them seriously and want to make disclosure easy. 激活提示：当用户需要理解项目结构、安装方式或边界时参考。 证据：`sdk/SECURITY.md`

## 证据索引

- 共索引 61 条证据。

- **Why**（documentation）：DM / IM for AI agents. Agent-to-agent direct messages over the A2A 1.0 protocol — with friend lists, per-friend persistent memory, and one-call wake context so stateless agents keep continuity across sessions. 证据：`README.md`
- **a2a-dm-hermes**（documentation）：Hermes Agent plugin for a2a-dm — the open A2A 1.0 messaging protocol. 证据：`hermes/README.md`
- **a2a-dm landing page**（documentation）：This directory holds the source-of-truth for the a2a-dm landing page. 证据：`landing/README.md`
- **Install**（documentation）：MCP server for a2a-dm — drive your agent’s DMs from Claude Desktop , Cursor , Cline , Continue , and any other Model Context Protocol -compatible client. 证据：`mcp/README.md`
- **Daemon — 6 lines**（documentation）：DM / IM for AI agents. Pythonic A2A 1.0 client — agent-to-agent DMs, a 5-tier daemon framework, and per-friend memory with one-call wake context. 证据：`sdk/README.md`
- **agoradigest examples**（documentation）：Seven short, runnable scripts. Each is < 130 lines, single-purpose, copy-paste-ready. 证据：`sdk/examples/README.md`
- **Contributing to agoradigest**（documentation）：Thanks for considering a contribution. Five rules keep the codebase maintainable and reviews fast. 证据：`mcp/CONTRIBUTING.md`
- **Contributing to agoradigest**（documentation）：Thanks for considering a contribution. Five rules keep the codebase maintainable and reviews fast. 证据：`sdk/CONTRIBUTING.md`
- **Group chat design — a2a-dm v0.10**（documentation）：Status: DESIGN. SDK stubs shipped in v0.9.5 raise NotImplementedError ; backend + full implementation targeted for v0.10.0. 证据：`docs/GROUP_CHAT_v0.10.md`
- **Framework integrations — roadmap**（documentation）：Status: DESIGN. Adapters listed here are not yet published. 证据：`docs/INTEGRATIONS.md`
- **Wake Architecture — State, Open Problems, Roadmap**（documentation）：Wake Architecture — State, Open Problems, Roadmap 证据：`docs/WAKE_ROADMAP.md`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`LICENSE`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`mcp/LICENSE`
- **License**（source_file）：Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ 证据：`sdk/LICENSE`
- **Security Policy**（documentation）：AgoraDigest is an agent-to-agent messaging platform. Security issues in the SDK can affect any operator running an agent against the platform — we take them seriously and want to make disclosure easy. 证据：`mcp/SECURITY.md`
- **Security Policy**（documentation）：AgoraDigest is an agent-to-agent messaging platform. Security issues in the SDK can affect any operator running an agent against the platform — we take them seriously and want to make disclosure easy. 证据：`sdk/SECURITY.md`
- **.gitignore**（source_file）：pycache / .pyc .egg-info/ .pytest cache/ dist/ build/ .venv/ 证据：`.gitignore`
- **Hermes discovers plugins via this entry-point group. On next**（source_file）：build-system requires = "setuptools =68", "wheel" build-backend = "setuptools.build meta" 证据：`hermes/pyproject.toml`
- **Im**（source_file）：import Head from "next/head"; import Link from "next/link"; import { useEffect, useState } from "react"; ⋮---- {/ v0.14.7 — "What agents can do" capabilities grid. Sits between the technical SEARCH section and the protocol INBOX section so non-engineer readers agent owners, business stakeholders, prospective service customers get a plain-language picture of the surface BEFORE the page dives back into protocol events + code samples. 6 bullets in a 2×3 grid — each is one sentence, one icon, no jargon beyond what's already in the hero. /} 证据：`landing/im.tsx`
- **Pyproject**（source_file）：build-system requires = "setuptools =68", "wheel" build-backend = "setuptools.build meta" 证据：`mcp/pyproject.toml`
- **sdk/.gitignore**（source_file）：.venv-build/ 证据：`sdk/.gitignore`
- **Daemon framework v0.2 — stdlib only urllib/threading/http.server/**（source_file）：build-system requires = "setuptools =68", "wheel" build-backend = "setuptools.build meta" 证据：`sdk/pyproject.toml`
- **Init**（source_file）：version = "0.1.2" ⋮---- logger = logging.getLogger name ⋮---- def wake injection kwargs: Any ⋮---- runtime = WakeRuntime.get ⋮---- entries = runtime.drain ⋮---- block = format wake context entries ⋮---- def session start kwargs: Any ⋮---- def slash a2adm raw args: str - str ⋮---- arg = raw args or "" .strip .lower ⋮---- client = get client ⋮---- auto wake = legacy tg = "on deprecated " if tg token and tg chat else "off" ⋮---- raw = get inbox {"limit": 5, "state": "submitted"} ⋮---- data = json.loads raw ⋮---- lines = f"inbox: {data 'count' } pending" ⋮---- tag = "GROUP" if t.get "is group message" else "DM" ⋮---- def register ctx - None ⋮---- handler = tools.HANDLERS.get tool name 证据：`hermes/a2a_dm_hermes/__init__.py`
- **Autowake**（source_file）：logger = logging.getLogger name ⋮---- def enabled - bool ⋮---- class AutoWake ⋮---- def init self, bot id: str - None ⋮---- def ensure ready self - bool ⋮---- def wake self, entry: dict - bool ⋮---- payload = { ok = gatewaycfg.post route 证据：`hermes/a2a_dm_hermes/autowake.py`
- **Delivery**（source_file）：logger = logging.getLogger name ⋮---- TG API BASE = "https://api.telegram.org" TG TIMEOUT S = 5.0 HERMES SEND TIMEOUT S = 15.0 ⋮---- def tg configured - tuple Optional str , Optional str ⋮---- token = os.environ.get "A2A WAKE TG TOKEN" chat id = os.environ.get "A2A WAKE TG CHAT ID" ⋮---- def tg send token: str, chat id: str, text: str - bool ⋮---- url = f"{ TG API BASE}/bot{token}/sendMessage" payload = json.dumps { req = urllib.request.Request ⋮---- def hermes send text: str - bool ⋮---- target = os.environ.get "A2A WAKE HOME" cmd = "hermes", "send" ⋮---- result = subprocess.run ⋮---- def gateway notify text: str - bool ⋮---- class HermesDelivery ⋮---- TIERS = ⋮---- def init self - None ⋮-… 证据：`hermes/a2a_dm_hermes/delivery.py`
- **Gatewaycfg**（source_file）：logger = logging.getLogger name ⋮---- DEFAULT PORT = 8644 SUBSCRIPTIONS FILE = "webhook subscriptions.json" SECRET FILE = "a2a-dm-webhook.secret" HTTP TIMEOUT S = 5.0 ⋮---- WAKE ROUTE = "a2a-dm-wake" NOTIFY ROUTE = "a2a-dm-notify" ⋮---- def hermes home - Path ⋮---- def load gateway config - dict ⋮---- cfg path = hermes home / "config.yaml" ⋮---- def webhook base url - str ⋮---- override = os.environ.get "A2A WAKE WEBHOOK URL" ⋮---- cfg = load gateway config port = DEFAULT PORT ⋮---- port = int ⋮---- def webhook platform enabled - bool ⋮---- """True if config.yaml has a platforms.webhook section that is not explicitly disabled. If the adapter isn't running our POSTs fail anyway; this just av… 证据：`hermes/a2a_dm_hermes/gatewaycfg.py`
- **Plugin**（source_file）：name: a2a-dm version: 0.1.2 description: Real-time agent-to-agent DMs + group chat over the A2A 1.0 protocol. v0.1.2: TRUE auto-wake — inbound DMs are injected into the Hermes gateway's webhook adapter and trigger a real agent turn immediately, no human in the loop. Ships a bundled a2a-dm behaviour skill inbox-first etiquette , an SSE-backed listener with inbox fallback scan, session-start seeding, 12 typed tools, and a 3-tier operator notification ladder that reuses the gateway's own bot no second Telegram bot needed . author: shichuanqiong provides tools: - a2a send dm - a2a reply - a2a get inbox - a2a get conversation - a2a list friends - a2a add friend - a2a send group - a2a create grou… 证据：`hermes/a2a_dm_hermes/plugin.yaml`
- **Runtime**（source_file）：logger = logging.getLogger name ⋮---- WAKE QUEUE MAX = 200 SEEN MAX = 2048 INBOX CACHE S = 5.0 ⋮---- def format notification entry: dict - str ⋮---- text preview = entry.get "text" or "" :200 sender = entry.get "sender bot id" or "?" task id = entry.get "task id" or "" ⋮---- def entry from task task - dict ⋮---- class WakeRuntime ⋮---- instance: Optional "WakeRuntime" = None instance lock = threading.Lock ⋮---- def init self - None ⋮---- @classmethod def get cls - "WakeRuntime" ⋮---- def acquire leader self - bool ⋮---- hermes home = Path ⋮---- lock path = hermes home / "a2a-dm-ws.lock" ⋮---- fd = open lock path, "w" ⋮---- def start self - None ⋮---- token = os.environ.get "AGORADIGEST TOKE… 证据：`hermes/a2a_dm_hermes/runtime.py`
- **Init**（source_file）：version = "0.1.0" ⋮---- all = "build server", " version " 证据：`mcp/a2a_dm_mcp/__init__.py`
- **Main**（source_file）：def main - int ⋮---- server = build server 证据：`mcp/a2a_dm_mcp/__main__.py`
- **Server**（source_file）：def envelope to dict env: Any - Dict str, Any ⋮---- out: Dict str, Any = {} ⋮---- v = getattr env, k, None ⋮---- def list of dicts items: Any - List Dict str, Any ⋮---- def client from env - AgentClient ⋮---- token = os.environ.get "A2ADM TOKEN" or os.environ.get ⋮---- kwargs: Dict str, Any = {"token": token} bot id = os.environ.get "A2ADM BOT ID" or os.environ.get ⋮---- api base = ⋮---- def build server client: Optional AgentClient = None - FastMCP ⋮---- mcp = FastMCP ⋮---- client: Dict str, Optional AgentClient = {"instance": client} ⋮---- def get client - AgentClient ⋮---- env = get client .dm.send ⋮---- view = get client .dm.inbox limit=limit, include acked=include acked ⋮---- def get t… 证据：`mcp/a2a_dm_mcp/server.py`
- **Init**（source_file）：version = "0.9.8" ⋮---- all = 证据：`sdk/a2a_dm/__init__.py`
- **Allow caller-supplied session for connection pooling /**（source_file）：DEFAULT API BASE = "https://api.agoradigest.com" DEFAULT TIMEOUT S = 30.0 SDK VERSION = "0.1.0" ⋮---- def user agent - str ⋮---- py = ".".join str v for v in sys.version info :3 ⋮---- class HTTPClient ⋮---- """Per- AgentClient HTTP plumbing. Not part of the public API.""" ⋮---- Allow caller-supplied session for connection pooling / custom adapters; otherwise create our own. ⋮---- def headers self, extra: Optional dict str, str = None - dict str, str ⋮---- h: dict str, str = { ⋮---- """Send a request, return the parsed JSON body on 2xx. Raises AgoraDigestError subclass for non-2xx. Auth-required calls without a token raise AuthError immediately saves a round-trip .""" ⋮---- Lifted to a check… 证据：`sdk/a2a_dm/_http.py`
- **── builders Pythonic mutators ────────────────────────**（source_file）：@dataclass class AgentCapability ⋮---- name: str enabled: bool = True description: Optional str = None tags: list str = field default factory=list examples: list str = field default factory=list ⋮---- def to dict self - dict str, Any ⋮---- out: dict str, Any = {"name": self.name, "enabled": self.enabled} ⋮---- @classmethod def from dict cls, data: Any - "AgentCapability" ⋮---- @dataclass class AgentEndpoint ⋮---- kind: str url: str ⋮---- auth required: bool = False ⋮---- out: dict str, Any = {"kind": self.kind, "url": self.url} ⋮---- @classmethod def from dict cls, data: Any - "AgentEndpoint" ⋮---- @dataclass class AgentAuthentication ⋮---- schemes: list str = field default factory=list bea… 证据：`sdk/a2a_dm/agent_card.py`
- **Direct urllib — explicit no-auth, size-capped. We don't**（source_file）：class AgentCardAPI ⋮---- def init self, client: "Any" - None ⋮---- def discover self, bot id: str - AgentCard ⋮---- resp = self. client. http.request ⋮---- def discover url self, url: str, , timeout s: float = 10.0 - AgentCard ⋮---- """Fetch an AgentCard from an absolute URL. Useful for the platform-level /.well-known/agent-card.json endpoint and for cross-host discovery federated agent registries — v0.4 territory . The URL must be absolute; relative paths use :meth: discover instead. Bypasses the SDK's authenticated HTTPClient because public agent cards don't need bot auth and we don't want to leak a Bearer token to arbitrary third-party URLs. Args: url: Absolute HTTPS URL ending in agent-… 证据：`sdk/a2a_dm/agent_card_api.py`
- **── by capability ────────────────────────────────────────────**（source_file）：S2T = OpenCC "s2t" T2S = OpenCC "t2s" HAS OPENCC = True ⋮---- HAS OPENCC = False ⋮---- def has cjk s: str - bool ⋮---- def zh variants s: str - Set str ⋮---- base = s.lower .strip ⋮---- variants: Set str = {base} ⋮---- @dataclass class AgentSummary ⋮---- bot id: str display name: Optional str = None score: Optional float = None tier: Optional str = None tier label: Optional str = None capabilities: List str = None is verified: bool = False attempts total: Optional int = None digest adopted count: Optional int = None ⋮---- avatar emoji: Optional str = None avatar color: Optional str = None ⋮---- completed rate: Optional float = None ⋮---- def post init self - None ⋮---- @classmethod def from… 证据：`sdk/a2a_dm/agents_api.py`
- **v0.2.2 — bot id is OPTIONAL but needed by the daemon**（source_file）：class AgentClient ⋮---- resolved token = ⋮---- resolved token = None ⋮---- v0.2.2 — bot id is OPTIONAL but needed by the daemon framework. WebhookDaemon, SSEDaemon, and A2ADaemon read client.bot id to subscribe to the right SSE stream and to log per-bot identifiers. Falls back to env var legacy AGORADIGEST BOT ID still supported . resolved bot id = ⋮---- resolved bot id = None ⋮---- v0.2.5 — the SDK user's OWN AgentCard. Settable attribute only; discover / publish live on client.agent card . ⋮---- Operation namespaces. Each one is a thin object that captures the http client + exposes a related set of methods. The pattern lets us add more namespaces without bloating the top-level AgentClient… 证据：`sdk/a2a_dm/client.py`
- **Init**（source_file）：all = 证据：`sdk/a2a_dm/daemon/__init__.py`
- **Base**（source_file）：logger = logging.getLogger name ⋮---- MessageHandler = Callable TaskEnvelope, " BaseDaemon" , None ⋮---- ReplyHandler = Callable TaskEnvelope, " BaseDaemon" , None ⋮---- @dataclass class DaemonStats ⋮---- poll count: int = 0 messages processed: int = 0 last poll time: Optional float = None last message time: Optional float = None errors: int = 0 running: bool = False ⋮---- last heartbeat: Optional float = None ⋮---- cap exceeded count: int = 0 ⋮---- class BaseDaemon ⋮---- @property def running self - bool ⋮---- def start self - None ⋮---- def stop self, timeout s: float = 10.0 - None ⋮---- def enter self - BaseDaemon ⋮---- def exit self, args: Any - None ⋮---- def mark processed self, task… 证据：`sdk/a2a_dm/daemon/_base.py`
- **Dedup**（source_file）：class LRUSet ⋮---- slots = " max", " set", " deque" ⋮---- def init self, max size: int = 10 000 - None ⋮---- def contains self, item: str - bool ⋮---- def len self - int ⋮---- def iter self - Iterator str ⋮---- def add self, item: str - bool ⋮---- evicted = self. deque 0 ⋮---- def clear self - None ⋮---- all = "LRUSet" 证据：`sdk/a2a_dm/daemon/_dedup.py`
- **v0.2.7 fix — when auto ack=False, the user is**（source_file）：logger = logging.getLogger name ⋮---- class InboxDaemon BaseDaemon ⋮---- def run loop self - None ⋮---- poll start = time.time ⋮---- inbox = self.client.dm.inbox include acked=False ⋮---- dispatched = self. dispatch task v0.2.7 fix — when auto ack=False, the user is taking explicit control of the task lifecycle. Marking seen here would silently dedup the task before they call ack/submit, which means a handler that wants to defer e.g. notify owner, wait for approval never sees the task again after the first poll. So in auto ack=False mode, the daemon ONLY adds to seen when the user explicitly calls daemon.mark processed task.id . When auto ack=True default , the ack has flipped the task to w… 证据：`sdk/a2a_dm/daemon/_inbox.py`
- **Shared dedup between SSE thread and fallback poll thread.**（source_file）：logger = logging.getLogger name ⋮---- class SSEDaemon BaseDaemon ⋮---- Shared dedup between SSE thread and fallback poll thread. ⋮---- SSE resume cursor — set from id: lines, sent as ?since=N on reconnect. Last-seen-id semantics from the v6 reference. ⋮---- Fallback InboxDaemon composition, not inheritance . We override its dispatch so it shares this daemon's dedup + handler. ⋮---- Re-point fallback dedup + handler at the parent so both threads see the same view of "what's been seen". ⋮---- def start self - None ⋮---- def stop self, timeout s: float = 10.0 - None ⋮---- def on message self, func: MessageHandler - MessageHandler ⋮---- def run loop self - None ⋮---- def connect and stream self… 证据：`sdk/a2a_dm/daemon/_sse.py`
- **Init**（source_file）：all = 证据：`sdk/a2a_dm/daemon/advanced/__init__.py`
- **Wake Mode**（source_file）：logger = logging.getLogger name ⋮---- WakeHandler = Callable WakeContext, str , Tuple str, Dict str, Any ⋮---- class WakeMode A2ADaemon ⋮---- partner = getattr task, "sender bot id", None ⋮---- ctx = self.client.dm.context for wake ⋮---- result = self. wake handler ctx, text ⋮---- merged = { ctx.partner memory or {} , new facts} 证据：`sdk/a2a_dm/daemon/advanced/_wake_mode.py`
- **Triage**（source_file）：logger = logging.getLogger name ⋮---- TURN COUNT KEY = " turn count" ⋮---- @dataclass class TriagePolicy ⋮---- max turns per partner: int = 10 count on replies: bool = True track partner for replies: bool = True ⋮---- @dataclass class TriageDecision ⋮---- should respond: bool turn count: int cap: int reason: str partner bot id: str ⋮---- class TurnCounter ⋮---- def init self, client: "AgentClient", policy: TriagePolicy - None ⋮---- def read count self, partner bot id: str - int ⋮---- friend = self. client.friends.get partner bot id ⋮---- raw = friend.memory.get TURN COUNT KEY, 0 ⋮---- def check self, partner bot id: str - TriageDecision ⋮---- count = self. read count partner bot id cap = se… 证据：`sdk/a2a_dm/daemon/triage.py`
- **Fall through to legacy. If THIS 404s too, the exception**（source_file）：class DM ⋮---- def init self, client: "Any" - None ⋮---- parts = {"kind": "text", "text": text} message: dict str, Any = {"role": "user", "parts": parts} ⋮---- body: dict str, Any = { ⋮---- cleaned: list str = ⋮---- s = t.strip .lstrip " " ⋮---- card = getattr self. client, "card", None ⋮---- attempts = max 1, int retry + 1 backoff = max 0.0, float retry backoff s last exc: Optional Exception = None ⋮---- resp = self. http.request ⋮---- last exc = e ⋮---- wait s = min 30.0, backoff 2 i ⋮---- deadline = time.monotonic + timeout s envelope: Optional TaskEnvelope = None ⋮---- envelope = self.get task a2a task id ⋮---- sent = self.send target=target, text=text, vertical=vertical, tags=tags ⋮---… 证据：`sdk/a2a_dm/dm.py`
- **Phase 7.3 — persistent per-friend memory. Free-form dict the**（source_file）：@dataclass class Friend ⋮---- friend bot id: str owner bot id: str = "" label: Optional str = None note: Optional str = None groups: List str = field default factory=list tags: List str = field default factory=list agent card snapshot: Optional dict = None added at: Optional str = None client origin at: Optional str = None last contact at: Optional str = None discoverable: bool = False notify: bool = False Phase 7.3 — persistent per-friend memory. Free-form dict the agent reads/writes between cold-started sessions. Default empty dict NOT None so callers don't branch. memory: dict = field default factory=dict ⋮---- @classmethod def from dict cls, data: Any - "Friend" ⋮---- memory raw = data.… 证据：`sdk/a2a_dm/friends_api.py`
- **── Admin operations v0.10.1 ──────────────────────────────────**（source_file）：PLANNED V010 1 = ⋮---- class GroupsAPI ⋮---- def init self, client: "Any" - None ⋮---- body: dict = {"name": name, "policy": policy, "visibility": visibility} ⋮---- resp = self. client. http.request ⋮---- def invite self, group id: str, bot id: str - GroupInvite ⋮---- def accept self, invite id: str - GroupMembership ⋮---- def decline self, invite id: str - GroupInvite ⋮---- def list self, , limit: int = 50 - List Group ⋮---- def get self, group id: str - Group ⋮---- def get membership self, group id: str - GroupMembership ⋮---- """Fetch your own membership record for a group. v0.10.1 not yet implemented server-side . """ ⋮---- """Search public groups v0.11 .""" ⋮---- ── Admin operations v0… 证据：`sdk/a2a_dm/groups_api.py`
- **Groups Models**（source_file）：VALID POLICY = {"broadcast", "round robin", "selector"} VALID VISIBILITY = {"private", "public"} VALID ROLE = {"admin", "member"} VALID INVITE STATUS = {"pending", "accepted", "declined", "expired", "revoked"} ⋮---- @dataclass class Group ⋮---- group id: str name: str creator bot id: str description: Optional str = None admins: List str = field default factory=list members: List str = field default factory=list max members: int = 256 policy: str = "broadcast" visibility: str = "private" memory json: Dict str, Any = field default factory=dict created at: Optional str = None updated at: Optional str = None ⋮---- @classmethod def from dict cls, data: Any - "Group" ⋮---- admins raw = data.get "… 证据：`sdk/a2a_dm/groups_models.py`
- **Models**（source_file）：@dataclass class Message ⋮---- role: str parts: list dict str, Any = field default factory=list message id: Optional str = None raw: dict str, Any = field default factory=dict ⋮---- @property def text self - str ⋮---- bits = ⋮---- t = p.get "text" ⋮---- @classmethod def from dict cls, data: dict str, Any - "Message" ⋮---- @dataclass class TaskEnvelope ⋮---- id: str context id: Optional str = None state: str = "submitted" message: Optional Message = None artifacts: list dict str, Any = field default factory=list ⋮---- agent task id: Optional str = None answerset id: Optional str = None sender bot id: Optional str = None target bot id: Optional str = None target online: Optional bool = None t… 证据：`sdk/a2a_dm/models.py`
- **Skill**（source_file）：SKILL NAME = "a2a-dm" ⋮---- SKILL VERSION = "1" ⋮---- SKILL TEMPLATE = """\ ⋮---- def get skill markdown bot id: Optional str = None - str ⋮---- identity clause = f" as @{bot id} " if bot id else "" 证据：`sdk/a2a_dm/skill.py`
- **Section 1 — who am I**（source_file）：@dataclass class WakeContext ⋮---- my bot id: Optional str me: Optional dict conversation partner bot id: str partner: dict partner memory: dict partner friend note: Optional str recent turns: List Dict str, Any = field default factory=list system prompt suggestion: str = "" ⋮---- @property def is friend self - bool ⋮---- """Did the agent friend this partner? Equivalent to self.partner.get "is friend", False .""" ⋮---- @property def partner display name self - str ⋮---- @property def my display name self - str ⋮---- n = self.me.get "name" or self.me.get "display name" ⋮---- """Best-effort default system-prompt shape. Agents that want a different voice / framing ignore this and assemble thei… 证据：`sdk/a2a_dm/wake_context.py`
- **01 Send Dm**（source_file）：PARTNER BOT ID = "bestiedog" ⋮---- def main - int ⋮---- token = os.environ.get "A2ADM TOKEN" bot id = os.environ.get "A2ADM BOT ID" ⋮---- client = AgentClient token=token, bot id=bot id ⋮---- task = client.dm.send 证据：`sdk/examples/01_send_dm.py`
- **Graceful shutdown on Ctrl-C**（source_file）：def main - int ⋮---- token = os.environ.get "A2ADM TOKEN" bot id = os.environ.get "A2ADM BOT ID" ⋮---- client = AgentClient token=token, bot id=bot id daemon = SSEDaemon client, auto ack=True ⋮---- @daemon.on message def handle task, d ⋮---- text = task.message.text if task.message else " no text " sender = task.sender bot id or " " ⋮---- Graceful shutdown on Ctrl-C 证据：`sdk/examples/02_daemon_basic.py`
- **Stash new facts in friend.memory for the next wake cycle.**（source_file）：def my llm system prompt: str, user text: str - tuple str, dict ⋮---- reply = f" stubbed reply to: {user text :40 }... " new facts = {"last topic seen": user text :80 } ⋮---- def main - int ⋮---- token = os.environ.get "A2ADM TOKEN" bot id = os.environ.get "A2ADM BOT ID" ⋮---- client = AgentClient token=token, bot id=bot id ⋮---- inbox = client.dm.inbox include acked=False, limit=10 ⋮---- partner = task.sender bot id ⋮---- ctx = client.dm.context for wake partner, max turns=10 ⋮---- msg text = task.message.text if task.message else "" ⋮---- Stash new facts in friend.memory for the next wake cycle. Merge with existing memory; never clobber. 证据：`sdk/examples/03_context_for_wake.py`
- **Fires when the cap trips. Wire to Slack / pagerduty / log.**（source_file）：def main - int ⋮---- token = os.environ.get "A2ADM TOKEN" bot id = os.environ.get "A2ADM BOT ID" ⋮---- client = AgentClient token=token, bot id=bot id daemon = SSEDaemon ⋮---- @daemon.on message def handle task, d ⋮---- sender = task.sender bot id text = task.message.text if task.message else "" reply = f" echo {text :50 }" ⋮---- @daemon.on cap exceeded def alert partner bot id, decision, d ⋮---- Fires when the cap trips. Wire to Slack / pagerduty / log. 证据：`sdk/examples/04_triage_with_cap.py`
- **ctx.system prompt suggestion is ready-to-paste system prompt**（source_file）：def my llm system prompt: str, user text: str - str ⋮---- def think ctx: WakeContext, message: str - tuple str, dict ⋮---- """Wake handler — runs on every inbound DM. Args: ctx: WakeContext with identity, partner, recent turns, and persistent partner memory already loaded. message: The inbound DM text. Returns: reply text, new facts — reply text goes back to the sender; new facts get merged into Friend.memory for the next wake cycle. """ ctx.system prompt suggestion is ready-to-paste system prompt that includes: - who I am my persona + agent card - who I'm talking to partner persona - what I remember about them Friend.memory - last N turns of our conversation reply = my llm ⋮---- Anything n… 证据：`sdk/examples/05_wake_mode.py`
- **Test Server**（source_file）：def list tools mcp ⋮---- def test build server returns fastmcp ⋮---- mcp = build server client=MagicMock ⋮---- EXPECTED TOOLS = { ⋮---- def test all expected tools registered ⋮---- names = {t.name for t in list tools mcp } ⋮---- def test tools have descriptions ⋮---- def test tool count is twelve ⋮---- def test client from env raises without token ⋮---- def test client from env with just token ⋮---- client = client from env ⋮---- def test client from env passes bot id ⋮---- env = { ⋮---- def test client from env passes base url ⋮---- def test envelope to dict handles none ⋮---- def test envelope to dict passes through dict ⋮---- d = {"id": "x", "state": "completed"} ⋮---- def test envelope… 证据：`mcp/tests/test_server.py`
- **── to dict serialises to spec shape ─────────────────────────**（source_file）：def test capability roundtrip ⋮---- cap = AgentCapability out = cap.to dict back = AgentCapability.from dict out ⋮---- def test capability defensive on non dict ⋮---- cap = AgentCapability.from dict "a2a-dm" ⋮---- def test endpoint roundtrip ⋮---- ep = AgentEndpoint back = AgentEndpoint.from dict ep.to dict ⋮---- def test authentication roundtrip ⋮---- auth = AgentAuthentication out = auth.to dict ⋮---- back = AgentAuthentication.from dict out ⋮---- def test card parses platform shape ⋮---- platform = { card = AgentCard.from dict platform ⋮---- dm = card.endpoint by kind "dm" ⋮---- def test card parses empty input ⋮---- card = AgentCard.from dict {} ⋮---- def test card parses non dict input… 证据：`sdk/tests/test_agent_card.py`
- **Test Triage**（source_file）：def friend payload memory=None, overrides - dict ⋮---- base = { ⋮---- def envelope , task id="msg 1", sender="bestiedog", recipient="me" - TaskEnvelope ⋮---- env = TaskEnvelope ⋮---- def make concrete daemon client, kwargs - BaseDaemon ⋮---- class TestDaemon BaseDaemon ⋮---- def run loop self ⋮---- def test triage policy defaults ⋮---- p = TriagePolicy ⋮---- def test triage policy custom ⋮---- p = TriagePolicy max turns per partner=5, count on replies=False ⋮---- def test triage decision fields ⋮---- d = TriageDecision ⋮---- @responses.activate def test turn counter check under cap returns ok ⋮---- client = AgentClient token="bt test", bot id="me" tc = TurnCounter client, TriagePolicy max t… 证据：`sdk/tests/test_triage.py`
- 其余 1 条证据见 `AI_CONTEXT_PACK.json` 或 `EVIDENCE_INDEX.json`。

## 宿主 AI 必须遵守的规则

- **把本资产当作开工前上下文，而不是运行环境。**：AI Context Pack 只包含证据化项目理解，不包含目标项目的可执行状态。 证据：`README.md`, `hermes/README.md`, `landing/README.md`
- **回答用户时区分可预览内容与必须安装后才能验证的内容。**：安装前体验的消费者价值来自降低误装和误判，而不是伪装成真实运行。 证据：`README.md`, `hermes/README.md`, `landing/README.md`

## 用户开工前应该回答的问题

- 你准备在哪个宿主 AI 或本地环境中使用它？
- 你只是想先体验工作流，还是准备真实安装？
- 你最在意的是安装成本、输出质量、还是和现有规则的冲突？

## 验收标准

- 所有能力声明都能回指到 evidence_refs 中的文件路径。
- AI_CONTEXT_PACK.md 没有把预览包装成真实运行。
- 用户能在 3 分钟内看懂适合谁、能做什么、如何开始和风险边界。

---

## Doramagic Context Augmentation

下面内容用于强化 Repomix/AI Context Pack 主体。Human Manual 只提供阅读骨架；踩坑日志会被转成宿主 AI 必须遵守的工作约束。

## Human Manual 骨架

使用规则：这里只是项目阅读路线和显著性信号，不是事实权威。具体事实仍必须回到 repo evidence / Claim Graph。

宿主 AI 硬性规则：
- 不得把页标题、章节顺序、摘要或 importance 当作项目事实证据。
- 解释 Human Manual 骨架时，必须明确说它只是阅读路线/显著性信号。
- 能力、安装、兼容性、运行状态和风险判断必须引用 repo evidence、source path 或 Claim Graph。

- **项目概览**：importance `high`
  - source_paths: README.md, landing/README.md, landing/im.tsx
- **系统架构**：importance `high`
  - source_paths: README.md, sdk/a2a_dm/client.py, mcp/a2a_dm_mcp/server.py, hermes/a2a_dm_hermes/runtime.py
- **Python SDK 核心 API**：importance `high`
  - source_paths: sdk/a2a_dm/__init__.py, sdk/a2a_dm/client.py, sdk/a2a_dm/dm.py, sdk/a2a_dm/models.py, sdk/a2a_dm/_http.py
- **守护进程框架与接收器层级**：importance `high`
  - source_paths: sdk/a2a_dm/daemon/__init__.py, sdk/a2a_dm/daemon/_base.py, sdk/a2a_dm/daemon/_dedup.py, sdk/a2a_dm/daemon/_inbox.py, sdk/a2a_dm/daemon/_sse.py
- **唤醒上下文与跨会话记忆**：importance `high`
  - source_paths: sdk/a2a_dm/wake_context.py, sdk/a2a_dm/daemon/advanced/_wake_mode.py, sdk/a2a_dm/skill.py, sdk/a2a_dm/models.py, docs/WAKE_ROADMAP.md
- **Agent Card 与代理发现**：importance `medium`
  - source_paths: sdk/a2a_dm/agent_card.py, sdk/a2a_dm/agent_card_api.py, sdk/a2a_dm/agents_api.py, sdk/a2a_dm/skill.py, README.md
- **MCP 服务器与聊天驱动集成**：importance `high`
  - source_paths: mcp/a2a_dm_mcp/__init__.py, mcp/a2a_dm_mcp/__main__.py, mcp/a2a_dm_mcp/server.py, mcp/README.md, mcp/pyproject.toml
- **Hermes Agent 插件**：importance `medium`
  - source_paths: hermes/README.md, hermes/a2a_dm_hermes/__init__.py, hermes/a2a_dm_hermes/autowake.py, hermes/a2a_dm_hermes/delivery.py, hermes/a2a_dm_hermes/gatewaycfg.py

## Repo Inspection Evidence / 源码检查证据

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `b3c2a1406c7d2cf316c55b2106146264a8f3314e`
- inspected_files: `README.md`, `docs/GROUP_CHAT_v0.10.md`, `docs/INTEGRATIONS.md`, `docs/WAKE_ROADMAP.md`

宿主 AI 硬性规则：
- 没有 repo_clone_verified=true 时，不得声称已经读过源码。
- 没有 repo_inspection_verified=true 时，不得把 README/docs/package 文件判断写成事实。
- 没有 quick_start_verified=true 时，不得声称 Quick Start 已跑通。

## Doramagic Pitfall Constraints / 踩坑约束

这些规则来自 Doramagic 发现、验证或编译过程中的项目专属坑点。宿主 AI 必须把它们当作工作约束，而不是普通说明文字。

### Constraint 1: 可能修改宿主 AI 配置

- Trigger: 项目面向 Claude/Cursor/Codex/Gemini/OpenCode 等宿主，或安装命令涉及用户配置目录。
- Host AI rule: 列出会写入的配置文件、目录和卸载/回滚步骤。
- Why it matters: 安装可能改变本机 AI 工具行为，用户需要知道写入位置和回滚方法。
- Evidence: capability.host_targets | https://github.com/shichuanqiong/a2a-dm | host_targets=mcp_host, hermes, claude_code, claude, cursor
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 2: 能力判断依赖假设

- Trigger: README/documentation is current enough for a first validation pass.
- Host AI rule: 将假设转成下游验证清单。
- Why it matters: 假设不成立时，用户拿不到承诺的能力。
- Evidence: capability.assumptions | https://github.com/shichuanqiong/a2a-dm | README/documentation is current enough for a first validation pass.
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 3: 维护活跃度未知

- Trigger: 未记录 last_activity_observed。
- Host AI rule: 补 GitHub 最近 commit、release、issue/PR 响应信号。
- Why it matters: 新项目、停更项目和活跃项目会被混在一起，推荐信任度下降。
- Evidence: evidence.maintainer_signals | https://github.com/shichuanqiong/a2a-dm | last_activity_observed missing
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

- Trigger: no_demo
- Evidence: downstream_validation.risk_items | https://github.com/shichuanqiong/a2a-dm | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 5: 存在评分风险

- Trigger: no_demo
- Why it matters: 风险会影响是否适合普通用户安装。
- Evidence: risks.scoring_risks | https://github.com/shichuanqiong/a2a-dm | no_demo; severity=medium
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 6: issue/PR 响应质量未知

- Trigger: issue_or_pr_quality=unknown。
- Host AI rule: 抽样最近 issue/PR，判断是否长期无人处理。
- Why it matters: 用户无法判断遇到问题后是否有人维护。
- Evidence: evidence.maintainer_signals | https://github.com/shichuanqiong/a2a-dm | issue_or_pr_quality=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。

### Constraint 7: 发布节奏不明确

- Trigger: release_recency=unknown。
- Host AI rule: 确认最近 release/tag 和 README 安装命令是否一致。
- Why it matters: 安装命令和文档可能落后于代码，用户踩坑概率升高。
- Evidence: evidence.maintainer_signals | https://github.com/shichuanqiong/a2a-dm | release_recency=unknown
- Hard boundary: 不要把这个坑点包装成已解决、已验证或可忽略，除非后续验证证据明确证明它已经关闭。
