Initial iSphere AI bridge prompts

.gitignore

@@ -0,0 +1,20 @@
+evidence/*
+!evidence/.gitkeep
+runs/*
+!runs/.gitkeep
+*.log
+*.pcap
+*.pcapng
+*.har
+*.db
+*.sqlite
+*.sqlite3
+*.ldb
+*.bak
+*.tmp
+node_modules/
+bin/
+obj/
+dist/
+__pycache__/
+.venv/

README.md

@@ -0,0 +1,39 @@
+# iSphere AI Bridge
+
+This project is for building a safe message-handoff gateway between iSphere/IMPlatformClient and AI digital employees.
+
+The first goal is not full automation. The first goal is to let AI reliably see authorized messages, normalize them, generate reply drafts, and keep evidence. Sending messages is a later step and must stay behind human approval until explicitly changed.
+
+## Scope
+
+- Authorized company communication software only.
+- Read messages first; do not modify the original client state in the first POC.
+- Prefer stable entry points in this order:
+  1. Local data/log/cache discovery.
+  2. Plugin or existing bridge API.
+  3. UI automation/RPA wrapper.
+  4. Network/API discovery.
+  5. Binary reverse engineering only when needed to explain or validate the above.
+
+## Existing Source Material
+
+- Previous iSphere work: `E:\coding\codex\isphere`
+- Installer: `E:\coding\codex\isphere\iSphere_4.1.2.6841.exe`
+- Existing bridge artifacts: `OpenClawBridge`, `MockCloud`, `XmppBridgeService`
+- Existing MCP spec: `E:\coding\codex\isphere\docs\MCP接口需求规格.md`
+
+## Key Directories
+
+- `docs/` - design notes and installed skill map.
+- `prompts/` - ready-to-use Codex prompts for each investigation phase.
+- `evidence/` - local evidence output; ignored by Git except `.gitkeep`.
+- `runs/` - run logs and temporary investigation outputs; ignored by Git except `.gitkeep`.
+- `src/` - future implementation code.
+
+## Safety Baseline
+
+- Do not bypass login, licensing, MFA, access control, TLS protections, or endpoint security.
+- Do not read non-authorized account data.
+- Do not send real messages without explicit human approval.
+- Do not modify production databases, logs, config files, registry keys, or binaries during discovery.
+- Redact secrets, tokens, private message content, and personal data in reports.

docs/design.md

@@ -0,0 +1,63 @@
+# iSphere AI Bridge Design
+
+## Objective
+
+Make iSphere messages usable by AI digital employees without rebuilding the full communication platform.
+
+## First POC
+
+The first POC reads recent authorized messages, converts them into a normalized JSON model, generates draft replies, and records evidence. It does not automatically send messages.
+
+## Architecture
+
+```text
+iSphere / MockCloud / existing bridge
+  -> source connector
+  -> message adapter
+  -> AI draft worker
+  -> human review
+  -> evidence log
+```
+
+## Source Connectors
+
+1. Local data connector: reads local databases, logs, or caches in read-only mode.
+2. Existing bridge connector: talks to `OpenClawBridge` or `XmppBridgeService` when available.
+3. UI/RPA connector: wraps the desktop client with limited actions.
+4. Network connector: wraps discovered HTTP/SOAP/XMPP/WebSocket APIs when validated.
+
+## Normalized Message Model
+
+```json
+{
+  "message_id": "string",
+  "conversation_id": "string",
+  "conversation_name": "string",
+  "conversation_type": "direct|group|system",
+  "sender_id": "string",
+  "sender_name": "string",
+  "content_type": "text|image|file|rich|notify",
+  "content_text": "string",
+  "attachments": [],
+  "created_at": "ISO-8601 or Unix ms",
+  "received_at": "ISO-8601 or Unix ms",
+  "source": "local-db|log|plugin|xmpp|http|rpa|mock",
+  "raw_ref": "path/table/packet/window reference"
+}
+```
+
+## Approval Policy
+
+- Read access can be automated in test scope.
+- Draft generation can be automated.
+- Sending messages requires human approval in the first implementation.
+- Every send action must log target, content hash, timestamp, approver, and connector.
+
+## Verification
+
+The project is successful when it can:
+
+1. Find at least one stable authorized message source.
+2. Extract recent messages into normalized JSON.
+3. Generate a reply draft.
+4. Produce an evidence report with exact paths, commands, and limits.

docs/installed-skills.md

@@ -0,0 +1,46 @@
+# Installed Skills
+
+## Official Curated Skills Installed
+
+- `aspnet-core`
+- `cli-creator`
+- `security-best-practices`
+- `security-threat-model`
+- `playwright-interactive`
+
+Codex must be restarted before newly installed skills are available in a fresh skills list.
+
+## Project-Specific Skills Created
+
+General reverse engineering:
+
+- `binary-triage`
+- `ghidra-static-analysis`
+- `decompiler-cleanup`
+- `android-apk-analysis`
+- `frida-observation-plan`
+- `legacy-code-reverse-doc`
+- `reverse-report`
+
+iSphere / communication handoff:
+
+- `comm-app-inventory`
+- `local-data-discovery`
+- `ui-automation-feasibility`
+- `network-protocol-discovery`
+- `rpa-wrapper-builder`
+- `message-adapter-design`
+- `reverse-evidence-report`
+
+## Suggested Order For This Project
+
+1. `comm-app-inventory`
+2. `local-data-discovery`
+3. `ui-automation-feasibility`
+4. `message-adapter-design`
+5. `reverse-evidence-report`
+6. `network-protocol-discovery`
+7. `binary-triage`
+8. `ghidra-static-analysis`
+9. `decompiler-cleanup`
+10. `rpa-wrapper-builder`

evidence/.gitkeep

@@ -0,0 +1 @@
+

prompts/00-master-prompt.md

@@ -0,0 +1,31 @@
+# Master Prompt
+
+Use this prompt to start a new focused Codex thread.
+
+```text
+目标：建设 `E:\coding\codex\isphere-ai-bridge`，让 AI 数字员工安全接管 iSphere/IMPlatformClient 消息。
+
+请先使用相关 skill：
+- `comm-app-inventory`
+- `local-data-discovery`
+- `ui-automation-feasibility`
+- `message-adapter-design`
+- `reverse-evidence-report`
+
+已有材料：
+- 旧项目：`E:\coding\codex\isphere`
+- 安装包：`E:\coding\codex\isphere\iSphere_4.1.2.6841.exe`
+- 已有桥接：`OpenClawBridge`、`MockCloud`、`XmppBridgeService`
+- MCP 规格：`E:\coding\codex\isphere\docs\MCP接口需求规格.md`
+- 新项目：`E:\coding\codex\isphere-ai-bridge`
+
+边界：
+- 只分析授权软件和授权账号。
+- 第一阶段只读消息、生成草稿，不自动发送。
+- 不绕过登录、授权、MFA、TLS、终端安全或许可控制。
+- 不修改生产客户端数据库、配置、注册表、二进制文件。
+- 证据报告要写清楚路径、命令、结果、置信度和未确认项。
+- 遵守工作区 AGENTS.md：不要使用 rg；优先 git ls-files、ag、grep -R。
+
+请先做入口摸底，判断最适合第一版 POC 的消息读取路径：本地数据、现有桥接、UI 自动化、网络接口，还是二进制逆向。
+```

prompts/01-inventory.md

@@ -0,0 +1,21 @@
+# Prompt: Communication App Inventory
+
+```text
+请使用 `comm-app-inventory` 对 iSphere/IMPlatformClient 做只读资产摸底。
+
+重点检查：
+- `E:\coding\codex\isphere`
+- `E:\coding\codex\isphere\iSphere_4.1.2.6841.exe`
+- 当前 Windows 安装项、快捷方式、进程、服务、计划任务
+- `Documents\isphere`
+- `%LOCALAPPDATA%`、`%APPDATA%`、`%PROGRAMDATA%`
+- `Temp\importal`、`Temp\opencode`
+- 插件目录、日志目录、配置目录、本地端口
+
+不要运行未知 exe，不要修改任何文件。输出：
+1. 已确认路径
+2. 运行时进程/端口
+3. 数据/日志/配置候选
+4. 插件/API/RPA 入口候选
+5. 下一步建议
+```

prompts/02-local-data-discovery.md

@@ -0,0 +1,13 @@
+# Prompt: Local Data Discovery
+
+```text
+请使用 `local-data-discovery`，基于入口摸底结果查找 iSphere 可只读读取的消息、联系人、群组、附件、日志和缓存数据。
+
+要求：
+- 只读，不写库，不改状态，不标记已读。
+- 优先 SQLite、LevelDB、XML、JSON、日志。
+- 如果遇到加密日志，只报告证据、格式和可能的解密类，不做未授权绕过。
+- 输出 3-5 条可脱敏的标准化 JSON 样例；如果没有可读消息，说明阻塞点。
+
+目标：判断第一版数字员工能否不通过发送接口就“看见消息”。
+```

prompts/03-ui-automation-feasibility.md

@@ -0,0 +1,19 @@
+# Prompt: UI Automation Feasibility
+
+```text
+请使用 `ui-automation-feasibility` 评估 iSphere 客户端是否能被 RPA/UI 自动化接管。
+
+重点：
+- 判断 UI 技术栈：Win32、WinForms、WPF、Qt、Electron、WebView、Java、自绘。
+- 检查消息列表、会话搜索、输入框、发送按钮、附件按钮能否稳定识别。
+- 优先 FlaUI / pywinauto；如果是 WebView 再考虑 Playwright。
+- 不发送真实消息，不输入账号密码，不改设置。
+
+输出：
+1. 控件树可读性
+2. 读取消息可行性
+3. 写草稿/发送可行性
+4. 推荐工具
+5. 稳定性评分 A/B/C/D
+6. 最小 RPA wrapper 动作清单
+```

prompts/04-network-protocol-discovery.md

@@ -0,0 +1,18 @@
+# Prompt: Network Protocol Discovery
+
+```text
+请使用 `network-protocol-discovery`，在用户明确批准抓包或代理前，只做只读网络线索整理。
+
+先检查：
+- iSphere 配置里的服务器、端口、协议名
+- 现有 `MockCloud` / `XmppBridgeService` / `OpenClawBridge` 代码里的 SOAP、XMPP、HTTP 端点
+- 本地监听端口和进程对应关系
+- 日志中出现的接口、域名、JID、XMPP、SOAP、WebSocket 线索
+
+不要绕过 TLS、登录、授权或终端安全。输出：
+1. 观察到的协议和端口
+2. 候选接口
+3. 消息收发路径证据
+4. 是否适合包装成内部 API
+5. 需要用户批准的下一步抓包/测试动作
+```

prompts/05-message-adapter-poc.md

@@ -0,0 +1,27 @@
+# Prompt: Message Adapter POC
+
+```text
+请使用 `message-adapter-design`，为 `E:\coding\codex\isphere-ai-bridge` 设计第一版消息适配器。
+
+输入来源要能兼容：
+- 本地数据读取
+- OpenClawBridge HTTP API
+- XmppBridgeService HTTP API
+- RPA wrapper
+- MockCloud 演示环境
+
+第一版只实现：
+- list_messages
+- watch_messages 或 poll_messages
+- generate_draft 的输入输出契约
+- archive_evidence
+
+不要实现自动发送。发送动作只保留接口草案 `send_after_approval`。
+
+输出：
+1. TypeScript 或 C# 数据模型建议
+2. Connector 接口
+3. Evidence log 结构
+4. 最小目录结构
+5. 第一轮测试用例
+```

prompts/06-rpa-wrapper.md

@@ -0,0 +1,28 @@
+# Prompt: RPA Wrapper Builder
+
+```text
+请使用 `rpa-wrapper-builder`，基于 UI 自动化评估结果设计 iSphere 的受控 RPA 包装层。
+
+目标不是让 AI 自由操作鼠标键盘，而是封装有限动作：
+- read_latest_messages(limit)
+- search_contact(query)
+- open_conversation(id_or_name)
+- write_draft(text)
+- send_after_approval(approval_id)
+- upload_file(path)
+- mark_processed(message_id)
+
+要求：
+- 发送真实消息必须人工确认。
+- 每个动作要有前置校验、结果校验和日志。
+- 优先控件 ID / automation id / role，不优先坐标。
+- 坐标和 OCR 只能作为最后兜底。
+
+输出：
+1. wrapper action contract
+2. selector strategy
+3. approval gates
+4. failure recovery
+5. audit log fields
+6. POC 实现计划
+```

prompts/07-reverse-deep-dive.md

@@ -0,0 +1,27 @@
+# Prompt: Reverse Deep Dive
+
+```text
+只有在本地数据、现有桥接、UI 自动化、网络接口都不足以支撑 POC 时，才进入这个阶段。
+
+请按顺序使用：
+- `binary-triage`
+- `ghidra-static-analysis`
+- `decompiler-cleanup`
+- `reverse-report`
+
+目标不是破解客户端，而是解释：
+- 消息数据库/日志路径
+- 配置加载逻辑
+- 插件加载逻辑
+- XMPP/SOAP/HTTP 接口线索
+- 消息收发函数
+- 文件上传下载路径
+
+禁止：
+- 绕过登录或授权
+- patch 授权逻辑
+- 生成隐藏注入或规避检测脚本
+- 修改生产二进制
+
+输出必须包含文件路径、hash、函数名、地址、字符串交叉引用、置信度和下一步验证方法。
+```

runs/.gitkeep

@@ -0,0 +1 @@
+