# memforge - 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 memforge 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

- **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

- **多宿主安装与分发**（需要安装后验证）：项目包含插件或 marketplace 配置，说明它面向一个或多个 AI 宿主的安装和分发。 证据：`public/ai-plugin.json` Claim：`clm_0001` supported 0.86
- **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：`CLAUDE.md`, `README.md` Claim：`clm_0002` supported 0.86

## 怎么开始

- `pip install memforge` 证据：`README.md` Claim：`clm_0004` supported 0.86
- `npm install @salishforge/memforge` 证据：`README.md` Claim：`clm_0005` supported 0.86
- `npm install -g @salishforge/memforge` 证据：`README.md` Claim：`clm_0006` supported 0.86
- `git clone https://github.com/salishforge/memforge.git` 证据：`README.md` Claim：`clm_0007` supported 0.86
- `curl -X POST http://localhost:3333/memory/agent-1/add \` 证据：`README.md` Claim：`clm_0008` supported 0.86
- `curl "http://localhost:3333/memory/agent-1/query?q=user+preferences&mode=hybrid" \` 证据：`README.md` Claim：`clm_0009` supported 0.86
- `curl -X POST http://localhost:3333/memory/agent-1/sleep \` 证据：`README.md` Claim：`clm_0010` supported 0.86
- `curl -X POST http://localhost:3333/memory/agent-1/active-recall \` 证据：`README.md` Claim：`clm_0011` supported 0.86
- `curl "http://localhost:3333/memory/agent-1/query?q=user+preferences" \` 证据：`CLAUDE.md` Claim：`clm_0009` supported 0.86, `clm_0012` 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**：角色质量和任务匹配不能直接相信。
- **继续会触碰**：角色选择偏差、命令执行、宿主 AI 配置

### 现在可以相信

- **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：多宿主安装与分发**（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：`public/ai-plugin.json` Claim：`clm_0001` 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：`CLAUDE.md`, `README.md` Claim：`clm_0002` 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

### 现在还不能相信

- **角色质量和任务匹配不能直接相信。**（unverified）：角色库证明有很多角色，不证明每个角色都适合你的具体任务，也不证明角色能产生高质量结果。
- **不能把角色文案当成真实执行能力。**（unverified）：安装前只能判断角色描述和任务画像是否匹配，不能证明它能在宿主 AI 里完成任务。
- **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。 证据：`CLAUDE.md`, `public/ai-plugin.json`
- **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。 证据：`public/ai-plugin.json`
- **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。

### 继续会触碰什么

- **角色选择偏差**：用户对任务应该由哪个专家角色处理的判断。 原因：选错角色会让 AI 从错误专业视角回答，浪费时间或误导决策。
- **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：`CLAUDE.md`, `README.md`
- **宿主 AI 配置**：Claude/Codex/Cursor/Gemini/OpenCode 等宿主的 plugin、Skill 或规则加载配置。 原因：宿主配置会改变 AI 后续工作方式，可能和用户已有规则冲突。 证据：`CLAUDE.md`, `public/ai-plugin.json`
- **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：`CLAUDE.md`, `README.md`, `public/ai-plugin.json`
- **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：`ARCHITECTURE.md`, `BENCHMARK_IMPLEMENTATION.md`, `CHANGELOG.md`, `CLAUDE.md` 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**：先用交互式试用验证任务画像和角色匹配，不要先导入整套角色库。（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。）
- **先备份宿主 AI 配置**：Skill、plugin、规则文件可能改变 Claude/Cursor/Codex 的默认行为。（适用：存在插件 manifest、Skill 或宿主规则入口时。）
- **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。
- **准备移除宿主 plugin / Skill / 规则入口**：如果试装后行为异常，可以把宿主 AI 恢复到试装前状态。
- **保留原始角色选择记录**：如果输出偏题，可以回到任务画像阶段重新选择角色，而不是继续沿着错误角色推进。
- **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_0013` inferred 0.45
- **宿主 AI 插件或 Skill 规则冲突**：新规则可能改变用户现有宿主 AI 的工作方式。 处理方式：安装前先检查插件 manifest 和 Skill 文件，必要时隔离测试。 证据：`public/ai-plugin.json` Claim：`clm_0014` supported 0.86
- **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：`CLAUDE.md`, `README.md` Claim：`clm_0015` 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

- **多宿主安装与分发**：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：`public/ai-plugin.json` Claim：`clm_0001` supported 0.86
- **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：`CLAUDE.md`, `README.md` Claim：`clm_0002` supported 0.86

### 上下文规模

- 文件总数：136
- 重要文件覆盖：40/136
- 证据索引条目：79
- 角色 / Skill 条目：27

### 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 memforge 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 memforge 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 memforge 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 27 个角色 / Skill / 项目文档条目。

- **MemForge**（project_doc）：! License: MIT https://img.shields.io/badge/License-MIT-blue.svg LICENSE ! npm version https://img.shields.io/npm/v/@salishforge/memforge.svg?label=npm&color=cb3837 https://www.npmjs.com/package/@salishforge/memforge ! npm downloads https://img.shields.io/npm/dm/@salishforge/memforge.svg?color=blue https://www.npmjs.com/package/@salishforge/memforge ! Status: Beta https://img.shields.io/badge/Status-Beta-yellow.svg… 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`
- **MemForge Benchmarks**（project_doc）：Retrieval quality and performance benchmarks for the MemForge memory system. 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：`benchmarks/README.md`
- **MemForge Python SDK**（project_doc）：Python client for MemForge https://github.com/salishforge/memforge — neuroscience-inspired memory for AI agents. 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：`python/README.md`
- **LongMemEval QA Accuracy Benchmark**（project_doc）：This benchmark runs the full LongMemEval pipeline : retrieve → generate answer → LLM judge. 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：`benchmarks/longmemeval-qa/README.md`
- **MemForge for Claude Desktop**（project_doc）：Persistent long-term memory for Claude Desktop via MCP. 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：`plugins/claude-desktop/README.md`
- **MemForge for Microsoft 365 Copilot**（project_doc）：Persistent long-term memory for M365 Copilot agents via API Plugin or Power Automate. 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：`plugins/m365-copilot/README.md`
- **MemForge — Project Instructions for AI Agents**（project_doc）：MemForge — Project Instructions for AI Agents 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：`CLAUDE.md`
- **Contributing to MemForge**（project_doc）：Thank you for your interest in contributing to MemForge. This document explains how to get started, what we look for in contributions, and how the project is organized. 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：`CONTRIBUTING.md`
- **MemForge — Architecture**（project_doc）：MemForge is a neuroscience-inspired memory system for AI agents. This document describes the internal architecture, data models, and processing pipelines. 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：`ARCHITECTURE.md`
- **MemForge — Memory Architecture Specification**（project_doc）：MemForge — Memory Architecture Specification 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：`SPECIFICATION.md`
- **MemForge Adversarial Security Assessment**（project_doc）：MemForge Adversarial Security Assessment 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：`ADVERSARIAL-ASSESSMENT.md`
- **Benchmark Implementation Progress**（project_doc）：Last updated: 2026-07-27 Status: Phase 1 P0 credibility fixes complete, QA accuracy harness implemented 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：`BENCHMARK_IMPLEMENTATION.md`
- **Changelog**（project_doc）：All notable changes to MemForge are documented here. 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：`CHANGELOG.md`
- **MemForge × Claude Dreaming — Feature Review & Integration Plan**（project_doc）：MemForge × Claude Dreaming — Feature Review & Integration Plan 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：`CLAUDE_DREAMING_PLAN.md`
- **Contributor Covenant Code of Conduct**（project_doc）：Contributor Covenant Code of Conduct 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：`CODE_OF_CONDUCT.md`
- **Development Guide**（project_doc）：- Node.js = 22 - PostgreSQL 16+ with extensions: - pgvector 0.5+ — vector similarity search; 0.5+ required for halfvec float16 storage - pg trgm — trigram fuzzy matching used for entity dedup and search fallback - Redis 7+ optional — MemForge works without it, just slower - Git 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：`DEVELOPMENT.md`
- **Git Commit Commands for Benchmark Plan**（project_doc）：Git Commit Commands for Benchmark Plan 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：`GIT_COMMIT_COMMANDS.md`
- **MemForge Security Hardening Plan**（project_doc）：Based on the Adversarial Security Assessment ADVERSARIAL-ASSESSMENT.md conducted 2026-04-08. 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：`HARDENING-PLAN.md`
- **Integrating MemForge with AI Agents**（project_doc）：Integrating MemForge with AI Agents 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：`INTEGRATION.md`
- **MemForge — Competitive Analysis & Next-Gen Recommendations**（project_doc）：MemForge — Competitive Analysis & Next-Gen Recommendations 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：`NEXTGEN-RECOMMENDATIONS.md`
- **Phase 5: Autonomous Knowledge Architecture — Plan & Funding Brief**（project_doc）：Phase 5: Autonomous Knowledge Architecture — Plan & Funding Brief 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：`PHASE_5_PLAN.md`
- **MemForge Roadmap**（project_doc）：AI agents today are stateless by default. Every conversation starts from zero. The few systems that add memory treat it as a passive store — save facts, retrieve facts, hope for the best. 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：`ROADMAP.md`
- **Security Policy**（project_doc）：Version Supported --------- -------------------- 2.1.x Yes 2.0.x Security fixes only < 2.0 No 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：`SECURITY.md`
- **MemForge Threat Model**（project_doc）：A comprehensive security analysis of MemForge's architecture, identifying attack vectors, bypass techniques, and integrity risks. This document is intended for security reviewers, contributors, and operators evaluating MemForge for production use. 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：`THREAT_MODEL.md`
- **MemForge Benchmark Results**（project_doc）：⚠️ Partial run — not an official result 42 of 500 questions stratified sample . Reported for development iteration only. Publishable figures require a full 500-question run; cite nothing from this page until then. 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：`benchmarks/RESULTS.md`
- **MemForge Tools for Claude**（project_doc）：This document describes all memory tools available to Claude when MemForge is connected via MCP. Use these tools to build persistent memory across conversations — remembering user preferences, tracking project context, learning from mistakes, and improving over time. 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：`plugins/claude-desktop/TOOLS.md`
- **MemForge Tools for Microsoft 365 Copilot**（project_doc）：MemForge Tools for Microsoft 365 Copilot 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：`plugins/m365-copilot/TOOLS.md`

## 证据索引

- 共索引 79 条证据。

- **MemForge**（documentation）：! License: MIT https://img.shields.io/badge/License-MIT-blue.svg LICENSE ! npm version https://img.shields.io/npm/v/@salishforge/memforge.svg?label=npm&color=cb3837 https://www.npmjs.com/package/@salishforge/memforge ! npm downloads https://img.shields.io/npm/dm/@salishforge/memforge.svg?color=blue https://www.npmjs.com/package/@salishforge/memforge ! Status: Beta https://img.shields.io/badge/Status-Beta-yellow.svg project-status ! Node.js https://img.shields.io/badge/Node.js-%3E%3D22-green.svg https://nodejs.org ! TypeScript https://img.shields.io/badge/TypeScript-5.7-blue.svg https://www.typescriptlang.org ! PostgreSQL https://img.shields.io/badge/PostgreSQL-16+-336791.svg https://www.pos… 证据：`README.md`
- **MemForge Benchmarks**（documentation）：Retrieval quality and performance benchmarks for the MemForge memory system. 证据：`benchmarks/README.md`
- **MemForge Python SDK**（documentation）：Python client for MemForge https://github.com/salishforge/memforge — neuroscience-inspired memory for AI agents. 证据：`python/README.md`
- **LongMemEval QA Accuracy Benchmark**（documentation）：This benchmark runs the full LongMemEval pipeline : retrieve → generate answer → LLM judge. 证据：`benchmarks/longmemeval-qa/README.md`
- **MemForge for Claude Desktop**（documentation）：Persistent long-term memory for Claude Desktop via MCP. 证据：`plugins/claude-desktop/README.md`
- **MemForge for Microsoft 365 Copilot**（documentation）：Persistent long-term memory for M365 Copilot agents via API Plugin or Power Automate. 证据：`plugins/m365-copilot/README.md`
- **Package**（package_manifest）：{ "name": "@salishforge/memforge", "version": "3.7.0", "description": "Neuroscience-inspired agent memory — sleep cycles, memory revision, knowledge graph, procedural learning, vector search, and MCP integration", "type": "module", "main": "./dist/index.js", "types": "./dist/index.d.ts", "exports": { ".": { "import": "./dist/index.js", "types": "./dist/index.d.ts" }, "./client": { "import": "./dist/client.js", "types": "./dist/client.d.ts" }, "./tools": { "import": "./dist/tool-definitions.js", "types": "./dist/tool-definitions.d.ts" } }, "bin": { "memforge": "./dist/server.js", "memforge-mcp": "./dist/mcp.js" }, "scripts": { "build": "tsc", "start": "node dist/server.js", "dev": "tsx watch… 证据：`package.json`
- **MemForge — Project Instructions for AI Agents**（documentation）：MemForge — Project Instructions for AI Agents 证据：`CLAUDE.md`
- **Contributing to MemForge**（documentation）：Thank you for your interest in contributing to MemForge. This document explains how to get started, what we look for in contributions, and how the project is organized. 证据：`CONTRIBUTING.md`
- **License**（source_file）：Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files the "Software" , to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: 证据：`LICENSE`
- **MemForge — Architecture**（documentation）：MemForge is a neuroscience-inspired memory system for AI agents. This document describes the internal architecture, data models, and processing pipelines. 证据：`ARCHITECTURE.md`
- **MemForge — Memory Architecture Specification**（documentation）：MemForge — Memory Architecture Specification 证据：`SPECIFICATION.md`
- **Schema**（source_file）：CREATE EXTENSION IF NOT EXISTS vector; CREATE EXTENSION IF NOT EXISTS pg trgm; CREATE TABLE IF NOT EXISTS agents id TEXT PRIMARY KEY, created at TIMESTAMPTZ NOT NULL DEFAULT now , last seen TIMESTAMPTZ NOT NULL DEFAULT now , metadata JSONB NOT NULL DEFAULT '{}', last sleep cycle TIMESTAMPTZ, sleep cycle cost tokens BIGINT NOT NULL DEFAULT 0, scoring weights JSONB ; CREATE TABLE IF NOT EXISTS hot tier id BIGSERIAL PRIMARY KEY, agent id TEXT NOT NULL REFERENCES agents id ON DELETE CASCADE, content TEXT NOT NULL, metadata JSONB NOT NULL DEFAULT '{}', created at TIMESTAMPTZ NOT NULL DEFAULT now , content hash TEXT, namespace TEXT NOT NULL DEFAULT 'default', session id TEXT NOT NULL DEFAULT 'def… 证据：`schema/schema.sql`
- **Classifier**（source_file）：export type Sensitivity = 'public' 'internal' 'confidential' 'restricted'; ⋮---- type RedactionAction = 'none' 'tag' 'redact' 'reject'; ⋮---- export interface Classification { type: string; sensitivity: Sensitivity; match: string; offset: number; confidence: number; action: RedactionAction; } ⋮---- export interface ClassificationResult { sensitivity: Sensitivity; findings: Classification ; redactedContent: string; wasRedacted: boolean; } ⋮---- export interface ContentClassifier { readonly name: string; readonly description: string; classify content: string : Classification ; } ⋮---- classify content: string : Classification ; ⋮---- function maxSensitivity a: Sensitivity, b: Sensitivity : Se… 证据：`src/classifier.ts`
- **Client**（source_file）：import type { AddResult, QueryResult, QueryMode, TimelineEntry, ConsolidateResult, ConsolidationMode, ClearResult, AgentStats, EntitySearchResult, GraphQueryResult, ReflectionResult, Reflection, ReflectionTrigger, Procedure, SleepCycleResult, MemoryHealth, FeedbackOutcome, FeedbackResult, MetaReflectionResult, ActiveMemoryResult, ResumeContext, EntityDeduplicationResult, HealthStatus, ColdTierSearchResult, RestoreColdTierResult, SleepAdvisory, SharedProcedure, ExpertiseResult, AgentRole, DriftReport, ProcedureOutcome, DreamRun, DreamStatus, DreamSource, DreamOutputMode, AnthropicMemoryStoreLink, AnthropicSyncState, SyncStrategy, CausalChainNode, PredictionResult, Abstraction, AbstractionLev… 证据：`src/client.ts`
- **Db**（source_file）：import { Pool } from 'pg'; import type { PoolConfig } from 'pg'; import { getLogger } from './logger.js'; ⋮---- export function getPool databaseUrl?: string : Pool ⋮---- export async function getVectorCast pool: Pool : Promise ⋮---- export async function closePool : Promise 证据：`src/db.ts`
- **Embedding**（source_file）：import { validateProviderUrl } from './schemas.js'; ⋮---- export interface EmbeddingProvider { embed text: string : Promise ; embedBatch texts: string : Promise ; readonly dimensions: number; readonly modelId: string; } ⋮---- embed text: string : Promise ; ⋮---- embedBatch texts: string : Promise ; ⋮---- interface OpenAIEmbeddingConfig { baseUrl?: string; apiKey?: string; model?: string; dimensions?: number; } ⋮---- export class OpenAIEmbeddingProvider implements EmbeddingProvider ⋮---- constructor config: OpenAIEmbeddingConfig = ⋮---- async embed text: string : Promise ⋮---- async embedBatch texts: string : Promise ⋮---- interface OllamaEmbeddingConfig { baseUrl?: string; model?: string; d… 证据：`src/embedding.ts`
- **Mcp**（source_file）：import { randomUUID } from 'crypto'; import { MemForgeClient } from './client.js'; import { VERSION } from './version.js'; import type { JsonSchemaProperty } from './types.js'; ⋮---- interface MCPRequest { jsonrpc: '2.0'; id: number string; method: string; params?: Record ; } ⋮---- interface MCPResponse { jsonrpc: '2.0'; id: number string; result?: unknown; error?: { code: number; message: string }; } ⋮---- interface MCPToolDefinition { name: string; description: string; inputSchema: { type: 'object'; properties: Record ; required?: string ; }; } ⋮---- function validateToolArgs name: string, args: Record : void ⋮---- async function executeTool client: MemForgeClient, name: string, args: Rec… 证据：`src/mcp.ts`
- **Memory Manager**（source_file）：import { createHash } from 'crypto'; import { Pool } from 'pg'; import { getPool, getVectorCast } from './db.js'; import { emitWebhookEvent } from './webhooks.js'; import { NoOpEmbeddingProvider } from './embedding.js'; import type { EmbeddingProvider } from './embedding.js'; import { REFLECTION SYSTEM PROMPT, PROCEDURE EXTRACTION PROMPT, wrapUserContent } from './llm.js'; import type { LLMProvider, ConsolidationSummary } from './llm.js'; import { safeParseLLMResponse, ReflectionResponseSchema, ProcedureExtractionSchema, NamespaceSchema, SessionIdSchema } from './schemas.js'; import { getConfig } from './config.js'; import { SleepCycleEngine, DreamCancellationError } from './sleep-cycle.js'… 证据：`src/memory-manager.ts`
- **Schemas**（source_file）：import { z } from 'zod'; import { getLogger } from './logger.js'; ⋮---- export type ValidatedOAuthIntrospect = z.infer ; ⋮---- // ─── Helper: Parse and validate LLM response ─────────────────────────────── ⋮---- / Strips markdown fences from LLM output, parses JSON, and validates against the provided Zod schema. Throws a descriptive error on failure. / export function safeParseLLMResponse schema: z.ZodType , raw: string : T ⋮---- function isPrivateHost hostname: string : boolean ⋮---- export function validateProviderUrl url: string, providerName: string, allowLocal = false : string 证据：`src/schemas.ts`
- **Server**（source_file）：import { MemoryManager } from './memory-manager.js'; import { createEmbeddingProvider } from './embedding.js'; import { createLLMProvider } from './llm.js'; import { closePool, getPool } from './db.js'; import { closeRedis } from './cache.js'; import { createDefaultRegistry } from './classifier.js'; import { wrapLLMProvider } from './llm-safety.js'; import { AuditChain } from './audit.js'; import { createApp } from './app.js'; import { DreamRunsWorker } from './dream-runs.js'; import { getLogger } from './logger.js'; import { configureWebhooks } from './webhooks.js'; import type { ConsolidationMode } from './types.js'; ⋮---- async function shutdown signal: string : Promise 证据：`src/server.ts`
- **Sleep Cycle**（source_file）：import type { Pool } from 'pg'; import { getVectorCast } from './db.js'; import { wrapUserContent } from './llm.js'; import type { LLMProvider } from './llm.js'; import { safeParseLLMResponse, RevisionResponseSchema, PrincipleExtractionSchema } from './schemas.js'; import type { EmbeddingProvider } from './embedding.js'; import type { SleepCycleConfig, SleepCycleResult, RevisionType, SharedPoolSleepCycleResult } from './types.js'; import type { AuditChain } from './audit.js'; import { getLogger } from './logger.js'; ⋮---- export class DreamCancellationError extends Error ⋮---- constructor public readonly runId: string ⋮---- export class SleepCycleEngine ⋮---- constructor pool: Pool, llm: LL… 证据：`src/sleep-cycle.ts`
- **Tool Definitions**（source_file）：import type { JsonSchemaProperty } from './types.js'; ⋮---- export interface ToolDefinition { name: string; description: string; input schema: { type: 'object'; properties: Record ; required?: string ; }; } ⋮---- export function toOpenAITools : Array< 证据：`src/tool-definitions.ts`
- **Client**（source_file）：class MemForgeError Exception ⋮---- def init self, message: str, status code: int = 0 ⋮---- class MemForgeClient ⋮---- headers: dict str, str = {"Content-Type": "application/json"} ⋮---- async def close self - None ⋮---- async def aenter self - "MemForgeClient" ⋮---- async def aexit self, args: Any - None ⋮---- async def get self, path: str, params: dict str, Any None = None - Any ⋮---- resp = await self. client.get path, params={k: v for k, v in params or {} .items if v is not None} data = resp.json ⋮---- async def post self, path: str, body: dict str, Any None = None - Any ⋮---- resp = await self. client.post path, json=body or {} ⋮---- body: dict str, Any = {"content": content} ⋮---- raw… 证据：`python/memforge/client.py`
- **Prefer summary if available more token-efficient**（source_file）：class ConversationMemory ⋮---- async def close self - None ⋮---- async def aenter self - "ConversationMemory" ⋮---- async def aexit self, args: Any - None ⋮---- tagged content = f" {role} : {content}" ⋮---- results = await self. client.query ⋮---- lines = ⋮---- Prefer summary if available more token-efficient text = r.summary if r.summary else r.content ⋮---- async def start session self, limit: int = 5 - str ⋮---- ctx = await self. client.resume self.agent id, limit ⋮---- hours = ctx.time since last activity ms / 3 600 000 ⋮---- async def end session self - None 证据：`python/memforge/conversation.py`
- **Resilient**（source_file）：logger = logging.getLogger "memforge" ⋮---- class ResilientMemForgeClient ⋮---- async def close self - None ⋮---- async def aenter self - "ResilientMemForgeClient" ⋮---- async def aexit self, args: Any - None ⋮---- def handle self, err: Exception - None ⋮---- async def add self, agent id: str, content: str, kwargs: Any - AddResult None ⋮---- async def query self, agent id: str, kwargs: Any - list QueryResult ⋮---- async def timeline self, agent id: str, kwargs: Any - list dict str, Any ⋮---- async def consolidate self, agent id: str, mode: str None = None, namespace: str None = None - ConsolidateResult None ⋮---- async def clear self, agent id: str - ClearResult None ⋮---- async def stats s… 证据：`python/memforge/resilient.py`
- **Types**（source_file）：T = TypeVar "T" ⋮---- def from response cls: type T , raw: Mapping str, Any - T ⋮---- known = {f.name for f in dataclasses.fields cls } ⋮---- @dataclass class AddResult ⋮---- id: int agent id: str created at: str ⋮---- deduplicated: bool = False ⋮---- @dataclass class QueryResult ⋮---- content: str metadata: dict str, Any consolidated at: str time start: Optional str time end: Optional str rank: float ⋮---- summary: Optional str = None ⋮---- context signals: dict str, Any = field default factory=dict ⋮---- epistemic status: Optional str = None ⋮---- evidence count: Optional int = None ⋮---- explanation: Optional list dict str, Any = None ⋮---- @dataclass class ConsolidateResult ⋮---- run id… 证据：`python/memforge/types.py`
- **MemForge Adversarial Security Assessment**（documentation）：MemForge Adversarial Security Assessment 证据：`ADVERSARIAL-ASSESSMENT.md`
- **Benchmark Implementation Progress**（documentation）：Last updated: 2026-07-27 Status: Phase 1 P0 credibility fixes complete, QA accuracy harness implemented 证据：`BENCHMARK_IMPLEMENTATION.md`
- **Changelog**（documentation）：All notable changes to MemForge are documented here. 证据：`CHANGELOG.md`
- **MemForge × Claude Dreaming — Feature Review & Integration Plan**（documentation）：MemForge × Claude Dreaming — Feature Review & Integration Plan 证据：`CLAUDE_DREAMING_PLAN.md`
- **Contributor Covenant Code of Conduct**（documentation）：Contributor Covenant Code of Conduct 证据：`CODE_OF_CONDUCT.md`
- **Development Guide**（documentation）：- Node.js = 22 - PostgreSQL 16+ with extensions: - pgvector 0.5+ — vector similarity search; 0.5+ required for halfvec float16 storage - pg trgm — trigram fuzzy matching used for entity dedup and search fallback - Redis 7+ optional — MemForge works without it, just slower - Git 证据：`DEVELOPMENT.md`
- **Git Commit Commands for Benchmark Plan**（documentation）：Git Commit Commands for Benchmark Plan 证据：`GIT_COMMIT_COMMANDS.md`
- **MemForge Security Hardening Plan**（documentation）：Based on the Adversarial Security Assessment ADVERSARIAL-ASSESSMENT.md conducted 2026-04-08. 证据：`HARDENING-PLAN.md`
- **Integrating MemForge with AI Agents**（documentation）：Integrating MemForge with AI Agents 证据：`INTEGRATION.md`
- **MemForge — Competitive Analysis & Next-Gen Recommendations**（documentation）：MemForge — Competitive Analysis & Next-Gen Recommendations 证据：`NEXTGEN-RECOMMENDATIONS.md`
- **Phase 5: Autonomous Knowledge Architecture — Plan & Funding Brief**（documentation）：Phase 5: Autonomous Knowledge Architecture — Plan & Funding Brief 证据：`PHASE_5_PLAN.md`
- **MemForge Roadmap**（documentation）：AI agents today are stateless by default. Every conversation starts from zero. The few systems that add memory treat it as a passive store — save facts, retrieve facts, hope for the best. 证据：`ROADMAP.md`
- **Security Policy**（documentation）：Version Supported --------- -------------------- 2.1.x Yes 2.0.x Security fixes only < 2.0 No 证据：`SECURITY.md`
- **MemForge Threat Model**（documentation）：A comprehensive security analysis of MemForge's architecture, identifying attack vectors, bypass techniques, and integrity risks. This document is intended for security reviewers, contributors, and operators evaluating MemForge for production use. 证据：`THREAT_MODEL.md`
- **MemForge Benchmark Results**（documentation）：⚠️ Partial run — not an official result 42 of 500 questions stratified sample . Reported for development iteration only. Publishable figures require a full 500-question run; cite nothing from this page until then. 证据：`benchmarks/RESULTS.md`
- **MemForge Tools for Claude**（documentation）：This document describes all memory tools available to Claude when MemForge is connected via MCP. Use these tools to build persistent memory across conversations — remembering user preferences, tracking project context, learning from mistakes, and improving over time. 证据：`plugins/claude-desktop/TOOLS.md`
- **MemForge Tools for Microsoft 365 Copilot**（documentation）：MemForge Tools for Microsoft 365 Copilot 证据：`plugins/m365-copilot/TOOLS.md`
- **Settings.Local**（structured_config）：{ "permissions": { "allow": "Bash grep -E \"\\\\.\\ ts js json yaml yml md\\ $\" ", "Bash gh api: ", "WebFetch domain:github.com ", "WebSearch", "mcp claude ai GitHub - Salishforge get file contents", "mcp claude ai GitHub - Salishforge search repositories", "Bash npx tsc: ", "Bash npm run: ", "Bash npm install: ", "Bash ./node modules/.bin/tsc --noEmit ", "Bash ./node modules/.bin/tsc ", "Bash wc -l src/ .ts schema/ .sql ", "Bash wc -l src/ .ts " } } 证据：`.claude/settings.local.json`
- **Tsconfig**（structured_config）：{ "extends": "../tsconfig.json", "compilerOptions": { "rootDir": "..", "noEmit": true }, "include": " / .ts", "../src/ / .ts" } 证据：`benchmarks/tsconfig.json`
- **Funding**（structured_config）：{ "$schema": "https://fundingjson.org/schema/v1.1.0/funding.schema.json", "entity": { "type": "individual", "role": "maintainer", "name": "John Brooke", "email": "john@salishforge.com", "description": "Solo maintainer building an open-source, MIT-licensed stack for long-running, multi-agent AI systems. The thesis is that frontier model labs are rightly focused on raw capability, but LLMs need a substrate that accumulates experience and revises beliefs over time to operate on human timelines. I'm building that substrate \u2014 security-first, zero-trust from day one \u2014 out of pocket on a modest VPS, with two autonomous agents running in production on the stack. Shipping weekly. Based in… 证据：`funding.json`
- **Glama**（structured_config）：{ "$schema": "https://glama.ai/mcp/schemas/server.json", "maintainers": "salishforge" } 证据：`glama.json`
- **Ai Plugin**（structured_config）：{ "schema version": "v1", "name for human": "MemForge Memory", "name for model": "memforge", "description for human": "Long-term memory for AI agents — store, search, and manage knowledge", "description for model": "MemForge provides persistent memory for AI agents. Use memforge tools to store memories, search past knowledge, and manage an agent's accumulated experience.", "auth": { "type": "service http", "authorization type": "bearer" }, "api": { "type": "openapi", "url": "/api/spec.json" } } 证据：`public/ai-plugin.json`
- **Server**（structured_config）：{ "$schema": "https://static.modelcontextprotocol.io/schemas/2025-12-11/server.schema.json", "name": "io.github.salishforge/memforge-mcp", "title": "MemForge", "description": "Neuroscience-inspired memory system for AI agents. Tiered storage hot/warm/cold , hybrid retrieval FTS + pgvector , knowledge graph, and a 10-phase sleep cycle that actively revises and improves stored knowledge during idle periods. 17 MCP tools.", "version": "3.0.0", "repository": { "url": "https://github.com/salishforge/memforge", "source": "github" }, "packages": { "registryType": "npm", "registryBaseUrl": "https://registry.npmjs.org", "identifier": "memforge-mcp", "version": "3.0.0", "runtimeHint": "npx", "transpo… 证据：`server.json`
- **Tsconfig**（structured_config）：{ "compilerOptions": { "target": "ES2022", "module": "NodeNext", "moduleResolution": "NodeNext", "outDir": "./dist", "rootDir": "./src", "declaration": true, "declarationMap": true, "sourceMap": true, "strict": true, "noUncheckedIndexedAccess": true, "noImplicitReturns": true, "noFallthroughCasesInSwitch": true, "verbatimModuleSyntax": true, "esModuleInterop": true, "skipLibCheck": true }, "include": "src/ / " , "exclude": "node modules", "dist" } 证据：`tsconfig.json`
- **Claude Desktop Config**（structured_config）：{ "mcpServers": { "memforge": { "command": "npx", "args": "memforge-mcp" , "env": { "MEMFORGE URL": "http://localhost:3333", "MEMFORGE TOKEN": "" } } } } 证据：`plugins/claude-desktop/claude_desktop_config.json`
- **Manifest**（structured_config）：{ "$schema": "https://developer.microsoft.com/json-schemas/copilot/plugin/v2.2/schema.json", "schema version": "v2.2", "name for human": "MemForge Memory", "description for human": "Long-term memory for AI agents — store, search, and manage knowledge that persists across sessions", "description for model": "MemForge provides persistent memory for AI agents. Use these tools to: store new memories facts, decisions, preferences, errors , search past knowledge with semantic understanding, get relevant context before tasks, and manage knowledge quality through consolidation and reflection. Memory persists across sessions and improves over time through active knowledge management.", "logo url": "… 证据：`plugins/m365-copilot/manifest.json`
- **Meeting Context**（structured_config）：{ "name": "MemForge: Pre-Meeting Context", "description": "5 minutes before a calendar event, retrieve relevant memories and send a Teams notification", "trigger": { "type": "Office365Outlook.CalendarEventSoon", "inputs": { "minutesBefore": 5 } }, "actions": { "name": "SearchMemories", "type": "Http", "inputs": { "method": "GET", "uri": "@{parameters 'MemForgeUrl' }/memory/@{parameters 'AgentId' }/query?q=@{encodeUriComponent triggerOutputs ? 'body/subject' }&mode=hybrid&max tokens=2000&limit=5", "headers": { "Authorization": "Bearer @{parameters 'MemForgeToken' }" } } }, { "name": "SendTeamsCard", "type": "MicrosoftTeams.PostMessageToConversation", "runAfter": "SearchMemories", "inputs": {… 证据：`plugins/power-automate/meeting-context.json`
- **Nightly Consolidation**（structured_config）：{ "name": "MemForge: Nightly Consolidation & Sleep Cycle", "description": "Run consolidation and sleep cycle at 2:00 AM daily to maintain memory quality", "trigger": { "type": "Recurrence", "inputs": { "frequency": "Day", "interval": 1, "startTime": "2026-01-01T02:00:00Z", "timeZone": "UTC" } }, "actions": { "name": "Consolidate", "type": "Http", "inputs": { "method": "POST", "uri": "@{parameters 'MemForgeUrl' }/memory/@{parameters 'AgentId' }/consolidate", "headers": { "Authorization": "Bearer @{parameters 'MemForgeToken' }", "Content-Type": "application/json" }, "body": {} } }, { "name": "SleepCycle", "type": "Http", "runAfter": "Consolidate", "inputs": { "method": "POST", "uri": "@{param… 证据：`plugins/power-automate/nightly-consolidation.json`
- **Store Emails**（structured_config）：{ "name": "MemForge: Store Important Emails", "description": "When an email tagged 'Important' arrives, store it as a memory in MemForge", "trigger": { "type": "Office365Outlook.OnNewEmail", "inputs": { "importance": "High", "includeAttachments": false } }, "actions": { "type": "Http", "inputs": { "method": "POST", "uri": "@{parameters 'MemForgeUrl' }/memory/@{parameters 'AgentId' }/add", "headers": { "Authorization": "Bearer @{parameters 'MemForgeToken' }", "Content-Type": "application/json" }, "body": { "content": "Email from @{triggerOutputs ? 'body/from' } — Subject: @{triggerOutputs ? 'body/subject' } — @{substring triggerOutputs ? 'body/bodyPreview' , 0, min length triggerOutputs ? 'b… 证据：`plugins/power-automate/store-emails.json`
- **.dockerignore**（source_file）：node modules dist .git .github .claude .env .env.local .env.production .env !.env.example !.env.docker tests coverage .md !README.md .log .DS Store Thumbs.db .editorconfig .nvmrc .eslintrc.json tsconfig.json 证据：`.dockerignore`
- **.editorconfig**（source_file）：indent style = space indent size = 2 end of line = lf charset = utf-8 trim trailing whitespace = true insert final newline = true 证据：`.editorconfig`
- **Docker Compose environment — copy to .env and set POSTGRES PASSWORD**（source_file）：Docker Compose environment — copy to .env and set POSTGRES PASSWORD POSTGRES DB=memforge POSTGRES USER=memforge POSTGRES PASSWORD=changeme POSTGRES PORT=5432 REDIS PORT=6379 PORT=3333 证据：`.env.docker`
- **─── MemForge Standalone — Environment Configuration ────────────────────────**（source_file）：─── MemForge Standalone — Environment Configuration ──────────────────────── 证据：`.env.example`
- 其余 19 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`, `benchmarks/README.md`, `python/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`, `benchmarks/README.md`, `python/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。

- **MemForge 概述与系统架构**：importance `high`
  - source_paths: README.md, ARCHITECTURE.md, SPECIFICATION.md, src/server.ts, src/sleep-cycle.ts
- **核心记忆系统与数据流**：importance `high`
  - source_paths: src/memory-manager.ts, src/db.ts, src/schemas.ts, src/embedding.ts, src/classifier.ts
- **SDK、平台集成与部署方式**：importance `high`
  - source_paths: src/client.ts, src/mcp.ts, src/tool-definitions.ts, python/memforge/client.py, python/memforge/resilient.py
- **运维、安全与社区关注问题**：importance `high`
  - source_paths: src/cache.ts, src/auth.ts, src/audit.ts, src/llm-safety.ts, src/webhooks.ts

## Repo Inspection Evidence / 源码检查证据

- repo_clone_verified: true
- repo_inspection_verified: true
- repo_commit: `756ff7faea127640e1c820060ac406595285e95d`
- inspected_files: `Dockerfile`, `README.md`, `docker-compose.yml`, `package.json`, `examples/claude_tools.py`, `examples/langchain_memory.py`, `examples/openai_tools.py`, `examples/quickstart.py`, `examples/quickstart.ts`, `examples/simple_chatbot.py`, `src/app.ts`, `src/audit.ts`, `src/auth.ts`, `src/cache.ts`, `src/classifier.ts`, `src/client.ts`, `src/config.ts`, `src/dashboard.ts`, `src/db.ts`, `src/dream-runs.ts`

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: 失败模式：installation: v3.0.0-beta.3

- Trigger: Developers should check this installation risk before relying on the project: v3.0.0-beta.3
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: v3.0.0-beta.3. Context: Observed when using python
- Why it matters: Upgrade or migration may change expected behavior: v3.0.0-beta.3
- Evidence: failure_mode_cluster:github_release | https://github.com/salishforge/memforge/releases/tag/v3.0.0-beta.3 | v3.0.0-beta.3
- 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: 失败模式：installation: v3.0.0-beta.4

- Trigger: Developers should check this installation risk before relying on the project: v3.0.0-beta.4
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: v3.0.0-beta.4. Context: Observed when using node, python
- Why it matters: Upgrade or migration may change expected behavior: v3.0.0-beta.4
- Evidence: failure_mode_cluster:github_release | https://github.com/salishforge/memforge/releases/tag/v3.0.0-beta.4 | v3.0.0-beta.4
- 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: 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/salishforge/memforge | host_targets=mcp_host, claude
- 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 4: 失败模式：configuration: MemForge v0.1.0-alpha

- Trigger: Developers should check this configuration risk before relying on the project: MemForge v0.1.0-alpha
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: MemForge v0.1.0-alpha. Context: Observed when using docker
- Why it matters: Upgrade or migration may change expected behavior: MemForge v0.1.0-alpha
- Evidence: failure_mode_cluster:github_release | https://github.com/salishforge/memforge/releases/tag/v0.1.0-alpha | MemForge v0.1.0-alpha
- 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: 失败模式：configuration: v3.0.0-beta.2 — CI Green, Shared Memory, Full Test Coverage

- Trigger: Developers should check this configuration risk before relying on the project: v3.0.0-beta.2 — CI Green, Shared Memory, Full Test Coverage
- Host AI rule: Before packaging this project, run the relevant install/config/quickstart check for: v3.0.0-beta.2 — CI Green, Shared Memory, Full Test Coverage. Context: Observed when using node, python, docker
- Why it matters: Upgrade or migration may change expected behavior: v3.0.0-beta.2 — CI Green, Shared Memory, Full Test Coverage
- Evidence: failure_mode_cluster:github_release | https://github.com/salishforge/memforge/releases/tag/v3.0.0-beta.2 | v3.0.0-beta.2 — CI Green, Shared Memory, Full Test Coverage
- 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: 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/salishforge/memforge | 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 7: 来源证据：Python SDK: QueryResult dataclass rejects unknown response keys (breaks on v3.8+ fields)

- Trigger: GitHub 社区证据显示该项目存在一个运行相关的待验证问题：Python SDK: QueryResult dataclass rejects unknown response keys (breaks on v3.8+ fields)
- Host AI rule: 来源显示可能已有修复、规避或版本变化，说明书中必须标注适用版本。
- Why it matters: 可能增加新用户试用和生产接入成本。
- Evidence: community_evidence:github | https://github.com/salishforge/memforge/issues/161 | 来源讨论提到 python 相关条件，需在安装/试用前复核。
- 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 8: 来源证据：REST: /entities and /graph return 500 for invalid agent ids; OpenAPI query path missing epistemic param

- Trigger: GitHub 社区证据显示该项目存在一个运行相关的待验证问题：REST: /entities and /graph return 500 for invalid agent ids; OpenAPI query path missing epistemic param
- Host AI rule: 来源显示可能已有修复、规避或版本变化，说明书中必须标注适用版本。
- Why it matters: 可能增加新用户试用和生产接入成本。
- Evidence: community_evidence:github | https://github.com/salishforge/memforge/issues/162 | 来源类型 github_issue 暴露的待验证使用条件。
- 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 9: 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/salishforge/memforge | 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/salishforge/memforge | 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。
