Apache-2.0、零依赖的 Git 故障经验记忆库（290 条课程），通过专属 MCP 适配器接入 DeepSeek Harness，提供 deepseek.recovery.* 故障恢复工具。
DSH 适配
兼容
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
Apache-2.0
01它能帮你完成什么？
Give DeepSeek Harness a shared failure-memory so it searches verified debugging lessons instead of re-debugging known errors
BM25 keyword matches across 290 indexed lessons (problem → root cause → fix → verify), served through the DeepSeekHarness MCP adapter as `deepseek.recovery.*` tools
DeepSeekHarness users who want harness-level failure recovery without a server, database or extra dependencies
Search 290 verified failure-recovery lessons locally from the command line
Zero-dependency BM25 keyword search results over Markdown lessons, runnable with Python stdlib only after `git clone`
Developers and agents debugging real failures (DCO, pip, token, MCP, encoding, CI) who want an offline-first lesson lookup
Feed unmatched failures back into the network without a GitHub account or email
Redacted intake submitted via the `misakanet_submit_intake` remote MCP tool, becoming a maintainer-visible GitHub issue labeled `intake`, `mcp-intake`, `pending-review`
Agents whose task stalls on a failure with no matching lesson; no GitHub account, email, Bearer token or browser required
02如何接入 DeepSeek Harness？
前置条件
- Python 3.10+ (badge: python-3.10+) — the core engine is zero-dep, pure Python stdlib
- git — lessons ship as Markdown files in the repository
安装步骤
- 01
Clone the repo: `git clone https://github.com/Ikalus1988/MisakaNet.git && cd MisakaNet`
$ git clone https://github.com/Ikalus1988/MisakaNet.git && cd MisakaNet
- 02
For DeepSeekHarness, run the recovery adapter: `python3 scripts/mcp_deepseek_adapter.py`
- 03
Alternative: run the standard MCP server `python3 scripts/mcp_server.py` and add it to your MCP config
- 04
Optional local search: `python3 search_knowledge.py "your error here"`
验证接入成功
- Run the smoke check: `python3 scripts/misakanet_cli.py smoke` — the README's 'Contribute in 3 minutes' flow uses it to verify it works
03DSH 适配与能力边界
MCP-compatible adapter (`python3 scripts/mcp_deepseek_adapter.py`) exposing `deepseek.recovery.*` tools for harness-level failure recovery
DeepSeekHarness recovery adapter
harness-level failure events from DeepSeekHarness→`deepseek.recovery.*` MCP tools for failure recovery inside the harness
runs a local MCP adapter process (`python3 scripts/mcp_deepseek_adapter.py`)Zero-dependency BM25 lesson search
an error message or query string→matching failure-recovery lessons (problem → root cause → fix → verify) via pure-Python BM25 retrieval
MCP server for any MCP client
MCP config of Claude / Codex / local agents→lesson search and intake tools, e.g. ask: "Search MisakaNet for pip install timeout"
runs a local MCP server process (`python3 scripts/mcp_server.py`)Remote MCP intake (no account)
a redacted failure description→`misakanet_submit_intake` call to `https://misakanet.org/mcp` creating a maintainer-visible GitHub issue for review
network call to the remote endpoint https://misakanet.org/mcpcreates a GitHub issue labeled intake / mcp-intake / pending-review; submissions must be redacted (no secrets, no raw logs)
04适合谁？何时不该用？
适合
- DeepSeekHarness users who want harness-level failure recovery without a server, database or extra dependencies
- Developers and agents debugging real failures (DCO, pip, token, MCP, encoding, CI) who want an offline-first lesson lookup
- Agents whose task stalls on a failure with no matching lesson; no GitHub account, email, Bearer token or browser required
不适合
- Lessons are community-contributed, so retrieved commands must be reviewed before execution — the README tells users to always sandbox their agent before running retrieved commands.
05兼容性、维护与安全提示
- Lessons are community-contributed, so retrieved commands must be reviewed before execution — the README tells users to always sandbox their agent before running retrieved commands.
- MisakaNet is purpose-built for failure-recovery knowledge: it is not a general-purpose memory system, not an agent runtime, and not a vector database or RAG system — BM25 keyword search only.
- Local clone + search is fully offline, but the remote MCP endpoint (https://misakanet.org/mcp) and lesson intake require network access; intake submissions are reviewed by maintainers before publication.
Apache-2.0 · actively maintained (latest release v2.17.1, 2026-08-16; last push 2026-08-19)
06常见问题
MisakaNet 如何接入 DeepSeek Harness？
通过专属 DeepSeekHarness MCP 适配器：克隆仓库后运行 `python3 scripts/mcp_deepseek_adapter.py`。这是一个 MCP 兼容适配器，暴露 `deepseek.recovery.*` 工具用于 harness 级故障恢复。完整的安装、验证与降级策略见 docs/integration/deepseek-harness.md。
它是 dsh 原生集成还是 MCP 工具？
以 MCP 工具集形式集成，而非原生运行时。标准 MCP 服务器（`python3 scripts/mcp_server.py`）同样适用于 Claude、Codex 等本地 agent；也可以直接使用托管端点 https://misakanet.org/mcp，无需克隆仓库。
使用前需要安装什么？
Python 3.10+ 和 git。核心引擎零依赖——纯 Python 标准库，无向量数据库、无服务器、无数据库。可选扩展通过 `pip install misakanet[semantic|hub|feishu]` 安装。
数据会上传吗？
本地克隆和搜索完全离线：`git clone` + `python3 search_knowledge.py`，不泄露提示词、不存储原始日志。只有可选的远程 MCP 提交会访问 https://misakanet.org/mcp，且提交内容必须脱敏，由维护者审核后才可能转为课程。
有什么限制？
MisakaNet 只做故障恢复知识这一件事——不是通用记忆系统、不是 agent 运行时、也不是 RAG 系统。课程由社区贡献，执行其中检索到的命令前务必自行审查，并让 agent 在沙箱中运行。
07相关的 DSH 工作流
archify
作者 tt-a1i
为编码智能体生成美观可验证的架构图、时序图与数据流图，输出自包含 HTML，支持动效与清晰导出。
openviking
作者 volcengine
为 AI 智能体打造的自进化上下文数据库，统一智能体记忆、知识 RAG 与技能。
nocobase
作者 nocobase
开源的 AI + 无代码应用搭建平台：AI 在久经生产验证的底层基础设施与所见即所得的无代码界面之上工作，帮你快速构建 CRM、ERP 等业务系统，兼顾速度与可靠性。
learn-harness-engineering
作者 walkinglabs
Harness 工程新手教程，从 0 到 1 系统学习智能体工作流框架。
08数据与来源
**DeepSeekHarness Adapter** | MCP-compatible adapter exposing `deepseek.recovery.*` tools for harness-level failure reco…
| DeepSeekHarness | `python3 scripts/mcp_deepseek_adapter.py` |
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。