Apache-2.0 协议的推理时认知控制技能套件，将 j-space/ 安装进 DeepSeek Harness 的 Skills 目录并用 verify_suite.py 校验，即可获得按需加载的深度推理、验证与恢复控制。
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？
Add deep reasoning, verification and recovery control to DeepSeek Harness tasks
A managed inference-time workspace (one entry, nine selectively loaded modules, three references) that routes fast / full / loop passes over deep reasoning, long-horizon work, tool use, verification, and recovery
Developers running DeepSeek Harness or any Skill-capable coding agent who want stronger reasoning and self-verification on complex tasks
Keep durable state across multi-stage, long-horizon agent tasks
A `.jspace/` ledger in the task workspace with goal / next-action notes, hub entries, checkpoints, questions, seams and resume, managed by the optional standard-library controller
Teams running multi-stage workflows (multiple files, turns, tools) where context must survive long gaps and interruptions
Verify that a Skill package installs intact before using it
An integrity check result from `verify_suite.py` run with any Python 3 interpreter, confirming SKILL.md and its relative module / reference / script paths are intact
Anyone installing the suite into a host's Skills directory who wants a quick post-install sanity check
02如何接入 DeepSeek Harness？
前置条件
- An AI host with a user-level Skills directory (or a chat / API environment where SKILL.md can be provided as a system- or developer-level instruction)
- A Python 3 interpreter on the host (commonly `python`, `python3`, or `py -3`) for the integrity check and the optional controller
安装步骤
- 01
Download or clone this repository
- 02
Locate the user-level Skills directory used by your AI host
- 03
Copy the complete `j-space/` directory into it so that the installed entry is `<skills-directory>/j-space/SKILL.md`
- 04
Run the integrity check: `<python-command> <skills-directory>/j-space/scripts/verify_suite.py`
- 05
Reload the host if it discovers Skills at startup
验证接入成功
- Run `<python-command> <skills-directory>/j-space/scripts/verify_suite.py` — the README's integrity check for the installed package
03DSH 适配与能力边界
Model-agnostic Skill (SKILL.md) installed into the host's user-level Skills directory; DeepSeek Harness loads it as a Skill and its evaluations were configured from the official DeepSeek Harness minimal-mode setup
Selective workspace loading with fast / full / loop passes
any task request via the host's Skill mechanism (Skill picker, `/j-space`, `$j-space`, or a direct request)→the entry gate selects the lightest suitable pass: fast (nothing extra), full (one or two modules, ship before delivery), or loop (ledger, seams, checkpoints, register audit, recovery)
Optional loop controller for durable task state
`jspace.py` CLI commands (note, seam, ship, resume) run by resolved Skill path with the task workspace as current directory→externalized loop state under `.jspace/`: goals, next actions, hub entries, checkpoints, questions, and resume after long gaps
writes working state under the task's `.jspace/` directory (Python standard library only)Verification and recovery mechanisms
in-flight reasoning, stalled derivations, or outgoing deliverables→bridge-before-conclusion intermediates, bounded empirical tests with a named verifier and coverage, metacognitive routing of confidence / inconsistency / failure signals, and `ship` register-leakage audits
Cross-model, host-agnostic integration
`j-space/SKILL.md` — installed directly in hosts with a native Skill loader, or provided as a system- / developer-level instruction in chat / API environments→the same protocol (workspace loading, selective routing, state externalization, verification, recovery) across DeepSeek, Qwen, GLM, GPT, and Claude model families
04适合谁？何时不该用？
适合
- Developers running DeepSeek Harness or any Skill-capable coding agent who want stronger reasoning and self-verification on complex tasks
- Teams running multi-stage workflows (multiple files, turns, tools) where context must survive long gaps and interruptions
- Anyone installing the suite into a host's Skills directory who wants a quick post-install sanity check
不适合
- The installed `j-space/` directory must remain intact: SKILL.md routes to relative paths under `modules/`, `references/`, and `scripts/`, so partial copies or moved files break the Skill.
- Effect size varies with base capability, context policy, tool harness, sampling configuration, and benchmark implementation; reported benchmarks were collected in the project's own evaluation environment, so scores are not directly comparable across harnesses.
05兼容性、维护与安全提示
- The installed `j-space/` directory must remain intact: SKILL.md routes to relative paths under `modules/`, `references/`, and `scripts/`, so partial copies or moved files break the Skill.
- Integrity checks and the optional controller require a Python 3 interpreter on the host; hosts without one can install the Skill but cannot run `verify_suite.py` or `jspace.py`.
- Effect size varies with base capability, context policy, tool harness, sampling configuration, and benchmark implementation; reported benchmarks were collected in the project's own evaluation environment, so scores are not directly comparable across harnesses.
Apache-2.0 · actively maintained (latest release v3.6.1, 2026-08-19)
06常见问题
J-Space 如何接入 DeepSeek Harness？
它以 Skill 形式分发：将完整的 `j-space/` 目录复制到 AI 宿主的用户级 Skills 目录，使入口为 `<skills-directory>/j-space/SKILL.md`，再通过宿主的技能机制调用（技能选择器、`/j-space`、`$j-space` 或直接请求）。其 DeepSeek 评测正是参照官方 DeepSeek Harness minimal-mode 配置完成的。
安装需要什么前置条件？
需要一个带有用户级 Skills 目录的宿主，以及一个 Python 3 解释器（常见为 `python`、`python3` 或 `py -3`）。Python 用于完整性检查和可选的循环控制器，后者仅依赖标准库。
如何验证安装成功？
复制目录后运行 `<python-command> <skills-directory>/j-space/scripts/verify_suite.py`，若宿主在启动时发现技能则重新加载。目录必须保持完整，因为 SKILL.md 通过 `modules/`、`references/`、`scripts/` 下的相对路径路由。
可选控制器会往磁盘写数据吗？
可选的 `jspace.py` 控制器会把 loop 状态外化到当前任务工作区的 `.jspace/` 目录。它只记录和汇报状态，方案选择仍由模型决定，且只使用 Python 标准库。
如果宿主没有原生技能加载器怎么办？
可以把 `j-space/SKILL.md` 作为系统级或开发者级指令提供，并通过文件或检索工具暴露 `modules/` 与 `references/`，按需检索所选文件——选择性加载本身就是该套件的设计原则。
07相关的 DSH 工作流
distilly
作者 titanwings
Distilly — Distill how they think into reusable Skills for any Agent or Bot. Formerly Colleague Skill（原同事 Skill）.
dsh-market
作者 dsh-market
The plugin market inside DeepSeek Harness — browse, search, one-click install · DSH 可视化插件市场
ai_animation
作者 unclecheng-li
本项目整理了用于生成[炫酷 HTML 动画网页]的 AI Prompts，涵盖动画效果、3D 可视化、PPT 风格演示、UI 美化等多个类别。
aegis
作者 ganyuanran
让 AI 编码智能体具备架构意识：基线优先、证据验证、漂移检查，保障长任务安全。
08数据与来源
It is packaged as a Skill for cross-platform use, selective loading, and low-friction integration.
The J-Space evaluations on DeepSeek were configured with reference to the official DeepSeek Harness minimal-mode setup,…
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。