Skip to content

Releases

全能模式 / Omni mode v1.7.0

Choose a tag to compare

@github-actions github-actions released this 14 Nov 08:53

全能模式 / Omni mode v1.7.0(正式版 / current release)

中文 · English

中文

同一个「全能模式」、同一个 omni 预设 ID。本次加入领域专长与用代码制作图片视频的能力,不调用生图或视频模型。v1.7.0 由 v1.7.0-rc.1 预发布转正:功能范围与预发布一致,兼容门禁放宽到当前最新的 DSH 0.2.2-rc.3。

本版已可正常使用,但仍带着一条未修复的已知限制。 用字节未改动的原 40 题、同一模型、同一工具与评分器复测,旧版通过 38 题、新版通过 34 题,有 4 题从通过变为不通过(research-02、review-02、recover-03、recover-04)。复核发现其中多数差异来自题面未规定的封闭取值或类型(如状态枚举、字符串数组与对象数组、引用列表是否允许附加项),但"题面不完整"不等于已经证明质量没有下降。这条限制在修复并重新冻结复测前一直有效,详见执行复核。

# 正式版安装 / Install current release
dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.7.0

也可下载本 Release 的 TGZ 后在插件管理器添加。安装后重启引擎、新建对话,选择「全能模式」;使用你实际的 profile。旧本地包与本包同为 omni ID,不能同时启用。上一个稳定版 v1.5.0 仍可回退。

新增内容

  • 十个领域技能:工程、游戏、安全网络与逆向、文档演示、小说叙事、研究学习、陪伴个性化、助理、分子生物数据、业务语义。按需加载,不常驻。
  • 能力索引:100 套来源配置引用与 23 个能力条目,用 verified / needs-dependency / method-only / uncovered 四种状态如实标注。
  • 11 个本地操作:数据校验与转换、游戏数值初筛、小说连续性、FASTA 统计、实体关系一致性、引用覆盖、本地指标提取、Markdown 大纲、待办依赖、统一格式日历草稿。只用 Node 内置模块,不新增生产依赖。
  • 代码视觉:场景源码 → SVG → 明确版本的 resvg → FFmpeg 出 PNG/GIF/MP4。有授权素材优先用素材,否则从零绘制;视频音频需求未明确会先询问。依赖需显式准备,不会在激活或渲染时静默安装。

这次转正实际改了什么

  • 兼容门禁从 >=0.2.2-rc.1 <=0.2.2-rc.2 放宽为 >=0.2.2-rc.1 <=0.2.2-rc.3,把当前最新的 rc.3 纳入。
  • 门禁不再只是"测试版本落在范围内":scripts/verify-package.mjs 现在把受支持版本写成冻结清单,既要求范围恰好接受这三版,也反过来断言 0.2.1-rc.5、0.2.2-alpha.3、0.2.2-rc.4、0.2.2、0.2.3-rc.1 都被拒绝——避免以后把"范围写得宽松"当成"已经验证"。
  • 开发依赖与 lock 同步到 0.2.2-rc.3,Node 受支持下限仍为 >=24.9.0。
  • 更正文文档中一处与实际不符的计数:真实渲染测试实测为 23 项(此前正文写作 22 项,与 VERIFICATION.md 的 23 项冲突)。数字以本次实际运行为准。

验证到什么程度

已实际执行:

  • 三种宿主版本的隔离安装与原生集成:0.2.2-rc.1、0.2.2-rc.2、0.2.2-rc.3 各自独立安装后,均跑通完整预设激活、11 个技能与 omni-method、grilling 正文逐字加载、推理选档、自适应切换、手动压缩事务、服务隔离与独立进程恢复。三者均 result: PASS,serviceLeaks: 0。
  • 263 项仓库测试、64 项能力契约测试、23 项真实渲染测试(真实 resvg 2.6.2 + FFmpeg 6.1.1,0 跳过)。
  • 从解包后的安装包独立完成中文 SVG/PNG/GIF/MP4 端到端产出,并完整解码。
  • 真实 Harness 中确认 11 个技能、能力索引与两个运行器可从安装包加载,且同名外部诱饵不能冒充。
  • 用读图工具查看实际画面并修复了字体证明图的截断显示问题。

没有验证、也不声称的事:集成里的模型是本地确定性模拟适配器,没有调用真实付费供应商(无隔离凭据/路由可用)。桌面应用实机 UI、Windows 打包更新、其他平台与三维渲染均未验证。原 40 题的 4 个关键警报仍在,不宣称整体无退步。

你的实际使用反馈

你在本版预发布期间已用它连续跑了多个真实任务,并据此要求转正。这属于使用者反馈,记录了"在实际任务中可用"这一事实;它不是独立行为评测,也不能替代上面那条未通过的整体门槛。两者分开记,不互相覆盖。

quality.js、adaptive.js、pool.js 与上游 grilling 原文自 v1.5.0 起字节未变;推理选档、自适应保留、权限与审批边界均无改动。详细证据与未证事项见增强证据。

渲染支持范围为 Linux arm64 实测,其他平台与三维未验证。附件提供 TGZ 与 SHA256;未发布 npm,不覆盖历史标签(v1.7.0-rc.1 预发布记录保留)。

English

The same Omni preset and the same omni preset ID. This release adds domain specialties and the ability to make images and video with code, using no image or video generation models. v1.7.0 promotes v1.7.0-rc.1 to a regular release: the feature set is unchanged, and the compatibility gate now covers the latest DSH 0.2.2-rc.3.

This version is usable, but it still ships with one known limitation. Re-running the byte-identical original 40 cases on the same model, tools and grader gives 38 baseline passes versus 34 candidate passes, with four cases moving from pass to fail (research-02, review-02, recover-03, recover-04). Review found most of those differences stem from closed values or types the prompts never specified (status enums, string versus object arrays, whether a citation list may carry extra items), but an under-specified prompt is not proof that quality did not drop. That limitation stands until the contracts are fixed, re-frozen and re-measured; see the execution review.

# Install current release
dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.7.0

Or add the release TGZ through the plugin manager. Restart, create a new conversation and select 全能模式; use your actual profile. The legacy local bundle declares the same omni ID and cannot be enabled simultaneously. The previous stable v1.5.0 remains available for rollback.

What is new

  • Ten domain skills: engineering, game, security/network/RE, documents, fiction, research, personalization, assistant, molecular-biology data and business semantics. Loaded on demand, not resident.
  • Capability index: references to 100 source configurations and 23 capability entries, honestly labelled verified / needs-dependency / method-only / uncovered.
  • 11 local operations: data validation and transformation, preliminary game arithmetic, fiction continuity, FASTA statistics, entity-relation consistency, citation coverage, local indicator extraction, Markdown outlines, todo dependencies and a calendar draft in a fixed format. Node built-ins only; no new production dependency.
  • Code media: scene source → SVG → pinned resvg → FFmpeg for PNG/GIF/MP4. Authorized assets are reused when available, otherwise drawn from scratch; audio for video is asked first. Dependencies are prepared explicitly and never installed silently at activation or render time.

What actually changed in this promotion

  • The gate widens from >=0.2.2-rc.1 <=0.2.2-rc.2 to >=0.2.2-rc.1 <=0.2.2-rc.3, covering the current rc.3.
  • The gate is no longer merely "the tested version happens to fall inside the range": scripts/verify-package.mjs now holds a frozen census of supported versions, requiring the range to admit exactly these three and asserting that 0.2.1-rc.5, 0.2.2-alpha.3, 0.2.2-rc.4, 0.2.2 and 0.2.3-rc.1 are all refused — so a wider range can never be mistaken for verified support.
  • The development dependency and lock move to 0.2.2-rc.3; the supported Node floor stays >=24.9.0.
  • Corrects one documentation count that disagreed with reality: the real render suite is 23 tests, where the prose said 22 and VERIFICATION.md said 23. The measured run now decides.

How far this is verified

Actually executed:

  • Isolated installs and native integration on three host versions: 0.2.2-rc.1, 0.2.2-rc.2 and 0.2.2-rc.3 were each installed separately and each passed full preset activation, verbatim loading of all 11 skills plus omni-method and grilling, reasoning selection, adaptive switching, the manual compaction transaction, service isolation and resume in an independent process. All three returned result: PASS with serviceLeaks: 0.
  • 263 repository tests, 64 capability contract tests and 23 real render tests (real resvg 2.6.2 + FFmpeg 6.1.1, 0 skipped).
  • Independent end-to-end Chinese SVG/PNG/GIF/MP4 production from the extracted package, with full decoding.
  • In a real Harness, 11 skills, the capability index and both runners load from the installed package, and same-name external decoys cannot substitute for them.
  • Actual frames viewed with an image-reading tool, which caught and fixed a truncation display issue in the font proof sheet.

Not verified, and not claimed: the integration model is a deterministic local fixture; no real paid provider was called (no isolated credential or route was available). Desktop-app UI, the Windows packaged updater, other platforms and 3D rendering remain unverified. The four critical warnings from the original 40 cases stand, so overall non-regression is not claimed.

Your own usage feedback

You have already used this build for multiple real tasks during its pre-release period and asked for promotion on that basis. That is user-reported feedback: it records that the build was usable in real work, and it is neither an independent behavioral evaluation nor a substitute for the unmet overall gate above. The two are recorded separately and neither overrides the other.

quality.js, adaptive.js, pool.js and the upstream grilling text are byte-for-byte unchanged since v1.5.0; reasoning selection, adaptive retention and permission/approval boundaries are untouched. See enhancement evidence for details and unproven items.

Rendering is validated on Linux arm64 only; other platforms and 3D remain unverified. The release attaches a TGZ and SHA256; nothing is published to npm and no historical tag is overwritten (the v1.7.0-rc.1 pre-release record is preserved).

全能模式 / Omni mode v1.7.0-rc.1

Pre-release

Choose a tag to compare

@github-actions github-actions released this 14 Nov 05:20

全能模式 / Omni mode v1.7.0-rc.1(实验性预发布 / experimental pre-release)

中文 · English

中文

同一个「全能模式」、同一个 omni 预设 ID。本次加入领域专长与用代码制作图片视频的能力,不调用生图或视频模型。

这是预发布,不是稳定版。 稳定版仍是 v1.5.0;本版为 v1.7.0-rc.1。功能可用,但按既定标准,关键无退步未达成,因此不宣布验收通过。

未达成、必须如实告知的部分: 用字节未改动的原 40 题、同一模型 gpt-6.1-sol、同一工具与评分器复测,旧版通过 38 题、新版通过 34 题,有 4 题从通过变为不通过(research-02、review-02、recover-03、recover-04)。复核发现其中多数差异来自题面未规定的封闭取值或类型(如状态枚举、字符串数组与对象数组、引用列表是否允许附加项),但"题面不完整"不等于已经证明质量没有下降。这条限制在修复并重新冻结复测前一直有效,详见执行复核。

# 预发布安装 / Pre-release install
dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.7.0-rc.1

也可下载本 Release 的 TGZ 后在插件管理器添加。安装后重启引擎、新建对话,选择「全能模式」;使用你实际的 profile。旧本地包与本包同为 omni ID,不能同时启用。预发布可能继续改动,不希望承担变动风险时请继续使用稳定版 v1.5.0。

新增内容

  • 十个领域技能:工程、游戏、安全网络与逆向、文档演示、小说叙事、研究学习、陪伴个性化、助理、分子生物数据、业务语义。按需加载,不常驻。
  • 能力索引:100 套来源配置引用与 23 个能力条目,用 verified / needs-dependency / method-only / uncovered 四种状态如实标注。
  • 11 个本地操作:数据校验与转换、游戏数值初筛、小说连续性、FASTA 统计、实体关系一致性、引用覆盖、本地指标提取、Markdown 大纲、待办依赖、统一格式日历草稿。只用 Node 内置模块,不新增生产依赖。
  • 代码视觉:场景源码 → SVG → 明确版本的 resvg → FFmpeg 出 PNG/GIF/MP4。有授权素材优先用素材,否则从零绘制;视频音频需求未明确会先询问。依赖需显式准备,不会在激活或渲染时静默安装。

验证到什么程度

已实际执行:

  • 263 项仓库测试、64 项能力契约测试、22 项真实渲染测试(真实 resvg + FFmpeg,0 跳过)。
  • 从解包后的安装包独立完成中文 SVG/PNG/GIF/MP4 端到端产出,并完整解码。
  • 真实 Harness 中确认 11 个技能、能力索引与两个运行器可从安装包加载,且同名外部诱饵不能冒充。
  • 用读图工具查看实际画面并修复了字体证明图的截断显示问题。

未通过整体门槛(本预发布即在此前提下提供):12 项指导式任务不公布总通过率。三个无能力提示的原生探针已有实际产物与自主方法选择证据,但首轮/恢复分列。原 40 题在加入通用恢复防护前后各跑一轮,防护后仍有上述 4 题关键警报,目标用例通过不等于整体无退步。仍不宣称专业水平提升或统计性涨分,详见执行复核。来源研究为静态审阅,未安装或运行任何被研究对象。

quality.js、adaptive.js、pool.js 与上游 grilling 原文自 v1.5.0 起字节未变;推理选档、自适应保留、权限与审批边界均无改动。详细证据与未证事项见增强证据。

兼容门禁为 engines.dsh: ">=0.2.2-rc.1 <=0.2.2-rc.2":rc.1 是官方 latest(桌面端实际提供的版本),rc.2 是 next 预发布;两者均已隔离安装并实测通过原生集成、11 技能加载、选档、压缩事务与服务隔离,范围外版本由宿主拒绝。渲染支持范围为 Linux arm64 实测,其他平台与三维未验证。附件提供 TGZ 与 SHA256;未发布 npm,不覆盖历史标签。

English

The same Omni preset and the same omni preset ID. This release adds domain specialties and the ability to make images and video with code, using no image or video generation models.

This is a pre-release, not a stable version. Stable remains v1.5.0; this release is v1.7.0-rc.1. The features work, but critical non-regression was not achieved under the agreed standard, so acceptance is not declared as passed.

What was not achieved and must be stated plainly: re-running the byte-identical original 40 cases on the same gpt-6.1-sol model, tools and grader gives 38 baseline passes versus 34 candidate passes, with four cases moving from pass to fail (research-02, review-02, recover-03, recover-04). Review found most of those differences stem from closed values or types the prompts never specified (status enums, string versus object arrays, whether a citation list may carry extra items), but an under-specified prompt is not proof that quality did not drop. That limitation stands until the contracts are fixed, re-frozen and re-measured; see the execution review.

# Pre-release install
dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.7.0-rc.1

Or add the release TGZ through the plugin manager. Restart, create a new conversation and select 全能模式; use your actual profile. The legacy local bundle declares the same omni ID and cannot be enabled simultaneously.

What is new

  • Ten domain skills: engineering, game, security/network/RE, documents, fiction, research, personalization, assistant, molecular-biology data and business semantics. Loaded on demand, not resident.
  • Capability index: references to 100 source configurations and 23 capability entries, honestly labelled verified / needs-dependency / method-only / uncovered.
  • 11 local operations: data validation and transformation, preliminary game arithmetic, fiction continuity, FASTA statistics, entity-relation consistency, citation coverage, local indicator extraction, Markdown outlines, todo dependencies and a calendar draft in a fixed format. Node built-ins only; no new production dependency.
  • Code media: scene source → SVG → pinned resvg → FFmpeg for PNG/GIF/MP4. Authorized assets are reused when available, otherwise drawn from scratch; audio for video is asked first. Dependencies are prepared explicitly and never installed silently at activation or render time.

How far this is verified

Actually executed:

  • 263 repository tests, 64 capability contract tests and 22 real render tests (real resvg + FFmpeg, 0 skipped).
  • Independent end-to-end Chinese SVG/PNG/GIF/MP4 production from the extracted package, with full decoding.
  • In a real Harness, 11 skills, the capability index and both runners load from the installed package, and same-name external decoys cannot substitute for them.
  • Actual frames viewed with an image-reading tool, which caught and fixed a truncation display issue in the font proof sheet.

Overall gate not passed (this pre-release ships under that caveat): the twelve guided tasks have no overall pass rate. Three native probes without capability hints have actual artifacts and spontaneous method-selection evidence, with first attempts/recovery separate. The original 40 cases ran once before and once after the general recovery guard; four critical warnings remain after it. A targeted pass is not overall non-regression. No professional or statistical improvement is claimed; see the execution review. Source research remains static-only, without installing or running researched subjects.

quality.js, adaptive.js, pool.js and the upstream grilling text are byte-for-byte unchanged since v1.5.0; reasoning selection, adaptive retention and permission/approval boundaries are untouched. See enhancement evidence for details and unproven items.

The gate is engines.dsh: ">=0.2.2-rc.1 <=0.2.2-rc.2": rc.1 is the official latest the desktop app ships, rc.2 the next pre-release. Both were installed in isolation and passed native integration, 11-skill loading, reasoning selection, compaction transactions and service isolation; the host refuses anything outside that range. Rendering is validated on Linux arm64 only; other platforms and 3D remain unverified. The release attaches a TGZ and SHA256; nothing is published to npm and no historical tag is overwritten.

全能模式 / Omni mode v1.5.0

Choose a tag to compare

@lxy0714 lxy0714 released this 11 Nov 13:26

全能模式 / Omni mode v1.5.0

中文 · English

中文

本次改变何时使用内置的 grilling 技能:从「你点名才用」变为「默认启用」。技能原文、许可证和归属不变。

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.5.0

也可下载本 Release 的 TGZ 后在 DSH 插件管理器添加。安装后重启引擎、新建对话,选择「全能模式」;使用实际 profile。旧本地包与本包同为 omni ID,不能同时启用。

默认启用范围:任务含有需要你拍板的设计取舍时,先走 grilling——方案/架构选型、技术栈或依赖替换、数据模型、权限与安全边界、破坏性或不可逆操作、交付范围与验收标准、成本与时间权衡,或需求仍有会改变实现方向的关键歧义。它会按设计树分轮提问,只问当前能决定的问题,每题附推荐答案,达成共识后再实施。

不会触发:需求已明确、步骤既定、机械改动、查证事实、修复已定位缺陷、纯执行或纯读取;也不会因为你只是要求集成或安装技能而启动访谈。你随时可以说跳过访谈。

验证:真实隔离 Harness 中从本包发现并逐字加载 grilling(创建与恢复两进程均通过);31 项策略测试;TGZ 解包后重复同一断言;技能正文 Git blob 仍为上游 d2f09a18。推理为本地确定性模拟,不代表真实模型行为;云端 CI 以 Actions 实际结果为准。

兼容门禁仍为 DSH 0.2.2-rc.1;未改选档、保留策略、权限或外发边界。本次是人设策略变更,属于行为变化,故按次版本发布。附件提供 TGZ 与 SHA256;未发布 npm,不覆盖历史标签。

说明:人设只描述触发条件,模型是否严格遵循取决于模型自身。若发现该问的没问、或不该问的追问了,请带最小复现到 Issues 反馈。

English

This release changes when the bundled grilling skill is used: from "only when you ask" to "enabled by default". The skill text, its license and provenance are unchanged.

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.5.0

Or add the release TGZ through the Harness plugin manager. Restart, create a new conversation and select 全能模式; use your actual profile. The legacy local bundle declares the same omni ID and cannot be enabled simultaneously.

When it applies by default: whenever a task contains a design trade-off you own — architecture or stack selection, dependency replacement, data model, permission/security boundaries, destructive or irreversible actions, delivery scope and acceptance criteria, cost or schedule trade-offs — or remaining ambiguity that would change the implementation direction. It interviews in rounds over a design tree, asks only what you can decide now with a recommended answer, and implements after shared agreement.

What does not trigger it: fully specified requirements, established steps, mechanical edits, fact lookups, fixes of an identified defect, plain execution or reads — and merely asking to integrate or install the skill. You can say to skip the interview at any time.

Verification: in a real isolated Harness, grilling is discovered from this package and loads verbatim in both create and resume processes; 31 policy tests; the same assertions repeat against the extracted TGZ; the body's Git blob still matches upstream d2f09a18. Inference is a deterministic local mock, not proof of real-model behavior. Check actual Actions results for cloud CI.

The compatibility gate remains DSH 0.2.2-rc.1 with no change to reasoning, retention, authorization or external-sending boundaries. This is a persona policy change, so it ships as a minor version. The release includes TGZ and SHA256 checksums; no npm publication and no historical tag rewriting.

Note: the persona states the trigger conditions; strict adherence depends on the model. If it interrogates when it should not, or skips a decision it should raise, please file a minimal reproduction in Issues.

全能模式 / Omni mode v1.4.0

Choose a tag to compare

@lxy0714 lxy0714 released this 11 Nov 12:35

全能模式 / Omni mode v1.4.0

中文 · English

中文

本次把 Matt Pocock 的 MIT grilling 技能内置进预设:随包安装、无需复制到 ~/.dsh/skills,通过官方 skill-filesystem 与 tool-skill 提供。

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.4.0

也可下载本 Release 的 TGZ 后在 DSH 插件管理器添加。安装后重启引擎、新建对话,选择「全能模式」;使用实际 profile。旧本地包与本包同为 omni ID,不能同时启用。

怎么用:直接说「grill 一下这个方案」「拷问我这个想法」或「压力测试这个决策」。它会按设计树分轮提问,只问当前能决定的问题,每个问题附推荐答案,达成共识后再实施。以下不会自动进入访谈:只是要求集成/安装技能、或已明确交代需求要求直接执行。

验证:真实隔离 Harness 中,从本包内 skills/grilling/SKILL.md 发现 grilling 并逐字加载成功(创建与恢复两个独立进程均通过);31 项策略测试;TGZ 解包后重复同一断言;技能正文的 Git blob d2f09a18 与上游一致。推理为本地确定性模拟,不代表真实模型行为;云端 CI 以 Actions 实际结果为准。

兼容门禁仍为 DSH 0.2.2-rc.1;未改选档、保留策略、权限或外发边界。上游为 MIT,版权与许可见 skills/grilling/LICENSE 和第三方声明;中文说明见grilling 使用说明。附件提供 TGZ 与 SHA256;未发布 npm,不覆盖历史标签。

English

This release bundles Matt Pocock's MIT-licensed grilling skill into the preset. It ships with the package — no copy into ~/.dsh/skills — and is exposed through the official skill-filesystem and tool-skill plugins.

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.4.0

Or add the release TGZ through the Harness plugin manager. Restart, create a new conversation and select 全能模式; use your actual profile. The legacy local bundle declares the same omni ID and cannot be enabled simultaneously.

Usage: ask to "grill this plan", "poke holes in this idea" or "stress-test this decision". It interviews in rounds over a design tree, asks only what you can decide now, recommends an answer per question, and implements after shared understanding. It does not auto-start when you merely ask to integrate/install the skill or when an already-specified task should simply run.

Verification: in a real isolated Harness, grilling is discovered from this package's skills/grilling/SKILL.md and loads verbatim in both the create and resume processes; 31 policy tests plus the new skill-discovery assertions; the same assertions repeat against the extracted TGZ; the body's Git blob matches upstream d2f09a18. Inference is a deterministic local mock, so this is not proof of real-model behavior. Check actual Actions results for cloud CI.

The compatibility gate remains DSH 0.2.2-rc.1 with no change to reasoning, retention, authorization or external-sending boundaries. Upstream is MIT: copyright and permission text are in skills/grilling/LICENSE and the third-party notices; see using grilling. The release includes TGZ and SHA256 checksums; no npm publication and no historical tag rewriting.

全能模式 / Omni mode v1.3.1

Choose a tag to compare

@lxy0714 lxy0714 released this 11 Nov 10:12

全能模式 / Omni mode v1.3.1

中文 · English

中文

本次为文档补丁:补齐仓库说明、全部用户文档、贡献/安全/许可证解释、Issue 表单、工作流显示文字和发行说明的中英双语,并增加语言与链接检查。不改变运行时选档、保留策略、权限或兼容范围。

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.3.1

也可下载本 Release 的 TGZ 后在 DSH 插件管理器添加。安装后重启引擎、新建对话,选择「全能模式」;使用实际 profile。旧本地包与本包同为 omni ID,不能同时启用。

支持门禁仍为 DSH 0.2.2-rc.1。验证包含双语/链接检查、31 项测试和真实 Harness+模拟推理的生命周期/发行包回归;具体云端状态以 Actions 实际结果为准。它不是付费模型效果或全插件兼容证明。

MIT 允许使用、修改、fork 和再分发,保留许可证与上游归属。标准英文法律文本不改,提供中文解释。TGZ 和 SHA256 校验文件随 Release 附带;源码档案由 GitHub 根据 tag 生成。没有发布 npm,也不覆盖历史标签。

English

This documentation patch completes Chinese/English repository text, user docs, contribution/security/license explanations, Issue forms, workflow labels and release notes. It adds language/link checks without changing runtime reasoning, retention, authorization or compatibility.

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.3.1

Or add the release TGZ through the Harness plugin manager. Restart, create a new session and select 全能模式; use the actual profile. The legacy local bundle declares the same omni ID and cannot be enabled simultaneously.

The compatibility gate remains DSH 0.2.2-rc.1. Verification includes bilingual/link checks, 31 tests and native Harness lifecycle/package regression with mock inference. Check actual Actions results for cloud status. This is not evidence of paid-model quality or universal plugin compatibility.

MIT permits use, modification, forks and redistribution with notices retained. The canonical English license is unchanged and has a Chinese explanation. The release includes TGZ and SHA256 checksums; GitHub generates source archives from the tag. No npm publication or historical tag rewriting.

全能模式 / Omni mode v1.3.0

Choose a tag to compare

@lxy0714 lxy0714 released this 11 Nov 09:48

全能模式 / Omni mode v1.3.0

中文

这是首次公开 GitHub 发行,采用 MIT 许可证,允许使用、修改、fork 和再分发;保留许可证和上游归属。

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.3.0

也可将本页 TGZ 添加到 DSH 插件管理器;重启引擎、新建会话,选择「全能模式」。使用实际 profile。旧本地包与本包均声明 omni,不能同时启用。

实测范围:DSH 0.2.2-rc.1、Node24;31 项测试、真实 Host 中的压力裁剪、切小窗口、手动压缩及两进程正常恢复;解包消费及公开 GitHub 安装检查已通过。推理是本地确定性模拟,不证明付费模型质量或所有平台/插件兼容。原版 CI已通过。

插件不需要独立 API key,不覆盖宿主工具权限与账号,也没有发布 npm。附件提供 TGZ 和 SHA256;源码由 GitHub 按 tag 生成。此为历史版本,完整双语文档请使用 v1.3.1;不覆盖本版本 tag 或安装包。

English

This first public GitHub distribution is MIT licensed: use, modify, fork and redistribute while retaining applicable notices.

dsh plugin --profile web add github:lxy0714/dsh-omni-mode#v1.3.0

Or add this release's TGZ through the Harness plugin manager. Restart, create a new conversation and select 全能模式; use your actual profile. The legacy local bundle declares the same omni ID and cannot be enabled together.

Verified scope: DSH 0.2.2-rc.1, Node24; 31 tests, real Host pressure pruning, smaller-window switching, manual compaction and normal two-process recovery. Extracted-package and public GitHub installation checks passed. Inference is a deterministic local mock, not proof of paid-model quality or universal platform/plugin support. Original CI passed.

No separate plugin API key is needed. Host tools, permissions and accounts remain authoritative. No npm publication. Attachments provide TGZ and SHA256; GitHub generates tagged source archives. This is historical: use v1.3.1 for complete bilingual docs. Its original tag and installation assets are not overwritten.