bpc-oss/dsh-verification

Name: bpc-oss/dsh-verification
Author: bpc-oss

Plugin插件 ⭐ 2 Apache-2.0 Productivity & Tasks效率与任务 Approval & Security审批与安全

Project Overview项目介绍

Edited by Zhiwei Agent (zhiwei)由知微Agent(zhiwei)编辑

This repository is a native verification plugin built exclusively for DeepSeek Harness (DSH), ported from the Conscience layer of the Bobby project. It adds a completion check gate to DSH agents that requires every acceptance criterion to be backed by real evidence captured from actual tool outputs, rather than relying only on the model’s own self-reporting. It ships as three separate packages: the core verification engine, pure Zod schemas for contracts and evidence, and a client-side UI settings panel for DSH. In enforce mode, it blocks unverified task completions and returns a list of defects, while advisory mode only records verification results without blocking any tasks.

It addresses the common problem of "false completion" that plagues many AI agents, where a model claims a task is done but actual evidence does not back that claim up. Common examples include a model saying a file exists when it does not, claiming tests passed when they actually failed, or stating a review was completed when there is no log of it running. This plugin catches all these cases before they can result in an undeliverable or broken task being marked as complete. It follows a clear workflow: selectors for evidence are frozen during the planning phase, tool outputs are automatically captured during execution, and each acceptance criterion is verified before a task can be marked done.

The project is released under the open-source Apache-2.0 license, and has been extensively tested on the ASI-Bench benchmark, with results showing that combining this verification layer with RSI boosts task completion rate from 83% to 100% while cutting average task time and token consumption by nearly half. When installing the plugin, users need to avoid adding extra DSH official runtime dependencies, as this can cause symbol conflicts that crash the tool scheduler. Built artifacts are already committed to the repository, so you can use it directly after cloning, and any changes to the GUI require a restart of the DSH web process to take effect. All 172 unit tests currently pass, and synthetic testing confirms 100% recall and 0% false positives.

dsh-verification 是专为 DeepSeek Harness (DSH) 开发的原生验证插件，移植自 Bobby 项目的良心层，核心作用是为 DSH 代理添加完成验证闸门，要求任务在标记完成前，每条验收标准都必须获得真实工具输出证据支持，而非仅依赖模型自我报告。在强制模式下，它会拦截未通过验证的任务并返回缺陷清单，建议模式仅记录所有结果不会拦截。项目包含三个核心包：验证引擎、 zod 模式定义和客户端设置面板。

该插件解决了代理常出现的“虚假完成”问题，比如模型声称完成任务但实际缺少对应文件、测试未通过或没有审查记录，这些问题都会被该插件拦截，避免未达标任务直接交付。用户可在规划阶段冻结每个验收标准对应的证据选择器，执行阶段工具调用会自动捕获为内容寻址证据，完成阶段逐一验证每个验收标准，全程事件可追溯重放。适合对任务完成质量有要求的 DSH 开发者和日常使用者。

该项目采用 Apache-2.0 开源许可证，已通过 ASI-Bench 测评，验证工作流结合 RSI 可将任务完成率从 83%提升至 100%，同时降低平均耗时和 token 消耗。安装时需要注意，不要额外安装 DSH 官方运行时依赖，避免符号冲突导致工具调度崩溃，构建产物已提交，克隆后可直接使用，GUI 更改后需要重启 web 进程生效。内部所有单元测试全绿，合成测试召回率 100%，误报率 0%。

View on GitHub在 GitHub 查看

Send to your DSH发给你的 DSH 安装

请帮我安装这个 DSH 插件。安装前先完成【兼容性检查 + 安全性检查】，检查通过再动手。

插件：dsh-verification（bpc-oss/dsh-verification）
仓库：https://github.com/bpc-oss/dsh-verification
本站详情页：https://www.yhbd.top/plugins/bpc-oss-dsh-verification/
本站登记：类型 plugin · 归类 原生 DSH 插件 · 许可证 Apache-2.0 · ⭐ 2 · 最近提交 2026-08-26 · 主语言 TypeScript · 未检测到 DSH 插件清单

按下面顺序执行，每步先把结论告诉我，再进入下一步：

【1 兼容性检查】
① 我这边：DSH 版本、Node 版本、操作系统、当前 profile（web / desktop）。
② 读它的 README、package.json、插件 manifest，列出它要求的 DSH 版本 / Node 版本 / 操作系统 / 外部依赖 / 需要另外先装的运行时。
③ 逐条比对，结论只写「满足 / 不满足 / 未知」三种；不满足的给出可行替代方案。
④ 检查是否和我已装的插件冲突：命令名重复、skill / tool 重名、端口占用、重复注册的 MCP server。

【2 安全性检查】
① 仓库可信度：和上面「本站登记」是否一致；star / fork 数、创建时间、最近提交，是否归档或长期停更。
② 安装脚本：逐行看 package.json 的 preinstall / install / postinstall，以及 install.sh、setup.ps1 之类脚本。出现 curl|bash、下载后直接执行、混淆代码、访问与插件功能无关的域名，立刻停下来告诉我，不要继续装。
③ 依赖：列出新增依赖，标出无人维护、或与知名包拼写近似的可疑包（typosquatting）。
④ 权限与副作用：它会读写哪些目录、访问哪些域名、需要哪些 DSH 权限（filesystem / network / shell / clipboard 等），以及怎么卸载和回滚。
⑤ 如果它要求 sudo / 管理员权限，或权限明显超出功能所需，先停下来问我。

【3 安装】
上面两步没有「不满足」和「高危项」时才执行；用官方推荐方式安装，不要自行提权。

【4 汇报】
用表格输出：检查项 / 结论 / 依据 / 是否需要我决策。拿不准的一律写「未知」并说明要我怎么确认——不要猜，也不要替我决定。

Send this message to DSH in your current session: it verifies compatibility and security first (answering met / not met / unknown item by item) and only installs once everything checks out — it will stop and ask you if it finds a high-risk item. The box scrolls; the copy is the full prompt. CLI install commands may not be accurate across systems, so DSH is the safer route.把上面这条消息直接发给当前会话里的 DSH：它会先核对兼容性与安全性（逐条给「满足 / 不满足 / 未知」），确认没问题再安装，有高危项会停下来问你。框内可滚动，复制到的是完整提示词；安装命令不一定准确，发给 DSH 更稳。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 2 warnings2 项注意

Only 2 stars - very few users, little community feedback星标只有 2，几乎没人在用，遇到问题缺少社区反馈
No DSH plugin manifest detected - it may only carry the dsh-plugin topic, so the install method must be confirmed on the spot未检测到 DSH 插件清单：可能只是打了 dsh-plugin 话题，安装方式要现场确认

DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时（Electron / Python / Docker 等）是否齐备
Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突：命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致，是否归档或长期停更
Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需（filesystem / network / shell / clipboard）
Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限，以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照，不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装（适合开发者）

CLI Install命令行安装

dsh plugin --profile web add github:bpc-oss/dsh-verification

把 bpc-oss/dsh-verification 加入你的 DSH 配置（web profile）即可启用。

READMEREADME

20.6 KB · 缓存于 2026-09-20 19:38:26 · 20.6 KB

dsh-verification

Verification engine for DeepSeek Harness (DSH) agents — a port of the Bobby "Conscience" layer.

"Claimed done" becomes "actually done." Before an agent can complete a goal, every acceptance criterion must be backed by server-stamped, real tool evidence (tool outputs, not model self-reports). The completion gate rejects unverified completions with a defect list; advisory mode records everything without ever blocking.

plan: freeze selectors (tool + args hash) for each AC
work: tool calls auto-capture as content-addressed evidence
complete: each AC needs a real-evidence pass, else the gate returns the defect list
audit: evidence/verdicts/gate are append-only session events, replayable

Package	What it does
`@bpc-oss/dsh-verification`	engine: task epochs, contracts, oracles, completion gate
`@bpc-oss/dsh-evidence`	pure zod schemas: contracts, evidence, verdicts, selectors
`@bpc-oss/dsh-client-ui-verification`	client: verification settings panel

Benchmarks (reproducible): synthetic recall 100% / false-positive 0% (pnpm --filter @bpc-oss/dsh-verification bench); real-task replay audit and completion-capability evaluation in docs/value-report.md and scripts/.

ASI-Bench Evaluation: enforce + RSI (seed31415, B3)

We evaluated enforce + RSI (verification workflow + Research Skill Intelligence) against official baselines on ASI-Bench (60 scientific tasks, Level B3, deepseek-v4-flash):

Condition	Completed	Format score	Avg time	Tokens/task
enforce + RSI 🏆	60/60	0.979	18.4 min	5.28M
enforce only	60/60	0.975	23.5 min	7.20M
enforce + RSI (masked)	48/59	0.790	17.9 min	6.36M
standard (official)	50/60	0.799	38.5 min	11.20M
minimal	7/59	0.080	12.9 min	7.78M

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要，完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

渲染基于仓库 bpc-oss/dsh-verification 的 README.md。在 GitHub 查看原文 →