测试与调试

A local audit tool for AI API relays and LLM proxies that checks prompt injection, model-substitution signals, context or tool-call rewriting, stream anomalies, and error leakage, then produces Markdown reports.

A DeepSeek Harness testing mode for organizing penetration-test goals, findings, assets, exploration links, and reports.

Flameox 通过 CLI 和 MCP 收集并分析本地性能、内存、执行、并发及可靠性证据，帮助智能体调查运行时问题。

一个只读检查 DSH 插件仓库清单、补丁格式、构建配置、生态合规性和 Hub 收录状态并生成诊断报告的工具。

在 DeepSeek Harness Web 中查看智能体修改的文件差异、撤销当前轮次更改，并通过行评论反馈修改意见。

面向鸿蒙开发者的设备调试插件，集成 hdc 设备 निरी查、截图、UI 操作、应用测试、崩溃诊断以及 DevEco 构建和模拟器控制。

A DSH code review skill that analyzes code through a multi-step workflow, grades issues, assigns quality scores, and generates text and HTML reports.

记录并整理 DeepSeek Harness 中的工具失败，去重统计后写入技能日志并提供针对常见错误的预防提示。

用于检查 DSH 插件的文件、依赖和配置，并通过隔离试跑验证插件启动情况，提供修复与回滚操作。

只读扫描 DSH 多帧 zstd 会话文件，诊断损坏、截断、空会话和残留文件并生成健康报告。

用于检查 DeepSeek Harness Web 的内置能力状态、筛选诊断条目，并管理少量经审阅的界面开关。

A CLI and agent-callable utility for checking DeepSeek Harness plugin manifests, builds, packages, installations, configuration, and selected environment conditions.

Maps code changes to relevant tests, checks changed-line coverage, and records verification results that become stale when the code changes.

在 Chrome 或 Chromium 中标注页面元素，收集 DOM、样式、可访问性信息、评论和可选截图，并将结构化反馈发送给智能体。

一个通过 checkup、fix 和 review 工具完成项目问题检查、修复交付与质量复查的工作流插件。

A DSH plugin diagnostic tool that searches, compares, audits, and prechecks plugins for overlap and dependency conflicts while providing usage analysis and relationship diagrams.

A DeepSeek Harness diagnostics and recovery plugin that scans configuration and startup issues and manages local checkpoints and repair actions through a web interface or CLI.

用于为 DSH 插件和技能运行 YAML 用例回归评测、检查会话轨迹与输出、对比基线并生成 CI 门禁结果。

Click an element in a local frontend development page to locate its Vue, React, Svelte, or Angular source file and line, then place the result in the DSH chat draft.

用于检查 DeepSeek Harness 插件与指定运行环境的依赖、安全公告及安装加载兼容性，并生成可审阅的影响与修复报告。

通过语言服务器为智能体和编辑器提供代码诊断、补全、格式化、代码操作、符号信息及跨工作区重命名功能。

A local evaluation workbench for running and comparing DeepSeek Harness candidate profiles with hard gates, Pareto ranking, and web-based experiment reports.

为 DeepSeek Harness 生成、运行并在有限迭代内修复项目单元测试。

Checks PowerShell tool calls for common syntax and command pitfalls before execution, with optional PSScriptAnalyzer support and deny-or-warn guidance.

为 DeepSeek Harness 提供安装环境与运行状态诊断、结构化报告及有限的本地修复功能。

A local rescue console for diagnosing failed DeepSeek Harness plugin loads and temporarily skipping or restoring affected plugins.

This plugin provides a process-isolated sandbox testing workflow for DeepSeek Harness, letting users create, run, inspect, stop, and destroy test instances with an explicit pre-merge validation gate.

Anchorlaw 是面向 DeepSeek Harness 的代码验证工具链，可扫描防御性编程模式、生成健康报告，并为 AI 辅助开发提供验证工作流工具。

DSH-WM is a DeepSeek Harness world-model plugin for locally inspecting rollout data, comparing predictions to ground truth, producing scoring and visual outputs, and guiding structured debugging and iteration workflows.

A DeepSeek Harness plugin for debugging and monitoring LLM requests and responses through reasoning control, stream analysis, thinking capture, and optional session audit persistence.

用于对 DeepSeek Harness 工具管线执行确定性故障注入和韧性测试，并生成带断言的结构化测试报告。

用于在授权范围内执行网络与应用层安全评估、验证发现并记录证据和生成报告的 DeepSeek Harness 插件。

用于验证 CLI、MCP 与 GUI 之间动作绑定、回放结果、内容哈希和状态转换一致性的测试工具。

用于对命令和 JSONL 运行器执行确定性基准测试、生成报告并进行基线回归比较。

用于验证恢复演练证据的测试调试插件，可检查清单、事件序列和对象哈希并生成验证报告。

用于验证发布清单中多个下载源的状态、大小、版本和 SHA-256，并生成确定性的证据文件。

A plugin that reviews recent Git changes, suggests focused code simplifications within changed ranges, and runs existing project tests.

监视 GitHub Actions 失败运行，将作业日志整理为结构化诊断，并记录重复出现的失败签名。

Checks AI-generated projects through builds, tests, feature verification, and user simulation, then reports detected issues back to the AI for fixing.

通过 SecurStack CLI 为 DeepSeek Harness 提供仓库安全扫描、策略检查和环境诊断能力。

这是一个面向 DeepSeek Harness 的本地故障诊断插件，可在工具或 Agent 失败时生成经过限量采集和脱敏处理的证据 ZIP 包。

这是一个为 DeepSeek Harness 提供兼容处理的插件，用于移除非升级沙箱请求中的冗余字段，避免 bash、pwsh、write 和 edit 调用被错误拒绝。

在代码编辑后自动运行本地 TypeScript 类型检查，并通过 code_check 工具返回结构化诊断信息。

A task-protection and debugging plugin that preflights injection declarations, detects tool-call loops and excessive workflow recursion, and provides status and safer restart assistance.

用于 DeepSeek Harness 的确定性回归测试插件，可通过必需和禁止片段评估输出并生成适合 CI 的结果报告。

为 DeepSeek Harness 提供代码审查、简化、计划执行、测试优先和冲突解决等工程实践技能。

A deterministic contract regression testing tool for DeepSeek Harness plugins that runs real pipelines in isolated subprocesses and validates tool calls, event streams, failures, cancellation, and concurrency.

用于扫描 MCP 配置和技能文件中的提示注入、隐藏 Unicode、危险 Shell 模式及凭据泄露，并提供可选的非阻断式工具调用运行时观察。

面向 QA 测试的 DeepSeek Harness 工作台，支持浏览器或 Electron 界面验证、视觉回归、会话分析和图像模型路由。

用于检查并验证 DSH 配置声明、pnpm 锁定导入器、已安装包清单与 bundle 补丁元数据是否一致的验证插件。

用于基于固定哈希的离线记录验证 DSH ToolRuntime、MCP JSON-RPC 与 CLI JSON 接口之间的语义契约一致性。

该插件根据明确的就绪要求验证预先收集且经过脱敏的 Windows 证据，并生成确定性的检查报告。

A read-only DSH plugin compatibility auditor that statically checks plugin combinations for common startup-conflict signals before installation.

通过宿主 shell 封装 Playwright CLI，支持检测版本、安装浏览器、运行测试并访问 HTML 报告。

Adds cooperative timeout protection for DeepSeek Harness ask_user_question calls so unanswered questions do not block a turn indefinitely.

通过 ADB 为 Android 设备和台架提供日志查看、APK 安装、文件传输、性能采集及崩溃与健康诊断能力。

通过运行公共 Falsify CLI，对文件或拉取请求执行静态检查、文档审查和门禁评估，并返回结果回执。

自动检测常见 JavaScript 和 Python 测试框架，执行指定测试并返回结构化的通过、失败与错误摘要。

A semantic linting plugin and CLI that checks LLM instruction files for overly broad wording and semantic traps, with optional rule injection and lifecycle-hook warnings.

这是一个用于 DeepSeek Harness 的只读运行时诊断插件，可检查智能体能力、工具、模型、技能、工作区和会话持久化状态并报告问题。

为模型提供面向部署指定目标的受限 HTTP GET 探测工具，并返回带状态、内容类型和截断标记的结构化结果。

Provides Python and JavaScript/TypeScript projects with language-server diagnostics and navigation, debugger integration, and discovery or execution of npm, Make, and just tasks.

用于静态检查 DeepSeek 兼容工具调用请求及录制流数据中的协议、架构、配置和拼接问题。

为 DeepSeek Harness 提供增量 Git 差异审查面板和 /review 命令，记录审查检查点并将反馈发送回智能体。

Opens Git diffs from the DSH web interface in a standalone side-by-side review page with syntax highlighting, file navigation, and inline comments.

这是一个面向 DeepSeek Harness 的只读故障复盘插件，可从既有会话事件日志生成脱敏报告、导出 JSON 并提供仅可复制的修复提示。

Popper is a DeepSeek Harness plugin for evidence-driven debugging that validates explicit claims, runs configured gates, and requires new hypotheses after falsification.

根据产品需求、Git 提交信息或用户故事生成带优先级的功能测试用例，并输出 Excel 风格测试报告。

用于核验设计文档与重构计划中的可验证声明，并对状态机模型执行结构化和对抗性逻辑检查。

A DeepSeek Harness startup guard that repairs session logs, validates plugin composition and client bundles, smoke-tests host apply(), and isolates problematic plugins.

This plugin for DeepSeek Harness provides agent trajectory replay by forking from turn, message, or tool checkpoints so you can run replay sessions, step through execution, and compare original versus replayed results.

用于调试和学习 DeepSeek Harness 的插件，可在专用界面中检查模型调用的提示词、消息、工具、选项及上下文组成。

用于检查 DeepSeek Harness 配置补丁和插件问题的诊断工具，可报告失效补丁、配置字段丢失及其他配置差异，并提供可选的定向修复。

Provides DeepSeek Harness with pre-execution rule interception and independent read-only delivery reviews for agent actions and changes.

A read-only DeepSeek Harness diagnostic tool that checks installed plugins for loader health, dependencies, compatibility, scripts, duplicates, patch integrity, and provenance.

在文件编辑成功后运行用户配置的类型检查器或代码检查器，并将诊断结果附加到同一工具结果中。

A DeepSeek Harness plugin for inspecting turn-level code changes, browsing file diffs, and performing guarded undo operations from recorded snapshots.

A DeepSeek Harness dependency diagnostic plugin that predicts installation conflicts and detects duplicate core packages, version drift, and vendored services after installation.

用于在 DeepSeek Harness 中可视化、回放、分析和调试智能体运行轨迹，并支持断点、分支比较及受控重运行。

用于分析智能体会话日志中的工具成功率、令牌效率、冗余调用和错误模式，并比较不同会话以发现回归。

记录 DSH 配置与插件状态快照及变更历史，帮助用户排查环境何时发生了变化。

将会话日志、失败命令和可用的 Git 差异进行脱敏并导出为可重放的问题复现包。

通过Lean内核重放验证数学和物理推导，并提供定理锁定、检查与验证证据存证工具。

Provides DeepSeek Harness with devecocli-backed HarmonyOS device and emulator management, UI automation, build and deployment, log diagnostics, linting, signing, and offline documentation search.

用于运行智能体基准测试、分析会话轨迹、计算脚本和模型评审指标，并生成报告及运行对比结果。

DSH Testkit runs deterministic lifecycle conformance tests for DeepSeek Harness plugins in disposable Docker environments, including packaging, installation, execution, removal, and recovery checks.

A DeepSeek Harness plugin for multi-round code review, issue triage, bounded fixes, validation and rollback, with read-only review and iteration progress panels.

用于诊断 DeepSeek Harness 环境、按范围修复配置与依赖问题、监控运行状态并支持回滚的命令行插件。

Automatically repairs invalid tool-call message pairings and ordering in DeepSeek Harness to help interrupted sessions continue running.

在安装 DSH 插件前检查本地版本、仓库状态、构建入口和 peer 依赖兼容性，并生成体检报告。

Lujo-MCP 是一个通过 MCP 为 AI 编程客户端提供浏览器、网络和运行时调试上下文的调试追踪服务。

A testing and debugging plugin that integrates Frida device discovery, process attachment, script loading, RPC calls, and event collection into DeepSeek Harness.

将本机原生 Playwright CLI 注册为 DSH 工具，供智能体执行浏览器测试、报告查看、截图和浏览器安装等命令。

Connects Xcode’s headless MCP capabilities to DSH so agents can build, test, debug, inspect Apple projects, and interact with simulators.

A plugin for cross-model review of code and AI responses with quote-based follow-up questions and review history.

这是一个通过多轮执行命令并统计中位数、耗时范围或成功率来辅助回归测试的命令行插件。

通过代码、数据流边界和 Git 提交二分法，帮助定位难以确定的缺陷根因。

用于调查 EADDRINUSE 端口冲突，并根据进程信息选择复用服务、切换端口或精确终止残留进程。

通过像素级比较两张已有截图，生成差异图、三联对比图和相似度指标。

一个用于观察和调试 DeepSeek Harness 事件流的本地审计面板，提供事件监听、计数、最近事件查看及 JSON 接口。

用于在本地运行 DSH 插件的运行时验证，检查 agent 循环行为并生成可复现报告。

用于诊断、修复并验证 AI 编程代理的安装、运行环境、配置、认证和网络问题的命令行与 MCP 工具集。

A local code-security review plugin that scans source files and Git diffs, verifies fixes, analyzes dependency risks, and generates policy-based reports.

Repeatedly runs Vitest, Jest, pytest, or node:test suites to identify flaky tests and manage detection history and quarantine records.

This plugin provides DeepSeek Harness with ROS2 diagnostics and debugging capabilities, covering inspection of nodes, topics, services, parameters, TF, graph, and bag info, and enabling build/install/parameter/recording workflows plus optional GUI observation and interaction after user approval.

通过视觉模型分析 Android 截图，并使用 adb 检查界面、执行点击滑动输入及应用安装启动，用于移动应用测试。

为智能体提供只读网络诊断工具，包括 DNS 查询、Ping、TCP 端口检查、TLS 证书检查、路由追踪和公网 IP 查询。

Provides an extensible model-facing diagnostic tool for inspecting live plugin, tool, credential metadata, approval policy, token usage, and subagent state.

为智能体提供结构化的代码和拉取请求审查流程，涵盖正确性、安全性、性能、代码风格与测试问题。

Provides isolated DeepSeek Harness test sandboxes for building and mounting local plugins, with GUI and agent controls for instance lifecycle and logs.

An acceptance-driven DeepSeek Harness plugin that independently checks tests, builds, and other project criteria before certifying completion.

Runs reproducible acceptance tests for agent-built web applications in real browsers and generates pass/fail reports with screenshot evidence.

A toolkit for checking workspace claims, validating JSON/YAML configuration, and running read-only URL, npm, and GitHub probes.

A DeepSeek Harness diagnostics plugin that provides bounded content-free in-memory request records, read-only queries, diffs, and health statistics for debugging model requests.

A DSH plugin for testing and debugging session messages by parsing context, editing and applying message changes, injecting assistant drafts, and importing or repairing session logs.

用于检查有序 DSH 配置包堆栈中的 Cordis 补丁归属冲突，并可在临时环境中运行有限运行时验证的本地命令行工具。

在结案前运行工作区验证命令并记录结果，且在发生修改后要求重新验证的插件。

A testing tool that checks whether supported Cordis-managed runtime resources return to baseline after fixture mount, exercise, unmount, and settling, then produces bounded audit receipts.

用于定义任务、运行智能体评测套件、自动评分并比较多次结果的自托管评估插件。

通过本地 Web 界面浏览和调试 Agent 历史会话，查看时间线、工具调用、错误及统计信息。

用于驱动浏览器或 Electron 应用执行 UI 操作、行为断言、可访问性检查和视觉回归测试。

用于读取 git diff 和指定代码片段，按检查清单生成结构化代码审查结果。

Provides on-demand engineering skills for test-driven development, code review, debugging, requirements specification, and domain modeling.

一个用于检查 TypeScript 和 JavaScript 代码变更中的语法、类型及可选 lint 问题，并将诊断结果展示给用户和 AI 助手的插件。

用于对代码变更或指定文件执行多视角审查，结合静态规则与模型分析生成带严重度、证据和修复建议的报告。

This plugin pre-checks DSH plugin conflicts (such as duplicate loader entries and patch issues) and optionally runs a local `dsh web` smoke launch check, with results viewable from the settings tab or via an Agent tool call.

A DeepSeek Harness rescue toolkit that diagnoses broken profile compositions, backs up and restores configuration, and can disable problematic entries during recovery.

A lightweight DeepSeek Harness plugin that registers a `test` smoke-test tool to help confirm the plugin is installed and callable.

通过授权门禁和预算限制调用 garak 对配置的 LLM REST 端点执行基线安全扫描，并生成可审计的证据报告。

An API debugging toolkit for sending and inspecting REST/HTTP requests, parsing or executing curl commands, and comparing responses within a session.

A regression-testing plugin that evaluates DSH policy fixtures, reports pass or fail results, and detects policy decision drift without invoking protected tool implementations.

为 DeepSeek Harness 提供实时的代码诊断、格式化、结构分析、导航和可配置自动修复功能。

A DeepSeek Harness testing tool that installs plugins in temporary profiles, runs boot and optional capability checks, and generates structured result reports.

Fixes delayed completion detection for the persistent bash tool in affected older DSH minimal presets and provides a minimal-fast preset.

Northstar 用多层确定性检查、参考实现比对和不变量验证来评估 AI 生成的 Python 实现及数据处理结果，并输出判定报告。

为 DeepSeek Harness 提供由用户手动运行的 Windows 环境诊断命令，检查运行时、工作区、代理设置、PowerShell 和原生文件夹选择器依赖。

This is a DeepSeek Harness client plugin that checks browser runtime capabilities such as document, canvas, MutationObserver, and Image, then outputs a JSON report to help validate whether DOM/pixel-based plugin approaches are feasible.

Replay Lab is a DeepSeek Harness web plugin that replays a completed session turn with selected candidates and compares request-surface evidence, trajectory, and execution metrics to support regression debugging.

Connects DSH Agent to Cheat Engine for single-player process debugging, memory scanning, pointer analysis, breakpoint operations, and controlled memory modification.

This DSH plugin parses OpenAPI 3.x and generates a runnable Node.js mock server with schema-based sample data through tools like mock_generate/mock_sample, for API simulation during development and testing.

Generates pytest or vitest API test skeletons from OpenAPI 3.x specifications for success, missing required path parameters, and invalid enum scenarios.

Analyzes local log files by aggregating errors, exceptions, and warnings and extracting matching lines with surrounding context for debugging.

Parses common code coverage reports, summarizes overall and per-file coverage, and locates uncovered line ranges in selected files.

A browser-based frontend verification plugin that lets agents inspect pages, check console and network issues, validate layout facts, interact with interfaces, and save screenshots.

Reviews code changes with parallel agents and adversarially verifies each finding before reporting confirmed defects.

dsh-mp-automator is a dsh plugin for WeChat Mini Program testing that provides mp_* tools to run real DevTools sessions with actions, inspections, screenshots, and console-based validation.

This plugin is for dsh agents to drive a real WeChat Mini Program in Developer Tools, using mp_* operations to run session management, inspection, actions, screenshots, and console checks for automated testing.

用于运行测试命令并解析多种测试框架及 TypeScript 编译失败输出的测试调试插件。

A web sidebar plugin for reviewing the latest agent-turn file diff, approving it or requesting changes by path, and optionally restoring selected files to the turn-start workspace state.

A repository linter that identifies chain-of-thought leakage and session-oriented narration in documentation, comments, JSDoc, and optional source files, with CI-friendly output.

This plugin reads existing AMDGPU HSACO or ELF artifacts and provides bounded JSON queries for code-object metadata and instruction information.

ReproLab is a DeepSeek Harness plugin that freezes repro tests and candidate fixes, runs RED/GREEN/REGRESSION checks, and produces reproducible evidence for fix validation.

通过向网页注入条件式 crypto.randomUUID poly填充，为纯 HTTP 局域网地址修复 DeepSeek Harness Web UI 的连接问题。

A JWT decoding utility for debugging that displays the header, payload, signature presence, and time claims without verifying signatures or other security claims.

A DeepSeek Harness plugin failure-protection tool that inspects plugin structure, attributes crashes, records local telemetry, and supports disabling and safe-mode recovery.

This plugin runs configured or auto-detected workspace verification commands before an agent completes a goal or turn and blocks completion when verification fails.