OpenQA Verifier
Verify the change. Then decide whether to accept it.
An independent verifier that runs locally or in CI and produces auditable merge and release readiness verdicts for every software change.
- Understand intent:read goals, requirements, diffs and repository rules.
- Analyse risk:find blast radius, out-of-scope edits and evidence gaps.
- Execute independently:call tests, analysis, scanners and real environments.
- Issue verdicts:state merge and release readiness and who must still review.
- 5
- task stages
- 7
- connected platforms
- 0
- installable skills
Do not trust completion claims. Verify the change directly.
Understand intent and change
Reads the task, requirements, repository and Git diff to extract acceptance conditions, permitted scope and what this change claims to complete.
Evidence goes to智能代码知识图谱 ↗Analyse impact and risk
Finds affected modules, APIs, screens and dependencies; checks scope drift, standards and evidence gaps; and determines the required verification depth.
Evidence goes to智能代码覆盖率系统 ↗Plan verification
Composes existing tests, static checks, security scans and organisation rules. When evidence is missing, it proposes reviewable checks or cases.
Evidence goes to智能用例管理平台 ↗Execute and collect evidence
Runs necessary checks locally, in CI, against APIs, browsers or devices, retaining direct command output, responses, logs, traces, screenshots and recordings.
Evidence goes to智能云真机 ↗Produce acceptance verdicts
Separates clear failures, insufficient evidence and human judgement, binds evidence to code and policy versions, and issues distinct merge and release readiness verdicts.
Evidence goes to智能 UI 自动化测试平台 ↗
Automated execution does not mean giving up control.
- 01
You choose the project and explicitly grant access to repositories, platforms and devices. Nothing outside that scope runs.
- 02
Intent, impact analysis, verification planning, execution and evidence collection are shown step by step. Stop the task whenever something looks wrong.
- 03
New checks, policy changes and high-risk writes enter review first. Every acceptance binds code, policy versions and direct evidence.
- Web
- iOS
- Android
- 小程序
- REST
- GraphQL
- gRPC
- Playwright
- pytest
- JUnit
One acceptance across the existing engineering stack.
- Internal
智能用例管理平台
Turn requirements into cases, inside the review and maintenance flow your team already has.
Case ManagementDue 2026 Q4
- Internal
智能接口自动化平台
One place for API definitions, cases, mocks, environments and regression results.
API TestingDue 2026 Q4
- Internal
智能 UI 自动化测试平台
Record, generate and self-heal web and app UI automation.
UI AutomationDue 2027 Q1
- Internal
智能云真机
Let the agent use a phone the way a person does.
Device & EnvironmentDue 2027 Q1
- Internal
智能代码知识图谱
Knowing what changed is what tells you what to test.
Code IntelligenceDue 2027 Q2
- Planned
智能代码覆盖率系统
Not just the percentage — what the untested part actually is.
Code IntelligenceDue 2027 Q2
- Planned
知识库
Your team's business rules, past defects and testing lore, readable by an agent.
Knowledge BaseDue 2027 Q2
Independent verification still needs explicit boundaries.
The verifier may need to run commands, reach the network or use credentials, but none is granted by default. The authoring agent cannot change protected acceptance policy; evidence comes directly from tools and environments, and any change to code, policy or evidence invalidates the prior verdict.
- Runs shell
- Network egress
- Needs credentials
It is not ready for public download. Progress stays public.
The desktop is still in private beta with a small number of teams. The dates below are the current plan, not promises; completed work, delays and changed plans remain on the record.
- 2026 Q3In Progress
Agent 桌面端内测
macOS / Windows 内测包，连接用例管理与接口平台；用例生成、接口用例两条链路先跑通。
- 2026 Q4Planned
Agent 公测
macOS / Windows 公开下载；接入 UI 自动化平台，浏览器内录制与自愈脚本。
- 2027 Q1Planned
Agent 1.0 正式版
五条能力链路稳定；技能库、五个平台全部打通。
- 2027 Q2Exploring
Agent Linux 版与无头 CLI
在 CI 容器里运行 agent，无桌面依赖。