Show HN: OQP – AI 에이전트를 위한 검증 프로토콜

hackernews | | 📰 뉴스
#agents #ai #ci/cd #openqa #verification #claude #review
원문 출처: hackernews · Genesis Park에서 요약 및 분석

요약

AI 에이전트 시대의 소프트웨어 검증을 위해 설계된 개방형 표준인 OQP는 코딩 에이전트와 CI/CD 파이프라인이 시맨틱 비즈니스 규칙을 조회하고 검증 워크플로우를 자동화할 수 있는 공통 언어를 제공합니다. 이 프로토콜은 OpenAPI 3.1 사양을 기반으로 에이전트가 테스트 기능을 발견하거나 코드 변경 위험을 평가하는 등 4가지 주요 상호작용을 정의합니다. OQP는 개발 구현부터 검증 실행까지 연결하는 '그린 계약' 역할을 하며, 중앙 위원회의 승인 없이 기업이 자유롭게 기능을 확장할 수 있는 오픈 바자르 개념을 추구합니다.

본문

The Open QA Protocol (OQP) is an open standard designed to facilitate agentic software verification. It provides a common language for coding agents, CI/CD pipelines, and human developers to query semantic business rules, assess release risk, and execute autonomous testing workflows against a centralized Knowledge Graph. In the era of agentic software development, the bottleneck is no longer writing code—it is verifying that the code satisfies business requirements. Traditional UI-driven testing tools and monolithic dashboards are incompatible with autonomous agents. OQP solves this by defining a standardized set of API primitives that allow any agent (e.g., Cursor, Claude Code, GitHub Copilot) to: - Discover testing capabilities available in a repository. - Query the semantic business rules and historical edge cases for a specific workflow. - Assess the risk of a proposed code change before merging. - Execute sandboxed, autonomous verification workflows. The core of OQP is an OpenAPI 3.1 specification. OQP defines four primary interactions: | Endpoint | Purpose | Caller | |---|---|---| GET /.well-known/oqp | Capability discovery. Agents query this to learn what the server supports. | Coding Agents | GET /context/workflows/{id} | Returns semantic business rules and historical incidents for a workflow. | Coding Agents (via MCP) | POST /verification/assess-risk | Submits a PR diff and returns a risk score, impacted workflows, and coverage gaps. | CI/CD Pipelines | POST /verification/execute | Triggers an autonomous sandboxed testing agent to verify a change. | Engineering Managers / Release Gates | OQP is designed to be the "Green Contract" between a coding agent and a verification agent. - Context Phase: A developer (or coding agent) begins working on the "Guest Checkout" flow. The agent queries the OQP server ( GET /context/workflows/checkout_flow ) to retrieve the business rules and known edge cases before writing code. - Implementation Phase: The code is written, guided by the semantic context. - Verification Phase: A pull request is opened. The CI/CD pipeline queries the OQP server ( POST /verification/assess-risk ) with the diff. The server maps the syntactic changes to the semantic Knowledge Graph and returns a risk score. - Execution Phase: If the risk is high, or if tests are missing, the pipeline triggers an autonomous verification agent ( POST /verification/execute ). The agent generates tests, runs them, and attempts recovery until the "Green Contract" is satisfied. OQP is designed to be an "open bazaar" of capabilities. No central committee is required to approve new features. Implementers can extend the protocol using reverse-domain naming (e.g., com.yourcompany.custom_capability ). Agents negotiate capabilities during the discovery phase (GET /.well-known/oqp ). If an agent doesn't support an extension, it gracefully degrades to the core specification. We welcome contributions from the community! Please see our Contributing Guide for details on how to propose changes to the specification. The Open QA Protocol specification is licensed under the Apache License 2.0. OQP is shaped by practitioners and researchers across the software quality, AI, and developer tooling communities. We are grateful to the following contributors for their review and endorsement of the specification: | Name | Organization | Role | |---|---|---| | Philip Lew | XBOSoft | Specification Reviewer & Endorser |

Genesis Park 편집팀이 AI를 활용하여 작성한 분석입니다. 원문은 출처 링크를 통해 확인할 수 있습니다.

공유

관련 저널 읽기

전체 보기 →