阶段化 AI 开发 Harness — 从模糊需求到交付的端到端 Claude Code 插件
npx claudepluginhub luagam/stage-harness自治模式:自动推进全流程(仅低/中风险 feature 适用)
ShipSpec → deep-plan 桥接(将规格产物合并为统一输入格式)
执行需求澄清全流程(含 domain-scout 前置 + 并行分析 + 决策分类 + 中断预算)
交付沉淀(发布议会 + 交付包 + 问题模式沉淀 + 候选 Skill 挖掘)
问题修复(针对 VERIFY REJECTED 的问题循环修复 + 重新审查)
JIT 即时诊断与补丁工作流:分析刚才的运行偏差,生成候选补丁,支持应用与回滚
实施计划生成(承载面调研 + task图谱 + coverage matrix + 计划议会)
审查与验收(技术review + spec compliance + 安全审查 + 对抗式补盲 + 验收议会)
规格定义(ShipSpec /feature-planning + 轻议会审查)
从模糊需求启动 stage-harness 全流程(PROJECT_PROFILE → CLARIFY → SPEC → PLAN)
查看所有 epic 的阶段进度、中断预算消耗、运行时健康度
开发执行(Worker 循环:re-anchor→preflight→TDD→smoke→commit+receipt)
CLARIFY specialist — stress-tests assumptions, surfaces gaps, and generates adversarial questions
代码审查 reviewer。审查代码质量、风格、可维护性、错误处理和 spec compliance。由 council/SKILL.md 并行调度。
PLAN stage scout — inventories configuration, environment variables, feature flags, and deployment constraints
CLARIFY specialist — investigates ambiguous requirements by deep-reading relevant code and producing clarification memos
PLAN stage scout — maps external dependencies, package versions, and import chains relevant to the epic
PLAN stage scout — extracts architecture patterns, interface contracts, and design decisions from existing codebase
PLAN stage scout — reads all documentation artifacts (PRD, SDD, TASKS.md, clarify-summary) to extract intent and constraints
CLARIFY Step 0 — product/domain framing before codebase impact analysis; no code reads
CLARIFY specialist — scans codebase to identify surfaces affected by the epic and assess blast radius
CLARIFY stage Lead Orchestrator — coordinates multi-role analysis, owns Decision Bundle, enforces Interrupt Budget
逻辑审查 reviewer。验证业务逻辑正确性、状态机完整性、边界条件处理。由 council/SKILL.md 并行调度。
计划审查 reviewer。审查实施计划的覆盖性、任务切分、依赖路径、测试策略。由 council/SKILL.md 并行调度。
CLARIFY specialist — maps requirements to specific codebase surfaces and file paths
质量审计 agent。全局质量检查,覆盖代码质量、测试覆盖率、文档完整性。可在任意阶段调用。
发布审查 reviewer。评估交付完整性、安全签署、文档质量、学习治理。由发布议会并行调度。
PLAN stage scout — maps the repository structure and identifies module boundaries, entry points, and key files
CLARIFY specialist — decomposes epic goals into structured functional requirements
运行时审计 agent。检查实现与规格的对齐情况,识别漂移。由 review/SKILL.md 调度。
CLARIFY specialist — expands high-risk semantic signals into structured scenario candidates
安全审查 reviewer。检查 OWASP Top 10、认证/授权、数据安全问题。由 council/SKILL.md 并行调度。
技能挖掘 agent。分析运行记录,提取可复用模式,生成候选 Skill。调度时机:DONE 阶段由 harness-done 调用。
PLAN stage scout — finds key symbols (functions, classes, types) that the epic will extend or depend on
JIT 进化观测 agent。读取 execution trace 与运行快照,重建事故窗口,生成 candidate patch 草稿。由 /harness:patch 命令或 harnessctl patch diagnose 触发。
测试审查 reviewer。评估测试覆盖率、测试质量、TDD 纪律遵守情况。由 council/SKILL.md 并行调度。
任务执行 worker。按 5 Phase 循环实现单个 task(re-anchor→preflight→implement→smoke→commit+receipt)。由主会话通过 Task 工具调度。
Autonomous execution mode — runs CLARIFY → SPEC → PLAN → EXECUTE → VERIFY without user interrupts (low/medium risk only).
ShipSpec → deep-plan 桥接技能。将 ShipSpec 产物(PRD、SDD、TASKS、unknowns-ledger)合并为 deep-plan 可接受的统一输入格式(bridge-spec.md)。
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
Compress all decisions from CLARIFY into a minimal user interrupt set.
Identify all codebase surfaces affected by an incoming change request.
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
Persist, retrieve, and evolve harness knowledge across sessions and epics.
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
Detect and describe the project's profile from codebase signals.
项目承载面缩圈 — 在产品/领域影响扫描之后,明确后续需要重点分析的具体落点。
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
在执行任何 `harnessctl` 命令前,先解析本地 CLI 路径:
AI-native product management for startups. Transform Claude into an expert PM with competitive research, gap analysis using the WINNING filter, PRD generation, and GitHub Issues integration.
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Intelligent prompt optimization using skill-based architecture. Enriches vague prompts with research-based clarifying questions before Claude Code executes them