@openprd/cli 0.1.0 → 0.1.8

package/.openprd/README.md

@@ -1,82 +1,56 @@
-# OpenPrd Workspace
+# OpenPrd 工作区
 
-`.openprd/` is the project-local source of truth for discovery, PRD synthesis, validation, freeze, and handoff.
+简体中文 | [English](./README_EN.md)
 
-## Lifecycle
+`.openprd/` 是项目内用于 discovery、PRD 合成、校验、freeze 和 handoff 的本地事实源。
+
+## 生命周期
 
 ```text
 classify -> interview -> synthesize -> validate -> freeze -> handoff
 ```
 
-## What Lives Here
-
-- `config.yaml` - runtime defaults and workflow policy
-- `schema/` - canonical PRD schema and validation rules
-- `schema/diagram-architecture.schema.yaml` - minimal contract schema for architecture diagrams
-- `schema/diagram-product-flow.schema.yaml` - minimal contract schema for product flow diagrams
-- `templates/` - starter packs and registry
-- `templates/diagram/` - contract templates for diagram artifacts
-- `standards/` - project standards contract and manual templates
-- `engagements/active/` - default active PRD draft, flow, role, and handoff docs
-- `engagements/active/decision-log.md` - durable decision history
-- `engagements/active/open-questions.md` - unresolved questions and discovery gaps
-- `engagements/active/progress.md` - append-only execution progress
-- `engagements/active/verification.md` - freeze and validation evidence
-- `engagements/active/architecture-diagram.html` - reviewable architecture diagram artifact
-- `engagements/active/architecture-diagram.json` - structured diagram contract for iteration
-- `engagements/active/product-flow-diagram.html` - reviewable product flow diagram artifact
-- `engagements/active/product-flow-diagram.json` - structured product flow contract for iteration
-- `artifacts/active/` - HTML + Markdown + patch bundles for human-in-the-loop review and playground workflows
-- `artifacts/archive/` - archived artifact bundles after demand review or handoff
-- `state/` - runtime status, version index, freeze snapshots, session metadata, and execution graph
-- `state/task-graph.json` - workflow/task graph, blockers, and next-ready node
-- `state/events.jsonl` - append-only lifecycle event stream
-- `state/versions/` - immutable version snapshots
-- `sessions/` - per-engagement working state
-- `exports/` - downstream outputs such as OpenSpec handoff bundles
-
-## Template Layers
+## 这里存放什么
+
+- `config.yaml`：运行时默认值与工作流策略。
+- `schema/`：规范 PRD schema 与校验规则。
+- `schema/diagram-architecture.schema.yaml`：架构图最小契约 schema。
+- `schema/diagram-product-flow.schema.yaml`：产品流程图最小契约 schema。
+- `templates/`：模板层与模板注册表。
+- `templates/diagram/`：图表 artifact 契约模板。
+- `standards/`：项目标准契约与说明书模板。
+- `engagements/active/`：当前默认 PRD 草稿、流程、角色与交接文档。
+- `engagements/active/decision-log.md`：可持续追加的决策记录。
+- `engagements/active/open-questions.md`：待解问题与 discovery 缺口。
+- `engagements/active/progress.md`：追加式执行进度。
+- `engagements/active/verification.md`：freeze 与验证证据。
+- `engagements/active/architecture-diagram.html`：可评审架构图 artifact。
+- `engagements/active/architecture-diagram.json`：用于继续迭代的结构化架构图契约。
+- `engagements/active/product-flow-diagram.html`：可评审产品流程图 artifact。
+- `engagements/active/product-flow-diagram.json`：用于继续迭代的结构化产品流程图契约。
+- `artifacts/active/`：人类评审与 playground 使用的 HTML、Markdown 与 patch bundle。
+- `artifacts/archive/`：需求评审或 handoff 之后归档的 artifact bundle。
+- `state/`：运行状态、版本索引、freeze 快照、session 元数据与执行图。
+- `state/task-graph.json`：工作流 / 任务图、阻塞关系与 next-ready 节点。
+- `state/events.jsonl`：追加式生命周期事件流。
+- `state/versions/`：不可变版本快照。
+- `sessions/`：按 engagement 组织的工作状态。
+- `exports/`：下游导出产物，例如 OpenSpec handoff bundle。
+
+## 模板层
 
 ```text
-base -> industry -> company -> project -> session
+core -> company -> industry -> project -> session
 ```
 
-Rules:
-
-- Core PRD fields are fixed.
-- Template layers may add or reorder sections.
-- Template layers may not remove required semantics.
-- The active engagement is the working draft that evolves over time.
-- `classify` sets the product type; `interview` loads discovery prompts.
-- `next` recommends the next action and discovery questions.
-- `synthesize` writes a new versioned PRD snapshot.
-- HTML review and playground artifacts should be paired with a markdown data source and a machine-readable capture patch.
-- `diff` compares version snapshots.
-- `history` lists the version index.
-- Freeze requires validation to pass.
-- `standards` verifies `docs/basic/`, file manual templates, and folder README templates.
-
-## Project Standards
-
-The canonical project baseline docs path is `docs/basic/`.
-
-Required docs:
-
-- `docs/basic/file-structure.md`
-- `docs/basic/app-flow.md`
-- `docs/basic/prd.md`
-- `docs/basic/frontend-guidelines.md`
-- `docs/basic/backend-structure.md`
-- `docs/basic/tech-stack.md`
-
-Use `openprd standards <path> --verify` before reporting a change as ready.
-
-## Supported Product Types
-
-- `consumer`
-- `b2b`
-- `agent`
+- `core`：OpenPrd 自带的基础字段、默认文案和共享规则。
+- `company`：团队或公司范围的共用术语、流程和评审偏好。
+- `industry`：行业特有字段、约束和验证重点。
+- `project`：单项目特有的补充规则、前提和交付方式。
+- `session`：单次协作临时覆盖，不改变长期事实源。
 
-## Handoff Principle
+## 日常规则
 
-The goal is not only to write a PRD. The goal is to keep a PRD alive as the product evolves, and to export a stable handoff package when execution starts.
+- 优先更新 `docs/basic/`，不要把长期说明堆回 `AGENTS.md`。
+- 新增工作流或导出物时，同步检查 `src/workspace-core.js`、`src/openprd.js` 和对应模板是否一致。
+- 对外确认前，优先运行 `openprd standards . --verify`、`openprd quality . --verify` 与 `openprd run . --verify`。

package/.openprd/README_EN.md

@@ -0,0 +1,84 @@
+# OpenPrd Workspace
+
+[简体中文](./README.md) | English
+
+`.openprd/` is the project-local source of truth for discovery, PRD synthesis, validation, freeze, and handoff.
+
+## Lifecycle
+
+```text
+classify -> interview -> synthesize -> validate -> freeze -> handoff
+```
+
+## What Lives Here
+
+- `config.yaml` - runtime defaults and workflow policy
+- `schema/` - canonical PRD schema and validation rules
+- `schema/diagram-architecture.schema.yaml` - minimal contract schema for architecture diagrams
+- `schema/diagram-product-flow.schema.yaml` - minimal contract schema for product flow diagrams
+- `templates/` - starter packs and registry
+- `templates/diagram/` - contract templates for diagram artifacts
+- `standards/` - project standards contract and manual templates
+- `engagements/active/` - default active PRD draft, flow, role, and handoff docs
+- `engagements/active/decision-log.md` - durable decision history
+- `engagements/active/open-questions.md` - unresolved questions and discovery gaps
+- `engagements/active/progress.md` - append-only execution progress
+- `engagements/active/verification.md` - freeze and validation evidence
+- `engagements/active/architecture-diagram.html` - reviewable architecture diagram artifact
+- `engagements/active/architecture-diagram.json` - structured diagram contract for iteration
+- `engagements/active/product-flow-diagram.html` - reviewable product flow diagram artifact
+- `engagements/active/product-flow-diagram.json` - structured product flow contract for iteration
+- `artifacts/active/` - HTML + Markdown + patch bundles for human-in-the-loop review and playground workflows
+- `artifacts/archive/` - archived artifact bundles after demand review or handoff
+- `state/` - runtime status, version index, freeze snapshots, session metadata, and execution graph
+- `state/task-graph.json` - workflow/task graph, blockers, and next-ready node
+- `state/events.jsonl` - append-only lifecycle event stream
+- `state/versions/` - immutable version snapshots
+- `sessions/` - per-engagement working state
+- `exports/` - downstream outputs such as OpenSpec handoff bundles
+
+## Template Layers
+
+```text
+base -> industry -> company -> project -> session
+```
+
+Rules:
+
+- Core PRD fields are fixed.
+- Template layers may add or reorder sections.
+- Template layers may not remove required semantics.
+- The active engagement is the working draft that evolves over time.
+- `classify` sets the product type; `interview` loads discovery prompts.
+- `next` recommends the next action and discovery questions.
+- `synthesize` writes a new versioned PRD snapshot.
+- HTML review and playground artifacts should be paired with a markdown data source and a machine-readable capture patch.
+- `diff` compares version snapshots.
+- `history` lists the version index.
+- Freeze requires validation to pass.
+- `standards` verifies `docs/basic/`, file manual templates, and folder README templates.
+
+## Project Standards
+
+The canonical project baseline docs path is `docs/basic/`.
+
+Required docs:
+
+- `docs/basic/file-structure.md`
+- `docs/basic/app-flow.md`
+- `docs/basic/prd.md`
+- `docs/basic/frontend-guidelines.md`
+- `docs/basic/backend-structure.md`
+- `docs/basic/tech-stack.md`
+
+Use `openprd standards <path> --verify` before reporting a change as ready.
+
+## Supported Product Types
+
+- `consumer`
+- `b2b`
+- `agent`
+
+## Handoff Principle
+
+The goal is not only to write a PRD. The goal is to keep a PRD alive as the product evolves, and to export a stable handoff package when execution starts.

package/.openprd/benchmarks/index.md

@@ -4,6 +4,7 @@
 
 - 项目级 approved benchmark 优先于 OpenPrd 内置 Source Map。
 - `inbox/` 里的 candidate 只表示待确认线索，不表示长期最佳实践。
+- 被采纳信源先累计证据，达到阈值后只推荐 approve，不自动晋级。
 - 每次只挑 1-3 个高相关来源；来源目录不是事实来源。
 
 ## Approved Sources
@@ -17,6 +18,9 @@
 - 不适用: 普通 PRD / 产品流程设计；与 CLI 无关的一次性 UI 视觉问题；默认给每个低风险 PR 拉起多 reviewer 并行审查
 - 研究方式: official_page_first
 - 来源: https://milvus.io/ar/blog/ai-code-review-gets-better-when-models-debate-claude-vs-gemini-vs-codex-vs-qwen-vs-minimax.md
+- 规范化信源: milvus-io-ai-code-review-gets-better-when-models-debate-claude-vs-gemini-vs-code
+- 最近 7 天采纳: 0
+- 累计采纳: 0
 - 备注: AI code review / PR review harness；reviewer agreement、交叉验证、误报过滤、hallucination filter、merge recommendation
 - 价值: AI code review / PR review harness；reviewer agreement、交叉验证、误报过滤、hallucination filter、merge recommendation
 
@@ -29,6 +33,9 @@
 - 不适用: 普通 PRD / 产品流程设计；与 CLI 无关的一次性 UI 视觉问题；默认给每个低风险 PR 拉起多 reviewer 并行审查
 - 研究方式: official_page_first
 - 来源: https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/
+- 规范化信源: nolanlawson-com-using-ai-to-write-better-code-more-slowly
+- 最近 7 天采纳: 0
+- 累计采纳: 0
 - 备注: AI code review / PR review harness；独立审查、主代理先不站队、汇总后再验证、critical/high/medium/low 分级、merge recommendation
 - 价值: AI code review / PR review harness；独立审查、主代理先不站队、汇总后再验证、critical/high/medium/low 分级、merge recommendation
 

package/.openprd/benchmarks/sources.yaml

@@ -1,12 +1,13 @@
 version: 1
 schema: openprd.benchmarks.v1
-updatedAt: 2026-05-28 11:30:07
+updatedAt: 2026-06-01 11:56:05
 sources:
   - id: milvus-io-ai-code-review-gets-better-when-models-debate-claude-vs-gemini-vs-code
     title: milvus.io/ai-code-review-gets-better-when-models-debate-claude-vs-gemini-vs-codex-vs-qwen-vs-minimax.md
     scope: project
     status: approved
     sourceType: engineering-article
+    sourceKey: null
     url: https://milvus.io/ar/blog/ai-code-review-gets-better-when-models-debate-claude-vs-gemini-vs-codex-vs-qwen-vs-minimax.md
     path: null
     repo: null
@@ -25,14 +26,25 @@ sources:
       recommendation
     value: AI code review / PR review harness；reviewer agreement、交叉验证、误报过滤、hallucination filter、merge
       recommendation
+    adoptedCount: 0
+    lastUsedAt: null
+    evidence: []
+    recentAdoptedCount: 0
+    promotion:
+      threshold: 3
+      windowDays: 7
+      recommended: false
+      recommendedAt: null
+      approveCommand: null
     addedAt: 2026-05-28 11:01:00
     approvedAt: 2026-05-28 11:01:06
-    lastVerified: 2026-05-28 11:30:06
+    lastVerified: 2026-06-01 11:56:02
   - id: nolanlawson-com-using-ai-to-write-better-code-more-slowly
     title: nolanlawson.com/using-ai-to-write-better-code-more-slowly
     scope: project
     status: approved
     sourceType: web
+    sourceKey: null
     url: https://nolanlawson.com/2026/05/25/using-ai-to-write-better-code-more-slowly/
     path: null
     repo: null
@@ -51,6 +63,16 @@ sources:
       recommendation
     value: AI code review / PR review harness；独立审查、主代理先不站队、汇总后再验证、critical/high/medium/low 分级、merge
       recommendation
+    adoptedCount: 0
+    lastUsedAt: null
+    evidence: []
+    recentAdoptedCount: 0
+    promotion:
+      threshold: 3
+      windowDays: 7
+      recommended: false
+      recommendedAt: null
+      approveCommand: null
     addedAt: 2026-05-28 11:00:52
     approvedAt: 2026-05-28 11:01:06
-    lastVerified: 2026-05-28 11:30:07
+    lastVerified: 2026-06-01 11:56:05

package/.openprd/discovery/config.json

@@ -1,5 +1,5 @@
 {
-  "activeChange": "verify-boundaries-and-historical-refresh",
+  "activeChange": "adaptive-project-framing",
   "taskSharding": {
     "maxItemsPerFile": 25,
     "handoffRequired": true,
@@ -14,7 +14,21 @@
     ],
     "optional": [
       "deps",
-      "type"
+      "type",
+      "test-layer",
+      "test-size",
+      "test-scope",
+      "evidence",
+      "evidence-plan",
+      "upgrade-reason",
+      "waiver",
+      "waiver-reason",
+      "execution-mode",
+      "parallel-group",
+      "write-scope",
+      "owner-role",
+      "local-verify",
+      "integration-owner"
     ],
     "dependencyOrder": "dependencies must appear before dependents"
   }

package/.openprd/engagements/active/flows.md

@@ -2,29 +2,34 @@
 
 ## 主流程
 
-- 维护者运行 openprd run . --context 或 openprd run . --verify 时，只看到当前项目主验证状态；仓库中存在 research、toolkit-sources、marketplace-candidates 等参考目录时，默认不会被当成本项目源码说明书缺口直接淹没结果；维护者可以继续按需显式 classify-external。
+- 用户第一次提需求时，OpenPrd 先输出一页项目画像：用户群体、产品形态、第一版先做、先不做、不能破坏、技术落点和风险探针
+- 如果用户表达已经足够清楚，Agent 直接用系统归纳的画像摘要请用户确认
+- 如果用户表达模糊，Agent 先追问少量高价值问题，必要时给 2 到 3 个方向或行业原型供用户选择
+- 画像确认后再进入后续 PRD、review、change 和 tasks 流程
 
 ## Mermaid 流程图
 
 ```mermaid
 flowchart LR
-  entry["入口触发<br/>维护者运行 openprd run . --context 或 openprd run . --verify 时，只看到当前项…"]
-  experience["产品内步骤<br/>修复 OpenPrd verify 边界，避免外部参考目录和 reference discovery 污染主验证，并刷新历史项…"]
-  decision{"决策点<br/>边界情况仍需澄清"}
-  success(["成功结果<br/>存在 reference discovery 时 openprd run . --context 不再默认展示它"])
-  failure[["失败与恢复<br/>reference discovery 仍然混入 run context 或 run verify"]]
-  entry -->|"维护者运行 openprd run . --context 或 openprd r…"| experience
-  experience -->|"系统处理请求"| decision
-  decision -->|"让主验证只关注当前项目源码和当前实现门禁"| success
-  decision -.->|"reference discovery 仍然混入 run context 或 ru…"| failure
+  entry["入口触发<br/>用户第一次提需求时，OpenPrd 先输出一页项目画像：用户群体、产品形态、第一版先做、先不做、不能破坏、技术落点和风险探针"]
+  experience["产品内步骤<br/>如果用户表达已经足够清楚，Agent 直接用系统归纳的画像摘要请用户确认"]
+  decision{"决策点<br/>用户一开始只有一句模糊想法，需要先给候选方向而不是硬追问技术细节"}
+  success(["成功结果<br/>首轮 clarify 能输出结构化的项目画像摘要，并包含用户群体、产品形态、第一版先做、先不做和不能破坏项"])
+  failure[["失败与恢复<br/>入口过重，导致 Vibe Coding 用户在第一次提需时无法回答"]]
+  entry -->|"用户第一次提需求时，OpenPrd 先输出一页项目画像：用户群体、产品形态、第一版…"| experience
+  experience -->|"如果用户表达已经足够清楚，Agent 直接用系统归纳的画像摘要请用户确认"| decision
+  decision -->|"在第一次提需时先形成一页轻量项目画像，而不是直接跳进局部需求修改"| success
+  decision -.->|"入口过重，导致 Vibe Coding 用户在第一次提需时无法回答"| failure
 ```
 
 ## 边界情况
 
-- 待补充
+- 用户一开始只有一句模糊想法，需要先给候选方向而不是硬追问技术细节
+- 已有项目上下文时应优先沿用已知事实，而不是把所有画像问题重问一遍
+- 局部 copy、样式或小交互修正不应被错误升级成重画像流程
 
 ## 失败模式
 
-- reference discovery 仍然混入 run context 或 run verify
-- standards verify 继续被 research 或 toolkit-sources 之类参考目录的说明书缺口淹没
-- fleet 更新后历史项目仍保留旧的 verify 噪音行为
+- 入口过重，导致 Vibe Coding 用户在第一次提需时无法回答
+- 入口过轻，导致 Agent 没有先确认项目边界就直接落地
+- 错误复用旧 active change 或旧 PRD 上下文，导致实现偏题

package/.openprd/engagements/active/handoff.md

@@ -1,9 +1,16 @@
 # 交接
 
-- 版本: v0001
+- 版本: v0005
 - 产品类型: agent
 - 模板包: agent
-- Digest: dc988d7a94c07d2a8070968a71ddacffb374eaeee12aa85eb93231e06dd7b077
+- Digest: 7f0e2d2730982a2ae96647d2377f6204e54f8adadf1b18b05f4bc7167046c575
 - 负责人: Codex
-- 下一步: 评审已生成的 PRD，并准备交接。
-- 目标系统: OpenSpec
+- 下一步: 生成新的 change 和 tasks，并按首轮项目画像流程修改 workflow、模板、文档和测试。
+- 目标系统: OpenPrd
+
+## 变化摘要
+
+- 新增：clarify 生成的 intake-reflection 应包含首轮项目画像和风险探针
+- 新增：inline clarification 应显式展示适用对象、产品形态、第一版先做、先不做、不能破坏和技术落点
+- 新增：冷启动 kickoff 问题改成画像导向问题，而不是抽象的通用提问
+- 新增：类型专项 intake 模板改成更贴近产品与业务语言的提问方式