{"id":"6590fd6f-cf7d-49f9-92c0-a98aaef42292","entityType":"agent","slug":"clawhub-mtty-ai-opphub","name":"opphub","canonicalUrl":"https://www.xpersona.co/agent/clawhub-mtty-ai-opphub","canonicalPath":"/agent/clawhub-mtty-ai-opphub","generatedAt":"2026-10-10T08:44:56.657Z","source":"CLAWHUB","claimStatus":"UNCLAIMED","verificationTier":"NONE","summary":{"evidence":{"source":"editorial-content","verified":true,"confidence":"high","updatedAt":"2026-10-10T04:28:01.898Z","emptyReason":null},"description":"偶合 OppHub - 把偶合账号接入 OpenClaw，录入公司、查商机、收推送。 Skill: opphub Owner: mtty-ai Summary: 偶合 OppHub - 把偶合账号接入 OpenClaw，录入公司、查商机、收推送。 Tags: latest:5.0.0 Version history: v5.0.0 | 2026-07-27T08:56:48.339Z | user OPC → User 全量重命名: /api/opc/ → /api/user/, token 字段 opc_id → user_id. check-update.js 加 stripPrefix 函数 (v4.0.1 漏定义, 修 plugin cron 'skill 没查到' 错误) Archive index: Archive v5.0.0: 49 files, 223936 bytes Files: _meta.json (125b), bin/opphub (7499b), bin/opphub-check-","descriptionLabel":"Technical summary","evidenceSummary":"Capability contract not published. No trust telemetry is available yet. 1.7K downloads reported by the source. Last updated 10/10/2026.","installCommand":"clawhub skill install s17ampw22fp45rj3rm81m8ssd989v9cx:opphub","sourceUrl":"https://clawhub.ai/mtty-ai/opphub","homepage":"https://clawhub.ai/mtty-ai/skills/opphub","primaryLinks":[{"label":"View on ClawHub","url":"https://clawhub.ai/mtty-ai/opphub","kind":"source"},{"label":"Homepage","url":"https://clawhub.ai/mtty-ai/skills/opphub","kind":"homepage"}],"safetyScore":84,"overallRank":62,"popularityScore":65,"trustScore":null,"claimedByName":null,"isOwner":false,"seoDescription":"偶合 OppHub - 把偶合账号接入 OpenClaw，录入公司、查商机、收推送。 Skill: opphub Owner: mtty-ai Summary: 偶合 OppHub - 把偶合账号接入 OpenClaw，录入公司、查商机、收推送。 Tags: latest:5.0.0 Version history: "},"coverage":{"evidence":{"source":"public-profile","verified":false,"confidence":"medium","updatedAt":"2026-10-10T04:28:01.898Z","emptyReason":null},"protocols":[{"protocol":"OPENCLEW","label":"OpenClaw","status":"self-declared","notes":"Declared in the public agent profile."}],"capabilities":[],"verifiedCount":0,"selfDeclaredCount":1,"capabilityMatrix":{"rows":[{"key":"OPENCLEW","type":"protocol","support":"unknown","confidenceSource":"profile","notes":"Listed on profile"}],"flattenedTokens":"protocol:OPENCLEW|unknown|profile"}},"adoption":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T04:28:01.898Z","emptyReason":null},"stars":null,"forks":null,"downloads":1686,"packageName":null,"latestVersion":"5.0.0","tractionLabel":"1.7K downloads"},"release":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"medium","updatedAt":"2026-10-10T04:28:01.898Z","emptyReason":null},"lastUpdatedAt":"2026-10-10T04:28:01.898Z","lastCrawledAt":"2026-10-10T04:28:01.898Z","lastIndexedAt":null,"nextCrawlAt":"2026-10-11T04:28:01.898Z","lastVerifiedAt":null,"highlights":[{"version":"5.0.0","createdAt":"2026-07-27T08:56:48.339Z","changelog":"OPC → User 全量重命名: /api/opc/ → /api/user/, token 字段 opc_id → user_id. check-update.js 加 stripPrefix 函数 (v4.0.1 漏定义, 修 plugin cron 'skill 没查到' 错误)","fileCount":49,"zipByteSize":223936}]},"execution":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No published capability contract is available yet."},"installCommand":"clawhub skill install s17ampw22fp45rj3rm81m8ssd989v9cx:opphub","setupComplexity":"low","setupSteps":["Setup complexity is LOW. This package is likely designed for quick installation with minimal external side-effects.","Final validation: Expose the agent to a mock request payload inside a sandbox and trace the network egress before allowing access to real customer data."],"contract":{"contractStatus":"missing","authModes":[],"requires":[],"forbidden":[],"supportsMcp":false,"supportsA2a":false,"supportsStreaming":false,"inputSchemaRef":null,"outputSchemaRef":null,"dataRegion":null,"contractUpdatedAt":null,"sourceUpdatedAt":null,"freshnessSeconds":null},"invocationGuide":{"preferredApi":{"snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/trust"},"curlExamples":["curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/snapshot\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/contract\"","curl -s \"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/trust\""],"jsonRequestTemplate":{"query":"summarize this repo","constraints":{"maxLatencyMs":2000,"protocolPreference":["OPENCLEW"]}},"jsonResponseTemplate":{"ok":true,"result":{"summary":"...","confidence":0.9},"meta":{"source":"CLAWHUB","generatedAt":"2026-10-10T08:44:56.655Z"}},"retryPolicy":{"maxAttempts":3,"backoffMs":[500,1500,3500],"retryableConditions":["HTTP_429","HTTP_503","NETWORK_TIMEOUT"]}},"endpoints":{"dossierUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/dossier","snapshotUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/snapshot","contractUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/contract","trustUrl":"https://www.xpersona.co/api/v1/agents/clawhub-mtty-ai-opphub/trust"}},"reliability":{"evidence":{"source":"runtime-metrics","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No trust, reliability, or runtime telemetry is available."},"trust":{"status":"unavailable","handshakeStatus":"UNKNOWN","verificationFreshnessHours":null,"reputationScore":null,"p95LatencyMs":null,"successRate30d":null,"fallbackRate":null,"attempts30d":null,"trustUpdatedAt":null,"trustConfidence":"unknown","sourceUpdatedAt":null,"freshnessSeconds":null},"decisionGuardrails":{"doNotUseIf":["Contract metadata is missing or unavailable for deterministic execution."],"safeUseWhen":[],"riskFlags":["missing_or_unavailable_contract","trust_data_unavailable","schema_references_missing"],"operationalConfidence":"low"},"executionMetrics":{"observedLatencyMsP50":null,"observedLatencyMsP95":null,"estimatedCostUsd":null,"uptime30d":null,"rateLimitRpm":null,"rateLimitBurst":null,"lastVerifiedAt":null,"verificationSource":null},"runtimeMetrics":{"successRate":null,"avgLatencyMs":null,"avgCostUsd":null,"hallucinationRate":null,"retryRate":null,"disputeRate":null,"p50Latency":null,"p95Latency":null,"lastUpdated":null}},"benchmarks":{"evidence":{"source":"no-benchmark-data","verified":false,"confidence":"low","updatedAt":null,"emptyReason":"No benchmark suites or observed failure patterns are available."},"suites":[],"failurePatterns":[]},"artifacts":{"evidence":{"source":"CLAWHUB","verified":false,"confidence":"high","updatedAt":"2026-10-10T04:28:01.898Z","emptyReason":null},"readme":"Skill: opphub\n\nOwner: mtty-ai\n\nSummary: 偶合 OppHub - 把偶合账号接入 OpenClaw，录入公司、查商机、收推送。\n\nTags: latest:5.0.0\n\nVersion history:\n\nv5.0.0 | 2026-07-27T08:56:48.339Z | user\n\nOPC → User 全量重命名: /api/opc/ → /api/user/, token 字段 opc_id → user_id. check-update.js 加 stripPrefix 函数 (v4.0.1 漏定义, 修 plugin cron 'skill 没查到' 错误)\n\nArchive index:\n\nArchive v5.0.0: 49 files, 223936 bytes\n\nFiles: _meta.json (125b), bin/opphub (7499b), bin/opphub-check-update.js (6347b), bin/opphub-configure.js (7552b), bin/opphub-cron-setup.js (10591b), bin/opphub-find-peers.js (4613b), bin/opphub-knowledge-add.js (5486b), bin/opphub-knowledge-autofill.js (1140b), bin/opphub-knowledge-card.js (15405b), bin/opphub-knowledge-discover.js (15498b), bin/opphub-knowledge-ingest-batch.js (8814b), bin/opphub-knowledge-match.js (13515b), bin/opphub-knowledge-relate.js (20077b), bin/opphub-knowledge-search.js (4394b), bin/opphub-knowledge-status.js (4246b), bin/opphub-knowledge-submit.js (11522b), bin/opphub-oauth-login.js (25839b), bin/opphub-plugin-check.js (2355b), bin/opphub-token-refresh.js (4067b), CHANGELOG.md (6362b), docs/runtime-channel-renderer-v31-design.md (10948b), docs/server-schema-v31-design.md (3745b), docs/server-schema-v32-design.md (16162b), docs/superpowers/plans/2026-07-23-phase1-nav-rewrite.md (32098b), docs/superpowers/plans/2026-07-23-phase2-confirm-and-group.md (24529b), docs/superpowers/plans/2026-07-23-phase3-modal-interactions.md (23297b), docs/superpowers/plans/2026-07-23-phase4-bug-fixes.md (27089b), docs/superpowers/plans/2026-07-23-phase5-v1-conflicts-radar.md (7761b), docs/superpowers/plans/2026-07-24-batch1-push-pipeline.md (25710b), docs/superpowers/plans/2026-07-24-dashboard-optimize.md (27740b), docs/superpowers/specs/2026-07-23-opphub-product-design.md (19197b), docs/superpowers/specs/2026-07-24-opphub-dashboard-optimize-design.md (12998b), docs/superpowers/specs/2026-07-24-opphub-discover-to-order-flow-design.md (46430b), docs/v3.1-architecture.md (30966b), flow/registration.md (2512b), INTERNAL.md (9831b), lib/error-format.js (5402b), lib/intent-card.js (6550b), lib/opphub-plugin-client.js (9380b), lib/opphub-server-client.js (1770b), memory/2026-07-24.md (11335b), package.json (863b), README.md (181b), scripts/check-version.js (2737b), skill-card.md (2776b), SKILL.md (4244b), tests/e2e-verify.js (16596b), tests/mock-knowledge-server.js (4038b), tests/unit-knowledge-card.js (10639b)\n\nFile v5.0.0:SKILL.md\n\n---\nname: opphub\nversion: 5.0.0\ndescription: 偶合 OppHub - 把偶合账号接入 OpenClaw，录入公司、查商机、收推送。\nauthor: mtty-ai\nhomepage: https://github.com/mtty-ai/opphub-skill\nentry: bin/opphub\ndefaultLocale: zh-CN\nrequires:\n bins:\n - node>=18\n - openclaw\n - jq\n - curl\n - security\n - openssl\n - xdg-open\n env: []\n platform:\n darwin: full\n linux: full\n windows: none\nmetadata:\n api: https://api.opphub.ruiplus.cn\n deviceFlow:\n authorize: https://api.opphub.ruiplus.cn/api/oauth/device/code\n token: https://api.opphub.ruiplus.cn/api/oauth/device/token\n userinfo: https://api.opphub.ruiplus.cn/api/oauth/userinfo\n refreshToken:\n endpoint: https://api.opphub.ruiplus.cn/api/oauth/token\n client_id: opphub-plugin\n scope: profile ws:read ws:write\n---\n\n# 偶合 OppHub\n\n把偶合账号接入 OpenClaw，录入公司、查商机、收推送。\n\n支持 3 种触发方式：\n- IM 群聊 / 私聊 @bot（飞书 / 微信 / 钉钉 / Telegram 等）\n- OpenClaw 后台直接调用\n- cron 定时任务\n\n具体 IM 取决于 OpenClaw runtime 已装的 channel plugin。\n\n---\n\n## 一句话介绍\n\n录入公司、查商机、收推送，3 件事 bot 都替你跑。\n\n---\n\n## 安装\n\n```bash\nopenclaw skills install @mtty-ai/opphub\nopenclaw plugins install clawhub:@mtty-ai/opphub\n```\n\n两条都要装：\n\n- `opphub` skill：在 IM 里跟 bot 对话\n- `opphub` plugin：装好后 bot 收得到推送，登录态写 macOS Keychain\n\n装完自动重启 gateway。\n\n依赖 runtime 自动校验，不缺。\n\n---\n\n## 五步上手\n\n### 1. 注册偶合账号\n\n去 ，邮箱/手机号 + 6 位验证码。\n\n### 2. 验证 plugin 装好\n\n```bash\nopphub plugin-check --json\n```\n\n返 `installed: true` 就 OK。没装好跑 `openclaw plugins install clawhub:@mtty-ai/opphub` 重装。\n\n### 3. @bot 说\"偶合登录\"\n\nbot 走 OAuth device flow：\n- bot 私聊发你激活链接（不会在群里贴验证码）\n- 你打开链接点同意\n- 登录态自动写 Keychain，下次不用再登\n\n### 4. @bot 说\"偶合配置\"\n\nbot 列出你本机已装的 IM 通道（飞书/微信/钉钉/Telegram 等），你选一个作**默认推送通道**。后续商机推送、cron 提醒都走这个通道。\n\n跳过这步也行——bot 会用 server 端上次选的通道；不放心就先跑一次确认。\n\n### 5. @bot 说\"偶合录入 [公司名]\"\n\nbot 自动 6 步跑完，只问你一次\"是否入库\"：\n\n- 联网查工商 / 业务 / 招聘 / 项目\n- 推断行业（MCN / SaaS / 律所 / 制造）\n- 拆能力卡片 + 上下游\n- 给你 1 张清单确认\n- 批量入库（幂等，重复跑不冲突）\n- 跑上下游匹配，列结果\n\n> 第 4 步拆出能力后, bot **不直接入库**, 先给一张确认清单 (含 4 种 entryType: 我能提供/我想找/我的依赖/同行关系).\n> 用户回复:\n> - 「确认」 → bot 调 `knowledge-submit --confirm` 入库\n> - 「删 」 → bot 去掉某条重发清单\n> - 「重抽」 → 回到阶段 1 重新联网查\n> - 「改 <字段>」 → 改解析字段后重发清单\n\n---\n\n## 常用命令\n\n| 触发方式 | bot 做的 |\n|---|---|\n| 偶合登录 | OAuth 登录 |\n| 偶合状态 | 看登录态 + 推送通道 + 知识库 |\n| 偶合配置 | 选默认推送通道 |\n| 偶合录入 [公司] | 录入公司能力画像 |\n| 偶合录入关联公司 + 合同 xls | 解析合同拆上下游 |\n\n**触发方式**：群里 @bot、私聊 bot、OpenClaw 后台调用、cron 定时都可以。\n\n---\n\n## 常见问题\n\n**没收到推送？**\n确认 plugin 装好：`opphub plugin-check --json`。没装就跑 `openclaw plugins install clawhub:@mtty-ai/opphub`。\n\n**token 过期了？**\n@bot 说\"偶合登录\"重新走一遍。\n\n**浏览器没弹？**\n复制 bot 私聊里的链接手动打开。\n\n**录入撞同行业了？**\nbot 会弹卡片让你选行业。\n\n**录入冲突了？**\nbot 会列新旧版本让你选保留哪个。\n\n---\n\n## 卸载\n\n```bash\nopenclaw plugins uninstall clawhub:@mtty-ai/opphub\nopenclaw skills uninstall @mtty-ai/opphub\n```\n\nKeychain 里的登录态要手动清：macOS Keychain 搜 `opphub:default` 删除。\n\n---\n\n## 反馈\n\nissue: https://github.com/mtty-ai/opphub-skill/issues\n\nFile v5.0.0:README.md\n\n# 偶合 OppHub · OpenClaw Skill\n\nOpenClaw bot skill · device flow 登录 · 与 opphub-plugin 共 Keychain\n\n详见 SKILL.md (本仓 bot skill 模式说明, 由 OpenClaw 加载)\n\nFile v5.0.0:_meta.json\n\n{\n \"ownerId\": \"kn7ethqpnt6p4vba8zp33jx2hh89vmkc\",\n \"slug\": \"opphub\",\n \"version\": \"5.0.0\",\n \"publishedAt\": 1785142608339\n}\n\nFile v5.0.0:CHANGELOG.md\n\n# Changelog\n\n公开版仅保留用户可感知的功能变更。内部迭代历史不入仓。\n\n## v5.0.0 (2026-07-24) ⚠️ BREAKING\n\n### 加 (New)\n- **架构重写**：skill 不再写死任何业务词 / 行业模板 / 平台字典\n - 删除 `INDUSTRY_TEMPLATES` (mcn/saas/law/mfg 4 个行业的所有 abilities/upstream/downstream/peer 列表)\n - 删除 `INDUSTRY_SIGNALS` (token 频率猜行业)\n - 删除 `PLATFORM_TERMS` / `GENERIC_TERMS` / `TAIL_GENERIC` / `MID_GENERIC` (所有 stop word 词表)\n - skill 现在认任何中文公司 rawText：律所 / 餐饮 / 制造 / 数字营销 / … 都用同一套代码\n- **LLM 任务包**：每张卡 + JSON 输出都含\n - `evidenceCandidates` (string[30]): n-gram 频次排名 + 长度加分\n - `evidenceAnswerPrompt` (string): 单卡 LLM 选择规则 + JSON 输出格式\n - `llmInstruction` (string): 顶层 LLM 一次性任务流 (read → pick → rewrite → submit)\n - 占位符 `(证据词: <待 LLM 填写>)` 在 rawText 里等 LLM 填\n- **heading 驱动拆卡**:\n - rawText `## 业务描述` 节下 `### 维度名` 直接生成 ability 卡片\n - `### 上游依赖 / 下游客户 / 同业联盟` 子标题自动判定 type (无需模板)\n - 其它行业 (律所/餐饮等) 写啥拆啥: `### 公司诉讼` / `### 火锅堂食` 都能识别\n\n### 改 (Changed)\n- 数据流从 \"skill 输出 → 用户确认 → LLM 再审 → submit\" 多步，简化为 \"skill 输出 LLM 任务包 → LLM 一次性完成 → submit\"\n- 无行业推断 (token 频率猜), 行业归属由 LLM 在对话里判断并传入\n\n### 弃用 (Deprecated)\n- 任何用户写死的维度词表 (e.g. 老 v4.x 模板里硬编码的\"达人营销\"/\"短视频内容制作\")\n\n### 测 (Test)\n- 11 个单元测试: 通用性守卫 (源码不许有业务词) / heading 拆 / n-gram 候选 / 占位符 / LLM 提示\n- 用 `npm test` 跑\n\n## v4.0.10 (2026-07-24)\n\n### 加 (New)\n- **knowledge-card evidence 提取加固** (extractEvidenceFromDesc):\n - 长度上限 6→8, 让 \"短视频内容制作\"/\"店铺代运营\" 等 7-8 字短语能拆出\n - 新增 PLATFORM_TERMS 优先扫 (抖音/快手/小红书/B站/视频号 等 14 平台)\n - 解决 \"精选小红书博主合作\" 短语被 6 字限制吞掉的 case, 拆出关键词 \"小红书\"\n - 尾部泛词剥离 (等等/行业/领域/场景/赛道/玩法/业务/服务/平台/资源)\n - 列表前缀剥离 (包括/比如/例如/以及)\n - 人称前缀剥除 (我们/你/他/她/它)\n\n### 测 (Test)\n- 单元测试 9 → 12 (新增 3 个边缘 case 覆盖人称/列表前缀/跨公司对比)\n- `npm test` 全部 12 测试通过 (~600ms)\n\n## v4.0.9 (2026-07-24)\n\n### 修 (Fixed)\n- **knowledge-card 证据词提取**: 之前用模板写死的 `purpose` 字符串当证据词源, 不同公司卡片雷达词趋同 (e.g. 都显示 \"KOL 投放\"/\"媒介代理\")。现在从 rawText 自然语言描述里挖真实关键词 (拆标点 + 滤停用前缀 + 限 2-6 字名词性片段), 两家公司 evidence 完全独立。\n- **证据词格式统一**: 修复 `buildCard()` 漏传 `evidenceList` 参数 bug。所有卡片现在统一输出 `(证据词: kw1, kw2, ...)` 格式, 去掉旧 `(证据: rawText 包含 \"${dim}\")` 兜底文案 (前端 regex 解析不到)。\n- **雷达不再有杂质**: 前端雷达原显示 \"蓝色光标\"/\"宝马等\"等公司名/团队背景, 现在都是真实能力词 (达人筛选/店铺搭建/活动策划等)。\n- **extractParsedFields 城市检测**: 不再被同业联盟里 `无忧传媒(北京...)` 这类字符串污染。限定到「地址字段 || 工商信息节」内查找城市。\n- **knowledge-discover 骨架校验**: 加 `SKELETON_PATTERNS` 检测, rawText 含 `(名称 / 法人 / 注册资本...)` 等填空内容时直接拒绝 (skeleton_unfilled), 强制 LLM 用 web_search 搜真实数据。\n- **knowledge-discover 搜索指令内嵌**: `--name` 模式输出 `_explicitInstructions` 字段, 列出每节搜索命令 + 禁止传空骨架警告。\n\n### 加 (New)\n- 9 个单元测试覆盖 `extractEvidenceFromDesc` / `extractDimDesc` / `extractParsedFields` 三大核心逻辑。用 `npm test` 跑 (`tests/unit-knowledge-card.js`)。\n\n### 测 (Test)\n- `node --test tests/unit-knowledge-card.js` 全部 9 测试通过 (~450ms)\n- `npm test` 端到端验证脚本 (`tests/e2e-verify.js`) 不动\n\n## v4.0.0 (2026-07-22)\n\n### 加 (New)\n\n- OAuth Device Flow 登录 + token 自动写 macOS Keychain / Linux AES-256-GCM 加密文件\n- 6 步闭环引导: 注册 → 登录 → 装 plugin → 配默认通道 → 知识库 → 商机匹配\n- 开放式知识库录入: discover (生成 rawText 骨架) + card (按行业拆能力卡片) + submit (idempotent 入库 + 冲突检测)\n- 关联公司录入: relate (xls/html 合同解析, 拆 top 客户 / 供应商 / 类别聚合)\n- bot turn 调用 LLM 工具 (web_search / web_fetch / image / pdf / memory_search / wiki_search) 作为数据源\n- 错误结构化: 统一返 `{ ok, stage, code, retryable, hint, traceId }`\n\n### 改 (Changed)\n\n- 通道元数据真源 = OpenClaw runtime (读 `openclaw channels list --json`)\n- 默认通道真源 = opphub-server (PATCH `/api/user/channels/default`)\n- token 写入方 = opphub-plugin (Keychain 单写入, skill 不双源)\n- OAuth device flow start 30 秒内复用 (避免重复申请 device_code)\n- ingest-batch 用 mkdtemp 每次独立目录 (避免并发竞态)\n\n### 弃用 (Deprecated)\n\n- `bin/opphub-knowledge-autofill.js` (DEPRECATED 2026-07-22) — 读本机敏感源违反产品红线\n- `bin/opphub-cron-setup.js` (DEPRECATED 2026-07-22) — v4 cron 改由 plugin 维护\n\n### 修 (Fixed)\n\n- knowledge-card 行业歧义时双 JSON 输出\n- token refresh 返字符串 (无法读 expires_at) 改为返完整对象\n- OAuth URL/path 改 spawn argv 避免 shell 注入\n- knowledge-relate 必需列校验 + 金额解析增强 (逗号/中文/括号负数/货币符号)\n- knowledge-relate 二进制 xls 检测 (拒绝 OLE2 二进制)\n\n### 安全 (Security)\n\n- skill 不再读 token / IM channel / openclaw.json / outbox / MEMORY (隐私红线)\n- skill 不再硬编码真实 user open_id (cron fallback 改为返 null)\n- 错误响应含 traceId 便于排查\n\n### 已知限制\n\n- cron 操作 (create/list/rm) 需 plugin 已装 (peer 由 plugin 维护)\n- knowledge POST 路径 v2 (opphub-web server 端合并待定)\n- refresh endpoint 探测待 plugin client 暴露\n\nFile v5.0.0:docs/runtime-channel-renderer-v31-design.md\n\n# opphub-skill v3.1 · OpenClaw runtime · channel-renderer 渲染层架构\n\n> \"原始内容归原始内容, 你得 写一个 飞书 card 的渲染层 当然未来也有其他通道需要特定的渲染\"\n\n> **含义**: runtime 渲染层是 OpenClaw 平台活, **不在 opphub skill v3.1 范围**\n> **本文状态**: 独立存档, 等 OpenClaw runtime 团队接\n\n---\n\n## 0. 文档归属\n\n| 项 | 归属 |\n|---|---|\n| **本档**(运行时渲染层) | OpenClaw runtime 团队 |\n| opphub skill v3.1 doc | skill 层 (我开干的范围) |\n| opphub-server schema v3.1 | server 层 (单独存档) |\n\n---\n\n\n> \"原始内容归原始内容, 你得 写一个 飞书 card 的渲染层 当然未来也有其他通道需要特定的渲染\"\n\n\n> \"只到 skill 开放完, 没开发完, 不会会话\"\n\n**含义**:\n- skill 层只发 \"IntentMessage 原始内容\"(channel-agnostic)\n- skill 层不拼飞书 card / Block Kit / InlineKeyboard\n- runtime 渲染层是 OpenClaw 平台活, **不在 opphub skill v3.1 范围**\n- server schema (`OpcKnowledgeEntry` 表) 也是单独一条线\n\n---\n\n\n| # | 任务 | 改动层 | 工时 | 依赖 |\n|---|---|---|---|---|\n| N10 | Skill 改造: 5 个 bin 都改用 `bot.skillApi.askInteractive/askFreeText/send`, 删硬编码飞书 JSON | Skill | 3h | 无 |\n| N11 | SKILL.md frontmatter: `requires.tools` 加 `minimax__web_search / web_fetch / minimax__understand_image / pdf / memory_search / wiki_search` | Skill | 0.5h | 无 |\n| N12 | OpenClaw runtime 新 API: `skillApi.send/askInteractive/askFreeText/getChannel` + 类型定义 + 8 个 Renderer (FeishuCard / FeishuGroup / WeixinText / SlackBlock / DiscordEmbed / TelegramInline / CliInquirer / WebForm) + FallbackText | Runtime | 8h | 无 (这是 OpenClaw runtime 改造, 不是 skill 改造) |\n| N13 | E2E 多 channel 测试: 飞书 DM / 飞书群 / 微信 iLink / CLI 各跑一遍偶合配置 + 偶合画像 6 步 | 全链路 | 2h | N10 N12 |\n| **N10-N13 合计** | | | **13.5h ≈ 1.7 天** | |\n\n**总工时** v3.1: 19.5h (skill 6 步) + 13.5h (renderer) + 1h (server schema N4-N5) + 1h (server API N5) = **约 35h ≈ 4.5 天**\n\n\n---\n\n\n\n### 关键分工\n\n```\n┌─────────────────────────────────────────────────────┐\n│ Skill / Plugin 层 │\n│ 发 \"原始内容\" (IntentMessage, channel-agnostic) │\n│ - prompt / options / actions / inputHint / links │\n└──────────────────┬──────────────────────────────────┘\n ↓\n┌─────────────────────────────────────────────────────┐\n│ OpenClaw runtime · Content Renderer 层 ★ 这里 ★ │\n│ 每个 channel 一个 renderer: │\n│ · FeishuCardRenderer (interactive card + actions) │\n│ · FeishuGroupRenderer (同上,群 channel) │\n│ · WeixinRenderer (文本 + emoji 协议) │\n│ · SlackRenderer (Block Kit) │\n│ · DiscordRenderer (Embed + Components) │\n│ · TelegramRenderer (InlineKeyboard) │\n│ · CliRenderer (inquirer.select / confirm) │\n│ · WebRenderer (HTML form, 走 opphub-web) │\n│ · FallbackTextRenderer (纯文本兜底) │\n└──────────────────┬──────────────────────────────────┘\n ↓\n Channel (IM / CLI / Web)\n```\n\n### Skill / Plugin API (IntentMessage 原始内容)\n\n```ts\ninterface IntentMessage {\n prompt: string; // 主文案 (\"选默认推送通道\")\n options?: Array<{ // 选项 (pickOne 场景)\n id: string; // 选中时回传 (feishu:pm)\n label: string; // 显示 (\"feishu:pm ⭐\")\n hint?: string; // 副文案 (\"pm bot\")\n isDefault?: boolean;\n }>;\n actions?: Array<{ // 按钮\n id: string; // 唯一 action id\n label: string; // 按钮文字 (\"确认\" / \"🚀 一键装 plugin\")\n style?: 'primary' | 'default' | 'danger';\n payload?: Record; // action 回调带的元数据\n }>;\n inputHint?: string; // 自由文本输入\n links?: Array<{ // 链接按钮 (install 引导)\n text: string;\n url: string;\n }>;\n markdown?: string; // 富文本 (lark_md / mrkdwn / html)\n header?: {\n title: string;\n color?: 'blue' | 'orange' | 'red' | 'green';\n };\n metadata?: Record;\n}\n```\n\n### 8 个 Renderer 渲染示例\n\n| channel | askInteractive / askFreeText 渲染 |\n|---|---|\n| 飞书 DM / 群 | interactive card · actions + button + url link |\n| 微信 iLink | 文本 + emoji + \"回复按钮序号\" + 文本链接 |\n| Slack | Block Kit · actions block + button + url |\n| Discord | Embed + Components (buttons + link) |\n| Telegram | InlineKeyboard + URL 按钮 |\n| CLI / tty | inquirer.select / confirm / input |\n| web (opphub-web) | HTML form (走 /admin/devices/:opcId) |\n| FallbackText | 纯文本兜底 (不识别的 channel) |\n\n### OpenClaw runtime skillApi 新增 API\n\n```ts\ndeclare global {\n interface SkillApi {\n /** 单向通知 (无操作) */\n send(intent: IntentMessage): Promise;\n\n /** 交互意图 - 等用户回复 (pickOne / actions) */\n askInteractive(intent: IntentMessage & {\n options?: Option[];\n actions?: Action[];\n }): Promise;\n\n /** 自由文本输入 */\n askFreeText(intent: IntentMessage & {\n inputHint: string;\n }): Promise;\n\n /** 当前 channel hint */\n getChannel(): {\n type: 'feishu' | 'weixin' | 'slack' | 'discord' | 'telegram' | 'cli' | 'web';\n account?: string;\n supportsCards: boolean;\n supportsInlineActions: boolean;\n supportsFreeText: boolean;\n };\n }\n}\n```\n\n### Skill ↔ Plugin ↔ Runtime 三者边界 (钉死)\n\n| 层 | 谁做 | 不做 |\n|---|---|---|\n| Skill (per-invocation) | 发 \"IntentMessage 原始内容\" + 调 plugin CLI | ❌ 不拼飞书 card JSON / Block Kit / InlineKeyboard |\n| Plugin (常驻) | 执行 spawn CLI + WS 通讯 + 直读 JSON | ❌ 不接 channel context (没 channelType) |\n| OpenClaw runtime ★ | 把 IntentMessage 转译成 channel 原生渲染 | 不解析业务意图 |\n| Channel | 飞书 OpenAPI / 微信 iLink / Slack / Discord / Telegram / tty / web | — |\n\n\n- ✅ skill 必须调 `bot.skillApi.askInteractive(...)` / `askFreeText(...)` / `send(...)`\n- ❌ skill **永远不写** 硬编码 飞书 card JSON (不是 skill 的活)\n- ❌ skill **永远不写** `feishu_mcp_create_card(...)` 这种飞书 OpenAPI 直调\n- ❌ skill **永远不** 自己 `bot.sendMessage(一段 ad-hoc JSON)`\n- ✅ OpenClaw runtime 8 个 renderer, 未来加 channel 加 renderer 就行 (renderer 由 runtime/插件编写者实现)\n- ❌ skill / plugin **不在 tty** 跑 inquirer (runtime CliRenderer 才跑)\n- ✅ CliRenderer 是 runtime 内置 (SSH/headless fallback)\n\n### Plugin CLI 保留 (再钉一遍)\n\n- ❌ plugin **不接** skill 的 IntentMessage (没 channel context)\n- ❌ plugin CLI 的 inquirer 姿势 **保留** (SSH/headless 用户绕过 skill 直接跑 plugin CLI)\n\n### 渲染示例 (每个 channel 一段)\n\n#### 飞书 (FeishuCardRenderer)\n\n```json\n{\n \"config\": { \"wide_screen_mode\": true },\n \"header\": { \"title\": { \"tag\": \"plain_text\", \"content\": \"选默认推送通道\" }, \"template\": \"blue\" },\n \"elements\": [\n { \"tag\": \"div\", \"text\": { \"tag\": \"lark_md\", \"content\": \"本机已配 5 条有效通道\" } },\n { \"tag\": \"select_static\", \"options\": [\n { \"text\": { \"tag\": \"plain_text\", \"content\": \"feishu:pm\" }, \"value\": \"feishu:pm\" }\n ], \"value\": { \"type\": \"text\", \"text\": \"feishu:pm\" } },\n { \"tag\": \"action\", \"actions\": [\n { \"tag\": \"button\", \"text\": { \"tag\": \"plain_text\", \"content\": \"确认\" }, \"type\": \"primary\", \"value\": { \"action\": \"confirm\" } },\n { \"tag\": \"button\", \"text\": { \"tag\": \"plain_text\", \"content\": \"取消\" }, \"type\": \"danger\", \"value\": { \"action\": \"cancel\" } }\n ]}\n ]\n}\n```\n\n#### 微信 iLink (WeixinRenderer · 文本 + emoji)\n\n```\n📋 偶合 OppHub · 选默认推送通道\n\n本机已配 5 条有效通道:\n 1️⃣ feishu:default 公司主账号\n 2️⃣ feishu:dev dev bot\n 3️⃣ ⭐ feishu:pm pm bot\n 4️⃣ feishu:frontend 前端\n 5️⃣ openclaw-weixin:eat-im-bot 微信 bot\n\n回复 [数字] 选择通道, 或:\n Y → 用默认 (feishu:pm)\n N → 取消\n\n(消息已加密, 2 分钟有效)\n```\n\n#### Telegram (TelegramRenderer · InlineKeyboard)\n\n```json\n{\n \"text\": \"📋 *偶合 OppHub · 选默认推送通道*\\n\\n本机已配 5 条有效通道\",\n \"reply_markup\": {\n \"inline_keyboard\": [\n [{ \"text\": \"feishu:default\", \"callback_data\": \"pick:feishu:default\" }],\n [{ \"text\": \"⭐ feishu:pm\", \"callback_data\": \"pick:feishu:pm\" }],\n [{ \"text\": \"feishu:dev\", \"callback_data\": \"pick:feishu:dev\" }],\n [{ \"text\": \"✅ 确认\", \"callback_data\": \"confirm\" }, { \"text\": \"❌ 取消\", \"callback_data\": \"cancel\" }]\n ]\n }\n}\n```\n\n#### CLI / tty (CliRenderer · inquirer)\n\n```js\nconst answer = await inquirer.prompt([{\n type: 'list',\n name: 'choice',\n message: '选默认推送通道',\n choices: [\n { name: 'feishu:default (公司主账号)', value: 'feishu:default' },\n { name: '⭐ feishu:pm (pm bot)', value: 'feishu:pm' },\n ...\n ],\n default: 'feishu:pm',\n}]);\n```\n\n#### Web (WebRenderer · HTML form, 走 opphub-web)\n\n```html\n
\n

选默认推送通道

\n \n \n \n
\n```\n\n### 改造 + 风险\n\n| 风险 | 缓解 |\n|---|---|\n| skill 误写硬编码飞书 card | SKILL.md 加 lint / require.js static check (CI grep) |\n| 现有 plugin CLI 用户绕开 skill | plugin CLI inquirer **完全保留** |\n| 微信 iLink 不支持原生 card | WeixinRenderer 走文本 + emoji + 编号协议 (兜底) |\n| tty 没 TTY 时 CliRenderer 失败 | FallbackTextRenderer 兜底 (纯文本提一个一次性的\"我帮你做\") |\n| renderer 自己 crash | renderer 异常时自动回退到 FallbackTextRenderer (try/catch 包裹) |\n\n### 关联档\n\n- 设计稿:`workspace/skills/opphub/docs/v3.1-architecture.md` (本节)\n- 后续整改:Skill 全部 bin 改用 `bot.skillApi.*` 替代硬编码 JSON\n\n---\n\nFile v5.0.0:docs/server-schema-v31-design.md\n\n# opphub-server schema v3.1 · 知识库改造设计\n\n> \"现在的 公司能力卡片, 是记录到数据库的, 是字段固定的, 但是我要改成开放式的, 整个录入的信息是存到知识库里, 也就是向量存储, 这样就不拘泥于字段\"\n\n> **含义**: server schema 改造是 opphub-server 团队活, **不在 opphub skill v3.1 范围**\n\n---\n\n## 0. 文档归属\n\n| 项 | 归属 |\n|---|---|\n| 本档 (server schema) | opphub-server 团队 |\n| opphub skill v3.1 doc | skill 层 (我开干的范围) |\n| OpenClaw runtime 渲染层 | runtime 团队 (单独存档) |\n\n---\n\n\n> \"现在的 公司能力卡片, 是记录到数据库的, 是字段固定的, 但是我要改成开放式的, 整个录入的信息是存到知识库里, 也就是向量存储, 这样就不拘泥于字段\"\n\n## 2. 与 7/15 蒸馏设计的关系\n\n- **7/15 web P0 蒸馏设计**: `OpcProfile.distilledJson` + `embedding vector(768)` + `lib/distill.ts` LLM 填空字段\n\n---\n\n#### 6.5 schema 改造 (新加 + 弱化)\n\n**新加 `OpcKnowledgeEntry`**:\n\n```prisma\nmodel OpcKnowledgeEntry {\n id String @id @default(cuid())\n opcId String @map(\"opc_id\")\n sourceType String @map(\"source_type\") // rawText / url / upload / auto / manual\n sourceUrl String? @map(\"source_url\")\n rawText String @map(\"raw_text\") @db.Text\n chunks Json? // [{text, start, end, embedding_ref}]\n embedding Unsupported(\"vector(768)\")?\n embeddingModel String? @map(\"embedding_model\")\n embeddingVersion Int? @map(\"embedding_version\")\n fulltext Unsupported(\"tsvector\")? // 全文索引\n distilledTags String[] @map(\"distilled_tags\") // LLM 抽, 可空, 可自举\n visibility String @default(\"PRIVATE\") // PRIVATE / PUBLIC\n createdAt DateTime @default(now()) @map(\"created_at\")\n updatedAt DateTime @updatedAt @map(\"updated_at\")\n deletedAt DateTime? @map(\"deleted_at\")\n\n opc OpcAccount @relation(fields: [opcId], references: [opcId])\n\n @@index([opcId, deletedAt])\n @@index([visibility])\n @@map(\"opc_knowledge_entry\")\n}\n```\n\n**弱化 `OpcProfile`**(字段保留但**不强制填**,作为蒸馏快照):\n\n```prisma\nmodel OpcProfile {\n // 砍: mainSkill, subSkills, experience, cases, priceMin, priceMax, differentiators\n // 留: distilledJson (快照, LLM 蒸馏结果, 可空)\n // 留: embedding + embeddingModel + embeddingVersion (跨 entry 聚合向量)\n // 新加: knowledgeCount Int @default(0) // 关联多少条 OpcKnowledgeEntry\n // 新加: lastKnowledgeAt DateTime? // 最近一次入知识库\n}\n```\n\n**弱化 `OpcSkillCard`**(从\"多张字段卡\"改成\"多 entry 的视图\"):\n\n```prisma\n// 方案 A: 保留 OpcSkillCard 表, 但字段全是可空, 只作为展示快照\n// 方案 B: 直接砍 OpcSkillCard, 改用 OpcKnowledgeEntry 聚合\n```\n\n**全文索引 + 向量索引**(§7 决策已钉,7/14 拍):\n- pgvector HNSW: `m=16, ef=64`(7/14 拍)\n- tsvector: `to_tsvector('simple', rawText)`,GIN 索引\n- 检索: `embedding <=> query_embedding` + `fulltext @@ to_tsquery('simple', query_text)`,**完全不需要 JOIN 表**\n\n\n---\n\n#### 6.6 检索姿势 (matching 时)\n\n```sql\n-- 候选 OPC 召回 (Stage 1: 向量召回)\nSELECT e.opcId, e.id AS entryId, e.rawText,\n e.embedding <=> $1::vector AS distance,\n e.distilledTags\nFROM opc_knowledge_entry e\nWHERE e.deletedAt IS NULL\n AND e.visibility = 'PUBLIC'\nORDER BY e.embedding <=> $1::vector\nLIMIT 50;\n\n-- 二次过滤 (Stage 2: 全文 + tag 过滤, 不 JOIN 表)\nSELECT * FROM (\n -- 上面的结果\n) candidates\nWHERE candidates.fulltext @@ to_tsquery('simple', $2)\n AND candidates.distilledTags && $3::text[]; -- 任意 tag 命中\n\n-- LLM 精排 (Stage 3: 跟 7/15 设计同)\n```\n\nFile v5.0.0:docs/server-schema-v32-design.md\n\n# server schema v3.2 设计 · idempotent ingest\n\n>\n> 关联 workboard: `v3.3-skill-server-separation-2026-07-20` (id=bc34d33f-...)\n>\n> **跟 v3.1 关系**: v3.1 schema 弱化了 OpcProfile/OpcSkillCard，加了 OpcKnowledgeEntry;\n> v3.2 在 OpcKnowledgeEntry 基础上加 idempotent ingest 能力，**不动 v3.1 已拍的结构**。\n\n---\n\n\n\n1. **无去重**：server `OpcKnowledgeEntry.create()` 直接 create，0 个 unique constraint（除 entry.id 主键）\n - 同 OPC 同 rawText 提交 2 次 → 2 条 entry\n\n2. **无幂等键**：改一个错别字只能新建 entry，原 entry 还在\n - 没 `idempotencyKey` 概念\n - 没 `contentHash` 概念\n\n3. **无冲突检测**：录入有矛盾字段时静默入库\n - 没版本管理 / source 优先级 / 时间戳覆盖策略\n\n---\n\n\n- **P1**: skill = 采集者 + 翻译者，只产 rawText。**不做任何去重 / 冲突判断 / 版本管理**\n- **P2**: server = 仓库 + 处理器，做去重 / 冲突检测 / 版本管理 / 蒸馏 / 嵌入 / 召回\n- **P3**: 幂等性 + 冲突处理都在 server 端按 **content-addressable + idempotencyKey** 双重保险\n\n---\n\n## 三、Schema 改造 (最小)\n\n### 3.1 OpcKnowledgeEntry 加字段\n\n```prisma\nmodel OpcKnowledgeEntry {\n // 现有字段 (v3.1) - 不动\n id String @id @default(cuid())\n opcId String @map(\"opc_id\")\n sourceType String @map(\"source_type\") // rawText / url / upload / auto / manual\n sourceUrl String? @map(\"source_url\")\n rawText String @map(\"raw_text\") @db.Text\n status String @default(\"pending\") // pending / processing / done / failed\n chunkSummary Json? @map(\"chunk_summary\")\n embeddingModel String? @map(\"embedding_model\")\n embeddingVersion Int? @map(\"embedding_version\")\n distilledTags String[] @default([]) @map(\"distilled_tags\")\n visibility String @default(\"PRIVATE\")\n retryCount Int @default(0) @map(\"retry_count\")\n lastError String? @map(\"last_error\")\n createdAt DateTime @default(now()) @map(\"created_at\")\n updatedAt DateTime @updatedAt @map(\"updated_at\")\n deletedAt DateTime? @map(\"deleted_at\")\n\n \n // idempotencyKey: SHA256(opcId + entryType + entryDimension)\n // - 同 OPC 同 type+dimension → 同 key → server 走 upsert 路径\n // - 改 rawText (改一个错别字) → contentHash 变但 key 不变 → server 比 hash\n idempotencyKey String? @map(\"idempotency_key\")\n \n // contentHash: SHA256(rawText) 当前内容指纹\n // - 用来判断 rawText 是否真变了\n contentHash String? @map(\"content_hash\")\n \n // entryType + entryDimension: skill 拆出来的 4 类卡 (ability/upstream/downstream/peer) + 维度名\n // - 这 2 个字段是去重的最低必要锚点, 不属于\"过度字段化\"\n // - skill knowledge-card 拆出来直接传过来, server 不解析 rawText 前缀\n entryType String? @map(\"entry_type\") // ability / upstream / downstream / peer\n entryDimension String? @map(\"entry_dimension\") // 达人营销 / KOL 资源 / 品牌方 / ...\n \n // previousEntryId: 软链 (被覆盖的旧 entry), 保留 history\n // - 跟 git 同款: 老 entry 标 supersededAt, 新 entry previousEntryId 链回老 entry\n // - 召回时不查 supersededAt != null 的 entry\n previousEntryId String? @map(\"previous_entry_id\")\n supersededAt DateTime? @map(\"superseded_at\") // 非 null 即被新 entry 取代\n \n opc OpcAccount @relation(fields: [opcId], references: [opcId])\n chunks OpcKnowledgeChunk[] @relation(\"EntryChunks\")\n \n @@unique([opcId, idempotencyKey]) // 幂等唯一键\n @@index([opcId, entryType, entryDimension]) // 检索\n @@index([opcId, deletedAt]) // 现有\n @@index([supersededAt]) // 召回过滤\n}\n```\n\n\n\n**v3.2 加的字段**:\n| 字段 | 是否破坏\"开放式\" |\n|---|---|\n| `idempotencyKey` | ❌ 不破坏。skill 算 hash 传过来, server 不解析内容 |\n| `contentHash` | ❌ 不破坏。纯 hash, 不存业务语义 |\n| `previousEntryId` + `supersededAt` | ❌ 不破坏。跟 git 同款, 不删 history |\n\n**核心**: rawText 仍然全量存 (这是 12:58 拍的精神), 加的结构化字段都是**辅助字段** (去重 + 索引 + 版本链), 不替代 rawText。\n\n---\n\n## 四、接口改造\n\n### 4.1 新接口: `POST /api/knowledge/ingest` v2 (idempotent)\n\n**位置**: `/app/api/user/knowledge/ingest/route.ts` (现有 v1 接口保留, 加 v2 路由或加 query param `?v=2`)\n\n**request body**:\n```json\n{\n \"opcId\": \"opc_1hz6wsjrmt1s\",\n \"rawText\": \"睿驰嘉禾 · 能力卡片 · 达人营销\\n\\n(证据: rawText 包含 \\\"达人营销\\\")\\n...\",\n \"entryType\": \"ability\",\n \"entryDimension\": \"达人营销\",\n \"idempotencyKey\": \"sha256_hash\",\n \"contentHash\": \"sha256_hash\"\n}\n```\n\n**response 4 种**:\n```json\n// 1. 新增成功\n{ \"ok\": true, \"action\": \"created\", \"entryId\": \"cmxxx\", \"chunkCount\": null, \"status\": \"pending\" }\n\n// 2. 幂等命中 (rawText 没变)\n{ \"ok\": true, \"action\": \"no_change\", \"entryId\": \"cmxxx\", \"message\": \"rawText 与已有 entry 完全一致\" }\n\n// 3. 软链覆盖 (rawText 变了但无关键字段冲突)\n{ \n \"ok\": true, \n \"action\": \"soft_chain_override\", \n \"entryId\": \"cmyyy\", // 新 entry\n \"previousEntryId\": \"cmxxx\", // 被取代的旧 entry\n \"supersededEntry\": { \"id\": \"cmxxx\", \"supersededAt\": \"...\" }\n}\n\n// 4. 冲突返报告 (rawText 变了 + 关键字段冲突, 不入库, 等用户拍)\n{ \n \"ok\": false, \n \"conflict\": true, \n \"conflictReport\": {\n \"entryId\": \"cmxxx\", // 已有 entry\n \"diffType\": \"key_field_conflict\" // vs \"incremental_addition\"\n }\n}\n```\n\n### 4.2 实现逻辑 (server 端伪代码)\n\n```typescript\n// app/api/user/knowledge/ingest/route.ts v2\nexport async function POST(req: NextRequest) {\n const opcId = getOpcId(req);\n if (!opcId) return unauthorized();\n \n const { rawText, entryType, entryDimension, idempotencyKey, contentHash } = await req.json();\n if (!rawText || !idempotencyKey || !contentHash) return missingArgs();\n \n // 1. 查 unique\n const existing = await prisma.opcKnowledgeEntry.findUnique({\n where: { opcId_idempotencyKey: { opcId, idempotencyKey } }\n });\n \n // 2. 不存在 → 直接 create\n if (!existing) {\n const entry = await prisma.opcKnowledgeEntry.create({\n data: { opcId, rawText, entryType, entryDimension, \n idempotencyKey, contentHash, status: 'pending' }\n });\n return { ok: true, action: 'created', entryId: entry.id };\n }\n \n // 3. 存在 → 比 contentHash\n if (existing.contentHash === contentHash) {\n // rawText 没变, 静默幂等命中\n return { ok: true, action: 'no_change', entryId: existing.id };\n }\n \n // 4. contentHash 变 → 比 rawText 关键字段\n const diff = diffRawText(existing.rawText, rawText);\n \n if (diff.hasKeyFieldConflict) {\n // 关键字段冲突 → 返 conflictReport, 不入库\n return { \n ok: false, \n conflict: true, \n conflictReport: {\n entryId: existing.id,\n oldRawText: existing.rawText,\n newRawText: rawText,\n conflictFields: diff.conflictFields,\n diffType: 'key_field_conflict'\n }\n };\n }\n \n // 5. 无关键字段冲突 → 软链覆盖\n // 5a. 老 entry 标 superseded\n await prisma.opcKnowledgeEntry.update({\n where: { id: existing.id },\n data: { supersededAt: new Date() }\n });\n // 5b. 新 entry, previousEntryId 链回老 entry\n const newEntry = await prisma.opcKnowledgeEntry.create({\n data: { \n opcId, rawText, entryType, entryDimension, \n idempotencyKey, contentHash, \n previousEntryId: existing.id,\n status: 'pending'\n }\n });\n return { \n ok: true, \n action: 'soft_chain_override',\n entryId: newEntry.id,\n previousEntryId: existing.id,\n supersededEntry: { id: existing.id, supersededAt: new Date().toISOString() }\n };\n}\n```\n\n### 4.3 冲突检测算法 `diffRawText`\n\n```typescript\n// lib/diff-raw-text.ts\nfunction diffRawText(oldText: string, newText: string) {\n // 1. 提取关键字段 (正则 + 简单启发式, 不调 LLM)\n const oldFields = extractKeyFields(oldText);\n const newFields = extractKeyFields(newText);\n \n // 2. 比对关键字段\n const conflictFields: string[] = [];\n for (const key of ['legal_person', 'registered_capital', 'business_direction', 'industry']) {\n if (oldFields[key] && newFields[key] && oldFields[key] !== newFields[key]) {\n conflictFields.push(`${key}: ${oldFields[key]} → ${newFields[key]}`);\n }\n }\n \n // 3. 关键字段冲突 vs 增量补充\n const hasKeyFieldConflict = conflictFields.length > 0;\n \n return { \n hasKeyFieldConflict, \n conflictFields,\n diffType: hasKeyFieldConflict ? 'key_field_conflict' : 'incremental_addition'\n };\n}\n\n// 简单启发式提取关键字段 (不调 LLM, 性能优先)\nfunction extractKeyFields(rawText: string) {\n const fields: Record = {};\n \n const legalPersonMatch = rawText.match(/法[定]?人[::]?\\s*([^\\s,，;；\\n]+)/);\n if (legalPersonMatch) fields.legal_person = legalPersonMatch[1];\n \n if (bossMatch) fields.legal_person = bossMatch[1];\n \n // 注册资本\n const capitalMatch = rawText.match(/注册资[本金][::]?\\s*([0-9,，.]+\\s*[万亿]?)/);\n if (capitalMatch) fields.registered_capital = capitalMatch[1];\n \n // 业务方向 (取第一条)\n const businessMatch = rawText.match(/业务方向[::]?\\s*([^\\n]+)/);\n if (businessMatch) fields.business_direction = businessMatch[1];\n \n return fields;\n}\n```\n\n### 4.4 召回过滤改动 (search 路由)\n\n```typescript\n// app/api/user/knowledge/search/route.ts 改动\nconst entries = await prisma.opcKnowledgeEntry.findMany({\n where: { \n opcId, \n deletedAt: null,\n supersededAt: null, // ← 新加: 召回时不查被取代的 entry\n },\n // ... 其他\n});\n```\n\n---\n\n## 五、Skill 端改动 (v3.3 skill 范围)\n\n### 5.1 新加 `bin/opphub-knowledge-submit.js`\n\n**职责**: 把 cards 提交给 server, 接收 conflictReport, **不做任何判断**\n\n```javascript\n#!/usr/bin/env node\n// bin/opphub-knowledge-submit.js · v3.3.0\n// \n// 本 bin 纯转发 + 透传 server 响应\n//\n// 输入: cards.json (knowledge-card 输出)\n// 输出: { submitted: [...], deduplicated: [...], conflicts: [...] }\n\nimport { readFileSync } from \"node:fs\";\nimport { createHash } from \"node:crypto\";\nimport { readToken } from \"../lib/opphub-plugin-client.js\";\n\nconst API_BASE = \"https://api.opphub.ruiplus.cn\";\n\nfunction sha256(s) {\n return createHash(\"sha256\").update(s, \"utf8\").digest(\"hex\");\n}\n\nasync function main() {\n const args = parseArgs(process.argv.slice(2));\n const cards = JSON.parse(readFileSync(args.cards, \"utf8\"));\n \n // 拿 token (plugin = source of truth, 14:21 拍)\n const tokenDoc = await readToken();\n const accessToken = tokenDoc?.access_token;\n if (!accessToken) {\n return out({ ok: false, error: \"no_token\", message: \"需要先偶合登录\" });\n }\n \n // 从 token 解 opcId (skill 不需要 opphub-server 二次确认)\n const opcId = decodeJwt(accessToken).opcId;\n \n const results = { submitted: [], deduplicated: [], conflicts: [] };\n \n for (const card of cards) {\n const idempotencyKey = sha256(`${opcId}|${card.type}|${card.dimension}`);\n const contentHash = sha256(card.text);\n \n const resp = await fetch(`${API_BASE}/api/knowledge/ingest`, {\n method: \"POST\",\n headers: { \n \"Authorization\": `Bearer ${accessToken}`, \n \"Content-Type\": \"application/json\" \n },\n body: JSON.stringify({\n opcId, \n rawText: card.text, \n entryType: card.type, \n entryDimension: card.dimension, \n idempotencyKey, \n contentHash\n })\n });\n const data = await resp.json();\n \n if (data.ok && data.action === \"no_change\") {\n results.deduplicated.push({ cardIndex: card.index, entryId: data.entryId });\n } else if (data.ok) {\n // created 或 soft_chain_override 都算 \"submitted\"\n results.submitted.push({ \n cardIndex: card.index, \n entryId: data.entryId,\n action: data.action,\n previousEntryId: data.previousEntryId || null\n });\n } else if (data.conflict) {\n results.conflicts.push({ \n cardIndex: card.index, \n conflictReport: data.conflictReport \n });\n }\n }\n \n out({\n ok: true,\n opcId,\n summary: {\n submitted: results.submitted.length,\n deduplicated: results.deduplicated.length,\n conflicts: results.conflicts.length\n },\n submitted: results.submitted,\n deduplicated: results.deduplicated,\n conflicts: results.conflicts,\n nextStep: results.conflicts.length > 0 \n ? \"用 bot.skillApi.askInteractive 让用户拍冲突项\" \n : \"全部成功, 跑 knowledge-match\"\n });\n}\n\nmain().catch(e => out({ ok: false, error: e?.message ?? String(e) }));\n```\n\n### 5.2 改 `bin/opphub-knowledge-ingest-batch.js`\n\n**改动**: 删\"循环调 knowledge-add\"的逻辑, 改成\"调 submit\"\n\n```javascript\nfor (const card of cards) {\n await opphub-knowledge-add --raw-text \"$card.text\" --source-type auto\n}\n\n// 新逻辑 (v3.3)\nopphub-knowledge-submit --cards cards.json\n```\n\ningest-batch 变成纯编排入口 (不做实际提交), 实际提交逻辑在 submit bin。\n\n### 5.3 不动的部分 (skill 端)\n\n- ✅ `knowledge-card` 不动 — 仍然按 MCN/SaaS 模板拆 card\n- ✅ `knowledge-discover` 不动 — 仍然产 queryPlan / 验 rawText\n- ✅ `knowledge-search` 不动 — 召回接口跟 server 同步过滤 superseded\n- ✅ `knowledge-relate` 不动 — xls 解析逻辑\n\n---\n\n## 六、Bot 端流程改造 (v3.3)\n\n按 SKILL.md §录入公司流程 阶段 4-5，bot 拿到 `knowledge-submit` 的结果:\n\n```\nbot 给你看:\n📋 录入睿驰嘉禾 · 结果报告\n\n✅ 新提交 7 条\n🔁 已有 3 条 (跳过)\n⚠️ 冲突 1 条 (要你拍):\n - 能力卡片 · 虚拟人 IP 孵化\n 老: 数字人形象设计\n 新: 虚拟主播代运营\n 冲突字段: 业务描述\n\n回 \"保留旧的\" / \"用新的\" / \"跳过\"\n```\n\n**bot 走 IntentMessage**:\n- `askInteractive` 列出冲突项 + 让用户点选\n- 用户选择后, bot 再调一次 submit (不传冲突项的 cards, 或传 `forceOverride: true`)\n\n**冲突覆盖流程**:\n```\n用户: 用新的\nbot: 收到, 重新提交 1 条 (跳过冲突检测)\nbot 调: opphub-knowledge-submit --cards cards.json --force-override-conflict --json\nserver: 跳过冲突检测, 强制 soft_chain_override\n```\n\n---\n\n## 七、部署 & 测试\n\n\n1. ✅ **本地 dev migrate** (不部署 ECS)\n2. ✅ server 端 schema 改造 + 接口实现 (本地 dev server)\n3. ✅ skill 端 submit + ingest-batch 改造 (本地开发)\n\n### 7.2 本地 E2E 测试 (mock server)\n\n```bash\n# 1. 启本地 mock server\nnode tests/mock-knowledge-server.js &\n\n# 2. 跑 skill 端 (用本地 mock URL)\nOPPHUB_API_BASE=http://localhost:4001 bin/opphub knowledge-submit --cards /tmp/cards.json\n\n# 3. 验证 4 种响应:\n# - 新增成功 (action: created)\n# - 幂等命中 (action: no_change)\n# - 软链覆盖 (action: soft_chain_override)\n# - 冲突报告 (conflict: true)\n```\n\n### 7.3 回滚\n\n- server schema 改造加字段是**可空**的 (idempotencyKey? contentHash? entryType? entryDimension? previousEntryId? supersededAt?)\n- 老 entry 没这 6 个字段 → 不影响老流程\n- 新 entry 没 idempotencyKey → 走\"非幂等路径\" (跟 v3.1 行为一致)\n- **回滚安全**: 直接撤 server 代码, schema 不动\n\n---\n\n## 八、不动的事 (反复钉的纪律)\n\n- ❌ 不动 plugin 仓 (plugin-cli-cleanup-2026-07-20 单独 backlog)\n\n---\n\n## 九、关联文档\n\n- **设计稿 (本文)**: `workspace/skills/opphub/docs/server-schema-v32-design.md`\n- **v3.1 设计稿**: `workspace/skills/opphub/docs/server-schema-v31-design.md` (平行, 不冲突)\n- **v3.1 架构**: `workspace/skills/opphub/docs/v3.1-architecture.md`\n- **运行时渲染层**: `workspace/skills/opphub/docs/runtime-channel-renderer-v31-design.md` (不动)\n- **workboard 卡**: `v3.3-skill-server-separation-2026-07-20` (id=bc34d33f-...)\n- **MEMORY 红线**: `~/.openclaw/workspace-dev/MEMORY.md` §v3.3 backlog\n\n---\n\n## 十、下次同样场景\n\nFile v5.0.0:docs/superpowers/plans/2026-07-23-phase1-nav-rewrite.md\n\n# OppHub 阶段 1：导航 + 角色表达改造实施计划\n\n> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.\n\n**Goal:** 把 opphub-web 当前的 7 个一级导航改造为 spec 里的 6 个 (`dashboard / knowledge / discover / matches / orders / wallet`), 移除路径重复 (`/me/capabilities` ↔ `/opc/profile`, `/demands` ↔ `/marketplace`), 旧路径 301 跳新路径\n\n**Architecture:** 保持 Next.js app router 不变, 只移文件 / 改名 / 加 redirect; 数据接口 (`/api/knowledge` 等) 不动, 这样改动可单独发版\n\n**Tech Stack:** Next.js 14 app router, TypeScript, `next.config.js` `redirects()`\n\n---\n\n## 总体任务地图\n\n```\nTask 1: 文件迁移 + 旧路径 301 跳新路径\n - /me/capabilities → /knowledge\n - /opc/profile → /knowledge/[id]\n - /marketplace → /discover?role=seller\n - /demands → /discover?role=buyer\n - /match → /matches\n\nTask 2: 新建 /knowledge, /discover, /matches (默认重定向列表)\n\nTask 3: TopBar 改 6 个一级导航\n\nTask 4: 验证 (本地 build + 浏览器手测)\n\nTask 5: 部署 ECS + 终验\n```\n\n---\n\n## Task 1: 加 next.config.js redirects 套所有旧路径\n\n**Files:**\n- Modify: `next.config.js` (新增 redirects 段, 若不存在则创建)\n- Test: curl 各旧路径能 301 到新路径\n\n- [ ] **Step 1: 读现有 next.config.js**\n\n```bash\ntest -f opphub-web/next.config.js && cat opphub-web/next.config.js || echo \"NOT_EXISTS\"\n```\n\n预期: 输出现有内容 (或 `NOT_EXISTS`)\n\n- [ ] **Step 2: 添加 redirects 段**\n\n文件 `opphub-web/next.config.js`:\n\n```js\n/** @type {import('next').NextConfig} */\nconst nextConfig = {\n // v3.4 spec 阶段 1: 老路径 301 到新导航 (录→撮→谈→成)\n async redirects() {\n return [\n { source: \"/me/capabilities\", destination: \"/knowledge\", permanent: true },\n { source: \"/opc/profile\", destination: \"/knowledge/profile\", permanent: true },\n { source: \"/marketplace\", destination: \"/discover?role=seller\", permanent: true },\n { source: \"/demands\", destination: \"/discover?role=buyer\", permanent: true },\n { source: \"/match\", destination: \"/matches\", permanent: true },\n ];\n },\n};\n\nmodule.exports = nextConfig;\n```\n\n如果原本有配置就把 redirects 段塞进顶层 const 内 (保留原 reactStrictMode、images 等其他配置项)。\n\n- [ ] **Step 3: 静态检查**\n\n```bash\ncd opphub-web && node -e \"console.log(JSON.stringify(require('./next.config.js'), null, 2))\" | head -40\n```\n\n预期: JSON 包含 `redirects` 是 array of 5 项\n\n- [ ] **Step 4: Commit**\n\n```bash\ncd opphub-web\ngit add next.config.js\ngit commit -m \"feat(nav): 旧导航路径 301 到新路径 (阶段 1)\"\n```\n\n---\n\n## Task 2: 新建 `/knowledge` 主页面 (Tab 表达 4 种角色)\n\n**Files:**\n- Create: `app/knowledge/page.tsx`\n- Create: `app/knowledge/profile/page.tsx` (替代 `/opc/profile`)\n- Create: `app/knowledge/[id]/page.tsx` (替代 `/opc/profile/[opcId]` 后续用)\n\n- [ ] **Step 1: 新建 knowledge 主页 (空壳)**\n\n文件 `app/knowledge/page.tsx`:\n\n```tsx\n\"use client\";\nimport { useState } from \"react\";\nimport TopBar from \"../_components/TopBar\";\n\ntype Tab = \"ability\" | \"downstream\" | \"upstream\" | \"peer\";\n\nconst TAB_LABEL: Record = {\n ability: \"我能提供\",\n downstream: \"我在找\",\n upstream: \"我的依赖\",\n peer: \"同行关系\",\n};\n\nconst TAB_DESC: Record = {\n ability: \"录入你的能力/服务, AI 帮你找到需求方\",\n downstream: \"录入你想找的东西, AI 帮你找到供应方\",\n upstream: \"你依赖的供应商/数据源/工具\",\n peer: \"同业联盟 / 同行参考 / 关联企业\",\n};\n\nexport default function KnowledgePage() {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const [mounted, setMounted] = useState(false);\n const [tab, setTab] = useState(\"ability\");\n const [opcId, setOpcId] = useState(\"\");\n const [entries, setEntries] = useState<{ id: string; dimension: string; rawText: string; updatedAt: string }[]>([]);\n const [loading, setLoading] = useState(false);\n\n React.useEffect(() => setMounted(true), []);\n React.useEffect(() => {\n if (!mounted || !opcId) return;\n setLoading(true);\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n fetch(`${apiBase}/api/knowledge?opcId=me&entryType=${tab}`, {\n headers: { authorization: \"Bearer \" + token },\n })\n .then((r) => r.json())\n .then((j) => setEntries(j?.ok && Array.isArray(j.data) ? j.data : []))\n .catch(() => setEntries([]))\n .finally(() => setLoading(false));\n }, [mounted, opcId, tab, apiBase]);\n\n React.useEffect(() => {\n try { setOpcId(localStorage.getItem(\"opcId\") || \"\"); } catch {}\n }, []);\n\n if (!mounted) {\n return (\n
\n \n
加载中...
\n
\n );\n }\n\n return (\n
\n \n
\n

📇 知识库

\n

\n 在聊天窗口 @偶合 说一句话录入, 或在这里浏览你已录入的条目\n

\n\n {/* Tabs */}\n
\n {(Object.keys(TAB_LABEL) as Tab[]).map((t) => (\n setTab(t)}\n style={{\n padding: \"8px 16px\",\n borderRadius: 8,\n border: \"none\",\n cursor: \"pointer\",\n background: tab === t ? \"white\" : \"rgba(255,255,255,0.2)\",\n color: tab === t ? \"#4f46e5\" : \"white\",\n fontWeight: tab === t ? 600 : 400,\n }}\n >\n {TAB_LABEL[t]}\n \n ))}\n
\n\n

\n {TAB_DESC[tab]}\n

\n\n {/* List */}\n {loading ? (\n
⏳ 加载中...
\n ) : entries.length === 0 ? (\n \n ) : (\n
\n {entries.map((e) => )}\n
\n )}\n
\n
\n );\n}\n\nfunction EmptyState({ tab }: { tab: Tab }) {\n return (\n
\n
📇
\n
\n {tab === \"ability\" && \"你还没说能干啥\"}\n {tab === \"downstream\" && \"你还没说要找啥\"}\n {tab === \"upstream\" && \"还没录入上游依赖\"}\n {tab === \"peer\" && \"还没录入同行关系\"}\n
\n
\n 去聊天窗口 @偶合 说\"{tab === \"ability\" ? \"录入我的服务\" : tab === \"downstream\" ? \"我需要 xxx\" : tab === \"upstream\" ? \"录入我的供应商\" : \"录入同行\"}\" 开始\n
\n
\n );\n}\n\nfunction EntryCard({ entry }: { entry: { id: string; dimension: string; rawText: string; updatedAt: string } }) {\n const preview = entry.rawText.slice(0, 200);\n return (\n \n
\n
{entry.dimension}
\n
{preview}{entry.rawText.length > 200 ? \"...\" : \"\"}
\n
更新: {new Date(entry.updatedAt).toLocaleString(\"zh-CN\")}
\n
\n \n );\n}\n```\n\n注意: `React.useEffect` 那一段用 `useEffect` (不要写 `React.useEffect`), 顶部要加 `import { useEffect, useState } from \"react\";`\n\n(此步只是把 API 数据展示出来, 编辑功能暂未接入; 详细录入正反馈/录入三选一按钮留到阶段 2-3)\n\n- [ ] **Step 2: 新建 knowledge/profile 替代 /opc/profile**\n\n文件 `app/knowledge/profile/page.tsx`:\n\n```tsx\n\"use client\";\nimport { Suspense, useEffect, useState } from \"react\";\nimport Link from \"next/link\";\nimport TopBar from \"../../_components/TopBar\";\nimport { useSearchParams } from \"next/navigation\";\n\nexport const dynamic = \"force-dynamic\";\n\ntype Company = {\n companyName?: string;\n companyType?: string;\n legalPerson?: string;\n teamSize?: number;\n creditCode?: string;\n address?: string;\n};\ntype Profile = {\n mainSkill?: string;\n subSkills?: string[];\n experience?: string;\n priceMin?: number;\n priceMax?: number;\n city?: string;\n trustScore?: number;\n completedOrders?: number;\n};\n\nexport default function KnowledgeProfilePage() {\n return (\n 加载中...}>\n \n \n );\n}\n\nfunction ProfileInner() {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const sp = useSearchParams();\n const opcIdQuery = sp?.get(\"opcId\");\n const [mounted, setMounted] = useState(false);\n const [me, setMe] = useState<{ opcId?: string } | null>(null);\n const [targetOpc, setTargetOpc] = useState(null);\n const [company, setCompany] = useState(null);\n const [profile, setProfile] = useState(null);\n\n useEffect(() => { setMounted(true); }, []);\n useEffect(() => {\n try { setMe({ opcId: localStorage.getItem(\"opcId\") || undefined }); } catch {}\n }, []);\n\n useEffect(() => {\n if (!mounted || !me) return;\n const target = !opcIdQuery || opcIdQuery === me.opcId ? me.opcId! : opcIdQuery;\n setTargetOpc(target);\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n if (target === me.opcId) {\n fetch(`${apiBase}/api/auth/me`, { headers: { authorization: \"Bearer \" + token } })\n .then((r) => r.json())\n .then((j) => setProfile(j?.profile ?? null))\n .catch(() => {});\n } else {\n fetch(`${apiBase}/api/opc/${target}/profile`, { headers: { authorization: \"Bearer \" + token } })\n .then((r) => r.json())\n .then((j) => setProfile(j?.profile ?? null))\n .catch(() => {});\n }\n }, [mounted, me, opcIdQuery, apiBase]);\n\n if (!mounted || targetOpc === null) {\n return
⏳ 加载中...
;\n }\n\n const isMe = me?.opcId === targetOpc;\n return (\n
\n \n
\n

\n {isMe ? \"我的公开资料\" : `对方资料: ${profile?.mainSkill ?? \"—\"}`}\n

\n
\n {profile ? 
{JSON.stringify(profile, null, 2)}
:
暂无资料
}\n {isMe && (\n
\n ← 返回知识库\n
\n )}\n
\n
\n
\n );\n}\n```\n\n这一版是简化的过渡页 — 显示从 `/api/auth/me` 或 `/api/opc/[opcId]/profile` 拉的 profile 原始 JSON. 阶段 3 重做美化.\n\n- [ ] **Step 3: 新建 knowledge/[id] 单条目查看 (替代需要时的 /opc/profile 查看页)**\n\n> 阶段 1 只要求把旧路径跳走, 不必现在就做每条知识详情。占位即可。\n\n文件 `app/knowledge/[id]/page.tsx`:\n\n```tsx\n\"use client\";\nimport { useEffect, useState } from \"react\";\nimport { useParams } from \"next/navigation\";\nimport TopBar from \"../../_components/TopBar\";\n\nexport default function KnowledgeEntryDetailPage() {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const params = useParams<{ id: string }>();\n const id = params?.id ?? \"\";\n const [data, setData] = useState<{ id: string; entryDimension: string; entryType: string; rawText: string; updatedAt: string } | null>(null);\n const [loading, setLoading] = useState(true);\n\n useEffect(() => {\n if (!id) return;\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n fetch(`${apiBase}/api/knowledge?opcId=me`, { headers: { authorization: \"Bearer \" + token } })\n .then((r) => r.json())\n .then((j) => {\n const found = (j?.data || []).find((e: any) => e.id === id);\n setData(found ?? null);\n })\n .catch(() => {})\n .finally(() => setLoading(false));\n }, [id, apiBase]);\n\n return (\n
\n \n
\n {loading &&
⏳ 加载中...
}\n {data && (\n
\n
{data.entryType}
\n

{data.entryDimension}

\n 
{data.rawText}
\n
\n )}\n
\n
\n );\n}\n```\n\n- [ ] **Step 4: 验证 build**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"(Compiled|error|knowledge|discover|matches)\" | head -10\n```\n\n预期: `Compiled successfully`, 路由里出现 `/knowledge`, `/knowledge/profile`, `/knowledge/[id]`\n\n- [ ] **Step 5: Commit**\n\n```bash\ncd opphub-web\ngit add app/knowledge\ngit commit -m \"feat(knowledge): 新建 /knowledge Tab 页 + 单条目详情 + profile 替代 /opc/profile\"\n```\n\n---\n\n## Task 3: 新建 `/discover` 替代 `/demands` 和 `/marketplace`\n\n**Files:**\n- Create: `app/discover/page.tsx` (整合 demands 表 + marketplace 表)\n\n- [ ] **Step 1: 新建 discover 主页 (Tabs 切买家/卖家视角)**\n\n文件 `app/discover/page.tsx`:\n\n```tsx\n\"use client\";\nimport { useEffect, useState } from \"react\";\nimport { useSearchParams, useRouter } from \"next/navigation\";\nimport TopBar from \"../_components/TopBar\";\n\ntype Demand = {\n demandId: string;\n title: string;\n budget: number;\n city?: string;\n deadline?: string;\n category?: string;\n status?: string;\n createdAt: string;\n};\n\ntype OPC = {\n opcId: string;\n displayName: string;\n mainSkill: string;\n city?: string;\n trustScore: number;\n priceMin: string | null;\n priceMax: string | null;\n};\n\ntype Role = \"buyer\" | \"seller\";\n\nexport default function DiscoverPage() {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const sp = useSearchParams();\n const router = useRouter();\n const initialRole = (sp?.get(\"role\") === \"seller\" ? \"seller\" : \"buyer\") as Role;\n const [role, setRole] = useState(initialRole);\n const [mounted, setMounted] = useState(false);\n const [loading, setLoading] = useState(false);\n const [demands, setDemands] = useState([]);\n const [opcs, setOpcs] = useState([]);\n\n useEffect(() => { setMounted(true); }, []);\n useEffect(() => {\n if (!mounted) return;\n setLoading(true);\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n if (role === \"buyer\") {\n // 我是买方 → 浏览供应方\n fetch(`${apiBase}/api/opc/marketplace`, { headers: { authorization: \"Bearer \" + token } })\n .then((r) => r.json()).then((j) => setOpcs(j?.data ?? [])).catch(() => setOpcs([]))\n .finally(() => setLoading(false));\n } else {\n // 我是卖方 → 浏览需求\n fetch(`${apiBase}/api/demands`, { headers: { authorization: \"Bearer \" + token } })\n .then((r) => r.json()).then((j) => setDemands(j?.demands ?? j?.data ?? []))\n .catch(() => setDemands([]))\n .finally(() => setLoading(false));\n }\n }, [mounted, role, apiBase]);\n\n function switchRole(r: Role) {\n setRole(r);\n router.replace(`/discover?role=${r}`);\n }\n\n if (!mounted) {\n return
加载中...
;\n }\n\n return (\n
\n \n
\n

🔍 发现

\n

\n {role === \"buyer\" ? \"看供应方 (你的潜在合作方)\" : \"看需求 (你的潜在客户)\"}\n

\n
\n \n \n
\n\n {loading &&
⏳ 加载中...
}\n\n {!loading && role === \"buyer\" && (\n
\n
共 {opcs.length} 个供应方
\n {opcs.length === 0 ? (\n
暂无供应方
\n ) : (\n \n )}\n
\n )}\n\n {!loading && role === \"seller\" && (\n
\n
共 {demands.length} 条需求
\n {demands.length === 0 ? (\n
暂无公开需求
\n ) : (\n \n )}\n
\n )}\n
\n
\n );\n}\n```\n\n- [ ] **Step 2: Build 验证**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"(Compiled|error|discover)\" | head -10\n```\n\n预期: `/discover` 出现在路由列表, 无 error\n\n- [ ] **Step 3: Commit**\n\n```bash\ncd opphub-web\ngit add app/discover\ngit commit -m \"feat(discover): 新建 /discover (合并 /demands + /marketplace 视角)\"\n```\n\n---\n\n## Task 4: 新建 `/matches` 替代 `/match`\n\n**Files:**\n- Create: `app/matches/page.tsx` (暂时代理 `/match` 内容, 但 URL 改成复数)\n\n- [ ] **Step 1: 复制 match/page.tsx 到 matches/page.tsx, 改 current=**\n\n```bash\ncd opphub-web\nsed 's|\"match\"|\"matches\"|g; s|/api/opc/match/today|/api/opc/match/today|g' app/match/page.tsx > app/matches/page.tsx\ngrep -n \"match\\|Match\" app/matches/page.tsx | head -10\n```\n\n预期: 看到 `/api/opc/match/today` 仍保留 (server 路径未改), `current=\"matches\"` 已替换\n\n- [ ] **Step 2: 在 matches/page.tsx 顶部注释里更新 spec 引用**\n\n文件 `app/matches/page.tsx`, 改注释行 (从原 file 的注释基础替换):\n\n```tsx\n// /matches · 撮合流 (v3.4 spec 阶段 1: 替代 /match, 旧路径 301 到此)\n// 当前阶段先复刻 /match 内容, 阶段 3 重做撮合详情/状态机\n// 数据源: GET /api/opc/match/today\n// 字段: matchingId/demandId/demandTitle/demandBudget/demandCity/demandDeadline/opc/scores/explainText/createdAt\n```\n\n(替换文件最顶部的注释段, 只改 comment 不动 logic)\n\n- [ ] **Step 3: Build 验证**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"(Compiled|error|/matches|/match)\" | head -10\n```\n\n预期: `/matches` 出现在路由列表, `/match` 仍然存在 (因为 redirect 在 server 级, 路由仍编译), 无 error\n\n- [ ] **Step 4: Commit**\n\n```bash\ncd opphub-web\ngit add app/matches\ngit commit -m \"feat(matches): 新建 /matches (复刻 /match 内容, 阶段 3 重做)\"\n```\n\n---\n\n## Task 5: 改 TopBar 用 6 个一级入口\n\n**Files:**\n- Modify: `app/_components/TopBar.tsx` (`PAGE_NAME` map + `desktopLinks` + `mobileLinks` 三处)\n\n- [ ] **Step 1: 改 PAGE_NAME 加新键值**\n\n文件 `app/_components/TopBar.tsx`, 修改 PAGE_NAME 常量 (在第 19 行附近):\n\n```tsx\nconst PAGE_NAME: Record = {\n \"\": \"首页\",\n dashboard: \"工作台\",\n matches: \"撮合\",\n orders: \"交易\",\n knowledge: \"知识库\",\n \"knowledge/profile\": \"我的资料\",\n discover: \"发现\",\n wallet: \"钱包\",\n messages: \"通知\",\n \"messages/legacy\": \"消息\",\n \"account/security\": \"账号安全\",\n admin: \"后台\",\n onboarding: \"引导\",\n login: \"登录\",\n register: \"注册\",\n activate: \"设备激活\",\n};\n```\n\n- [ ] **Step 2: 改 desktopLinks (替换为 6 个)**\n\n文件 `app/_components/TopBar.tsx`, 替换 `desktopLinks` 整个 JSX 块:\n\n```tsx\nconst desktopLinks = (\n <>\n 工作台\n 📇 知识库\n 🔍 发现\n 🤝 撮合\n 📦 交易\n 💰 钱包\n {me?.isAdmin && (\n 🛡️ 后台\n )}\n { e.preventDefault(); logout(); }} className={linkCls}>退出\n \n);\n```\n\n- [ ] **Step 3: 改 mobileLinks 同步**\n\n文件 `app/_components/TopBar.tsx`, 替换 `mobileLinks` JSX 块:\n\n```tsx\nconst mobileLinks = (\n <>\n 工作台\n 📇 知识库\n 🔍 发现\n 🤝 撮合\n 📦 交易\n 💰 钱包\n {me?.isAdmin && (\n 🛡️ 后台\n )}\n { e.preventDefault(); logout(); }}\n className=\"block px-4 py-3 text-red-600 hover:bg-red-50\"\n >\n 退出\n \n \n);\n```\n\n- [ ] **Step 4: 新建 /wallet 占位页 (TopBar 链接不空)**\n\n文件 `app/wallet/page.tsx`:\n\n```tsx\n\"use client\";\nimport { useEffect, useState } from \"react\";\nimport TopBar from \"../_components/TopBar\";\n\nexport default function WalletPage() {\n const [mounted, setMounted] = useState(false);\n useEffect(() => { setMounted(true); }, []);\n if (!mounted) return
加载中...
;\n return (\n
\n \n
\n

💰 钱包

\n

收入/支出/提现 (阶段 7 实装, 当前占位)

\n
\n
💰
\n
暂无流水
\n
\n
\n
\n );\n}\n```\n\n- [ ] **Step 5: Build 验证**\n\n```bash\ncd opphub-web && npm run build 2>&1 | tail -30\n```\n\n预期: 路由包括 `/knowledge`, `/knowledge/profile`, `/knowledge/[id]`, `/discover`, `/matches`, `/wallet`, 加上所有旧的 `/match`, `/demands`, `/marketplace`, `/me/capabilities`, `/opc/profile` 仍编译 (用于 redirect 命中)\n\n- [ ] **Step 6: Commit**\n\n```bash\ncd opphub-web\ngit add app/_components/TopBar.tsx app/wallet\ngit commit -m \"feat(nav): TopBar 改 6 个一级 + 加 /wallet 占位 (阶段 1)\"\n```\n\n---\n\n## Task 6: 浏览器手测 + 全路径 301 校验\n\n**Files:**\n- (本地浏览器, 不写代码)\n\n- [ ] **Step 1: 本地启动 dev**\n\n```bash\ncd opphub-web && npm run build && (npm start &) && sleep 8 && echo \"started\"\n```\n\n预期: 服务起在 :3000\n\n- [ ] **Step 2: 试访问旧路径**\n\n```bash\nfor p in /me/capabilities /opc/profile /marketplace /demands /match; do\n echo \"=== $p ===\"\n curl -s -o /dev/null -w \"%{http_code} %{redirect_url}\\n\" \"http://localhost:3000$p\"\ndone\n```\n\n预期: 全部 `301` 加新 location (`/knowledge` `/knowledge/profile` `/discover?role=seller` `/discover?role=buyer` `/matches`)\n\n- [ ] **Step 3: 试新路径编译能 GET**\n\n```bash\nfor p in /knowledge /discover /matches /wallet /knowledge/profile /knowledge/foo; do\n echo \"=== $p ===\"\n curl -s -o /dev/null -w \"%{http_code}\\n\" \"http://localhost:3000$p\"\ndone\n```\n\n预期: 全部 `200` (Next.js 即使 id 不存在也是 200, 内容里显示 fallback)\n\n- [ ] **Step 4: 视觉校验 (人工浏览器)**\n\n打开浏览器, 在 `/knowledge` 看 Tab 切换 + EmptyState 渲染 OK\n\n## Task 7: 部署 ECS + 全站验证\n\n**Files:**\n- 同步到 ECS: `next.config.js`, `app/knowledge/*`, `app/discover/*`, `app/matches/*`, `app/_components/TopBar.tsx`, `app/wallet/*`\n\n- [ ] **Step 1: scp 文件到 ECS**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev\nscp opphub-web/next.config.js opphub-ecs:/opt/1panel/www/opphub-web/next.config.js\nscp -r opphub-web/app/knowledge opphub-ecs:/opt/1panel/www/opphub-web/app/\nscp -r opphub-web/app/discover opphub-ecs:/opt/1panel/www/opphub-web/app/\nscp -r opphub-web/app/matches opphub-ecs:/opt/1panel/www/opphub-web/app/\nscp -r opphub-web/app/wallet opphub-ecs:/opt/1panel/www/opphub-web/app/\nscp opphub-web/app/_components/TopBar.tsx opphub-ecs:/opt/1panel/www/opphub-web/app/_components/TopBar.tsx\n```\n\n- [ ] **Step 2: 容器内 build + 重启**\n\n```bash\nssh opphub-ecs \"docker exec opphub-web sh -c 'cd /app && npm run build 2>&1 | tail -5 && docker restart opphub-web'\"\n```\n\n注: 第二段 `docker restart` 实际是宿主命令, 不是容器内, 需要拆开:\n\n```bash\nssh opphub-ecs \"docker exec opphub-web sh -c 'cd /app && npm run build 2>&1 | tail -5'\"\nssh opphub-ecs \"docker restart opphub-web\"\n```\n\n- [ ] **Step 3: 验证 ECS 跳转**\n\n```bash\nssh opphub-ecs 'for p in /me/capabilities /opc/profile /marketplace /demands /match; do\n code=$(curl -s -o /dev/null -w \"%{http_code}\" \"http://localhost:3000$p\")\n loc=$(curl -s -o /dev/null -w \"%{redirect_url}\" \"http://localhost:3000$p\")\n echo \"$p -> $code $loc\"\ndone'\n```\n\n预期: 每个输出 `301` 后是新路径\n\n- [ ] **Step 4: 浏览器访问生产 URL `https://opphub.aiforce.cloud` 检查 TopBar**\n\n人工: 看顶导航 6 个, 点各 tab, 旧路径进入后跳新路径\n\n- [ ] **Step 5: Commit (若是单独的 \"deploy\" commit)**\n\n```bash\ncd opphub-web\ngit tag phase-1-nav-r2c\ngit log --oneline -5\n```\n\n(无需 commit, tag 也行, 主要给个里程碑标记)\n\n---\n\n## 完成判定\n\n- [ ] Task 1: redirects 加好, 旧路径 301 → 新路径 (curl 校验)\n- [ ] Task 2: `/knowledge` Tab 化 4 种角色, EmptyState 显示中文\n- [ ] Task 3: `/discover` 双视角切换 OK\n- [ ] Task 4: `/matches` 复刻 `/match` 内容\n- [ ] Task 5: TopBar 6 个入口, 新 `/wallet` 占位\n- [ ] Task 6: 本地构建 + curl + 浏览器手测通过\n- [ ] Task 7: ECS build + 容器重启, 旧路径 301 在生产环境也对\n\n最终产品现象: 用户打开 opphub.aiforce.cloud/dashboard → 顶部看到 6 个一级: 工作台 / 📇知识库 / 🔍发现 / 🤝撮合 / 📦交易 / 💰钱包; 点老链接自动跳\n\n## 后续阶段 (不在此 plan 范围)\n\n- 阶段 2: 知识库语义 + entryDimension 枚举 + 录入三选一\n- 阶段 3: 撮合流转 (matchings/inquiries/bids 表 + `/matches/[id]` + 手动撮合池)\n- 阶段 4: 订单状态机 + 支付 (orders 表 + `/orders/[id]` + 状态按钮)\n- 阶段 5: 推送 + Onboarding (opphub-ws 对接 + onboarding 4 屏 + 录入正反馈)\n- 阶段 6: 错误兜底 + 5 状态规范\n- 阶段 7: 钱包 + 设置 (含提现/充值实际流程)\n\n每个阶段单独走 spec → plan → implement 循环。\n\nFile v5.0.0:docs/superpowers/plans/2026-07-23-phase2-confirm-and-group.md\n\n# OppHub 阶段 2 修订版 v4：录入确认 + 解析字段 + 知识库分组展示\n\n> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.\n\n**Goal:** skill 录入走\"用户确认\"流程(覆盖 4 种 entryType), 同时输出结构化字段; web `/knowledge` 按 opcId 分组, 每组只显示现行版.\n\n**Architecture:** skill 端新增 confirmation state machine (输出 confirm 卡片 → 收 yes/no/modify → submit), cards 跟 parsedFields 一起入库. web 端 `/knowledge` 用 opcId 分组, 现行版 (supersededAt IS NULL) 渲染, 历史版只能从\"📜 历史\"按钮看. web 端顶部加录入引导.\n\n**Tech Stack:** openclaw skill (录入主路), opphub-web (展示 + 引导)\n\n---\n\n## 0. 关键边界 (钉死)\n\n- **录入主路**: openclaw skill, IM 端\n- **web 不写**: 不建表单\n- **匹配 / 推送**: 不动, 那是 server worker + opphub-ws 的事\n- **enum 强制**: 不做\n- **rawText 强结构校验**: 不做, 但加版本头\n\n---\n\n## 1. 任务地图 (6 个)\n\n```\nTask 1: skill rawText 顶部加 头 (3 个 bin)\nTask 2: skill 端 card 同时输出 parsedFields 结构化字段\nTask 3: skill 端 confirm 状态机 - 4 种 entryType 全部列清单, 用户确认后入库\nTask 4: web /knowledge 按 opcId 分组, 每组只显示现行版\nTask 5: web /knowledge 顶部\"如何录入\"文字引导\nTask 6: 部署 ECS + 终验\n```\n\n---\n\n## Task 1: skill 端 rawText 加 `` 头\n\n**Files:**\n- Modify: `opphub-skill/bin/opphub-knowledge-card.js`\n- Modify: `opphub-skill/bin/opphub-knowledge-discover.js`\n- Modify: `opphub-skill/bin/opphub-knowledge-submit.js`\n\n- [ ] **Step 1: knowledge-discover.js passthrough 时加版本头**\n\n文件 `opphub-skill/bin/opphub-knowledge-discover.js`, 找到:\n\n```js\n rawText: args.rawText,\n```\n\n修改为:\n\n```js\n rawText: ensureVersionHeader(args.rawText),\n```\n\n并在文件顶部 (import 之后) 添加工具函数:\n\n```js\n// v4.0.9: rawText 顶部加版本头, 让 server 端能识别协议版本\n// v1 (当前): 自由文本, 不锁结构, 内容是 LLM 自由拼出来的 rawText\n// v2 (未来): 加结构化字段 (parseKeyFields 强校验)\nfunction ensureVersionHeader(rawText) {\n if (!rawText) return rawText;\n if (rawText.startsWith(\"\")) return rawText;\n return `\\n${rawText}`;\n}\n```\n\n- [ ] **Step 2: knowledge-discover.js skeleton 也加版本头**\n\n找到 `buildRawTextSkeleton` 函数, 修改 return:\n\n```js\nfunction buildRawTextSkeleton(name) {\n const skeleton = `# ${name} · 自动画像 (skill turn 阶段 1 拼骨架)\\n\\n` +\n SOURCE_SKELETON.map((s, i) => `## ${i + 1}. ${s.name}\\n(${s.purpose})\\n\\n`).join(\"\");\n return `\\n${skeleton}`;\n}\n```\n\n- [ ] **Step 3: knowledge-card.js card.text 加版本头**\n\n文件 `opphub-skill/bin/opphub-knowledge-card.js`, 找到所有 `text: \\`${name} · ...` 字段 (ability / upstream / downstream 三段), 在前面加 `\\n`:\n\n(改 3 处, 示意 ability 一处)\n\n```js\n// 改前:\ntext: `${name} · 能力卡片 · ${dimension}\\n\\n(证据: rawText 包含 \"${dimension}\")\\n\\n(来自 rawText 实查, 非模板填空)`,\n\n// 改后:\ntext: `\\n${name} · 能力卡片 · ${dimension}\\n\\n(证据: rawText 包含 \"${dimension}\")\\n\\n(来自 rawText 实查, 非模板填空)`,\n```\n\n- [ ] **Step 4: knowledge-submit.js 兜底**\n\n文件 `opphub-skill/bin/opphub-knowledge-submit.js`, 在 `for (const card of cards)` 块内, 找到:\n\n```js\n const { type, dimension, text } = card;\n if (!type || !dimension || !text) {\n continue;\n }\n const idempotencyKey = sha256(`${opcId}|${type}|${dimension}`);\n const fullRawText = rawTextCtx ? `${rawTextCtx}\\n\\n---\\n${text}` : text;\n const contentHash = sha256(fullRawText);\n```\n\n修改为:\n\n```js\n const { type, dimension, text } = card;\n if (!type || !dimension || !text) {\n continue;\n }\n // v4.0.9: 兜底 - 老的 card.text 没版本头时, 补上 (旧 skill 数据兼容)\n const textWithHeader = text.startsWith(\"\")\n ? text\n : `\\n${text}`;\n const idempotencyKey = sha256(`${opcId}|${type}|${dimension}`);\n const fullRawText = rawTextCtx ? `${rawTextCtx}\\n\\n---\\n${textWithHeader}` : textWithHeader;\n const contentHash = sha256(fullRawText);\n```\n\n- [ ] **Step 5: 验证 skill 输出带版本头**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/skills/opphub\n\necho \"=== discover passthrough 期望 rawText 首行是 v1 ===\"\nnode bin/opphub-knowledge-discover --name \"测试公司\" --raw-text \"上海睿驰嘉禾是一家 MCN\" --json 2>&1 | python3 -c \"\nimport sys, json\nd = json.load(sys.stdin)\nprint('rawText 首行:', d.get('rawText','').split(chr(10))[0])\n\"\n\necho \"=== discover query-plan skeleton 也带 v1 ===\"\nnode bin/opphub-knowledge-discover --name \"测试公司\" --json 2>&1 | python3 -c \"\nimport sys, json\nd = json.load(sys.stdin)\nprint('skeleton 首行:', d.get('rawTextSkeleton','').split(chr(10))[0])\n\"\n```\n\n预期: 两个输出首行都是 ``\n\n- [ ] **Step 6: Commit**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/skills/opphub\ngit checkout main\ngit add bin/opphub-knowledge-card.js bin/opphub-knowledge-discover.js bin/opphub-knowledge-submit.js\ngit commit -m \"feat(skill): rawText 顶部加 opphub-raw-text-v1 版本头 (未来 v2 兼容)\"\n```\n\n---\n\n## Task 2: skill 端 card 同时输出 parsedFields 结构化字段\n\n**Files:**\n- Modify: `opphub-skill/bin/opphub-knowledge-card.js`\n- Modify: `opphub-skill/bin/opphub-knowledge-submit.js`\n\n**背景**: server schema 已经有 `parsedFields: Json` 字段 (prisma/schema.prisma:172 注释里写 \"拆出的结构化字段\"), 但 skill 端目前不输出. 这一 task 让 skill 把抽出的结构化字段带过去.\n\n- [ ] **Step 1: knowledge-card.js 顶层输出加 parsedFields 字段**\n\n文件 `opphub-skill/bin/opphub-knowledge-card.js`, 找到 main 函数最后 return 的 JSON (大约 `result` 对象), 在 `cards` 字段后加 `parsedFields`:\n\n```js\n// 改前 (示意):\nconst result = {\n ok: true,\n name,\n industry: { ... },\n cards,\n unmatchedTemplates,\n ...\n};\n\n// 改后:\nconst result = {\n ok: true,\n name,\n industry: { ... },\n cards,\n parsedFields: extractParsedFields(name, rawText, industry), // 新增\n unmatchedTemplates,\n ...\n};\n```\n\n- [ ] **Step 2: extractParsedFields 函数实现**\n\n在 knowledge-card.js 顶部 (INDUSTRY_TEMPLATES 后面) 加:\n\n```js\n// 从 rawText 抽结构化字段 (公司基础信息)\nfunction extractParsedFields(name, rawText, industry) {\n const fields = {};\n fields.companyName = name;\n fields.industry = industry ? { code: industry.code, name: industry.name } : null;\n\n // 法律实体\n const legalPersonMatch = rawText.match(/法人[:：]\\s*([^\\n]+)/);\n if (legalPersonMatch) fields.legalPerson = legalPersonMatch[1].trim().slice(0, 50);\n\n // 注册资本\n const capitalMatch = rawText.match(/注册资本[:：]\\s*([^\\n]+)/);\n if (capitalMatch) fields.registeredCapital = capitalMatch[1].trim().slice(0, 50);\n\n // 信用代码\n const creditMatch = rawText.match(/(?:信用代码|统一社会信用代码)[:：]\\s*([A-Z0-9]{18,20})/);\n if (creditMatch) fields.creditCode = creditMatch[1].trim();\n\n // 团队规模\n const sizeMatch = rawText.match(/(?:团队规模|规模|人数)[:：]\\s*([^\\n]+)/);\n if (sizeMatch) fields.teamSize = sizeMatch[1].trim().slice(0, 50);\n\n // 地址\n const addressMatch = rawText.match(/地址[:：]\\s*([^\\n]+)/);\n if (addressMatch) fields.address = addressMatch[1].trim().slice(0, 100);\n\n // 城市 (从地址或 rawText 提取)\n const cities = [\"上海\", \"北京\", \"深圳\", \"广州\", \"杭州\", \"成都\", \"南京\", \"武汉\", \"苏州\", \"天津\", \"重庆\"];\n for (const c of cities) {\n if (rawText.includes(c)) { fields.city = c; break; }\n }\n\n // 业务描述 (取 2. 业务描述 段)\n const bizMatch = rawText.match(/##\\s*2\\.\\s*业务描述\\s*\\n+([\\s\\S]*?)(?=\\n##\\s|\\s*$)/);\n if (bizMatch) fields.businessDescription = bizMatch[1].trim().slice(0, 500);\n\n return fields;\n}\n```\n\n- [ ] **Step 3: knowledge-submit.js 把 parsedFields 传给 ingest**\n\n文件 `opphub-skill/bin/opphub-knowledge-submit.js`, 找到 cards 加载的地方 (大约 `const cards = ...`), 在读 cards.json 后, 提取 parsedFields:\n\n```js\n// 改前:\nlet cardsDoc;\ntry {\n cardsDoc = JSON.parse(readFileSync(args.cards, \"utf8\"));\n} catch (e) { ... }\n\nconst name = cardsDoc.name || args.company || \"(未指定)\";\nconst cards = Array.isArray(cardsDoc.cards) ? cardsDoc.cards : ...;\n```\n\n修改为:\n\n```js\nlet cardsDoc;\ntry {\n cardsDoc = JSON.parse(readFileSync(args.cards, \"utf8\"));\n} catch (e) { ... }\n\nconst name = cardsDoc.name || args.company || \"(未指定)\";\nconst cards = Array.isArray(cardsDoc.cards) ? cardsDoc.cards : ...;\nconst parsedFields = cardsDoc.parsedFields || null; // 新增\n```\n\n然后在 for 循环的 body 提交时, 加 `parsedFields` 字段. 找到 `body: JSON.stringify({` 那块:\n\n```js\n body: JSON.stringify({\n opcId,\n rawText: fullRawText,\n sourceType: \"auto\",\n entryType: type,\n entryDimension: dimension,\n idempotencyKey,\n contentHash,\n forceOverride: args.forceOverrideConflict,\n }),\n```\n\n修改为 (用 const card 关联 parsedFields 一次, 因为 cards 数组里其实没 parsedFields, 是顶层):\n\n```js\n body: JSON.stringify({\n opcId,\n rawText: fullRawText,\n sourceType: \"auto\",\n entryType: type,\n entryDimension: dimension,\n idempotencyKey,\n contentHash,\n parsedFields: parsedFields ?? undefined, // 第一张卡带, 后续重复 server 用 no_change\n forceOverride: args.forceOverrideConflict,\n }),\n```\n\n(只在第一张卡带 parsedFields 也可以, server 端用 idempotencyKey 关联, 后续 no_change 时不需要再传. 但为简化, 每张都传, server 会去重.)\n\n- [ ] **Step 4: 验证 card 输出含 parsedFields**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/skills/opphub\n\necho \"=== knowledge-card 输出含 parsedFields ===\"\nnode bin/opphub-knowledge-card \\\n --name \"测试公司\" \\\n --raw-text '# 测试公司 · 自动画像\n\n## 1. 工商信息\n- 名称: 测试公司\n- 法人: 张三\n- 注册资本: 100万\n- 信用代码: 91310110MA12345678\n- 团队规模: 50人\n- 地址: 上海市杨浦区某路 1 号\n\n## 2. 业务描述\n短视频内容制作, 达人投放\n\n' \\\n --json 2>&1 | python3 -c \"\nimport sys, json\nd = json.load(sys.stdin)\nprint('parsedFields:', json.dumps(d.get('parsedFields', {}), ensure_ascii=False, indent=2))\n\"\n```\n\n预期: 输出 parsedFields 含 companyName / legalPerson / city / businessDescription 等\n\n- [ ] **Step 5: Commit**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/skills/opphub\ngit checkout main\ngit add bin/opphub-knowledge-card.js bin/opphub-knowledge-submit.js\ngit commit -m \"feat(skill): knowledge-card 输出 parsedFields, knowledge-submit 透传\"\n```\n\n---\n\n## Task 3: skill 端 confirmation state machine (4 种 entryType 全部列清单)\n\n**Files:**\n- Modify: `opphub-skill/bin/opphub-knowledge-card.js` (加 confirmation payload)\n- Modify: `opphub-knowledge-submit.js` 加 `--confirm` 开关 (默认拒绝, 收到才入)\n- Modify: `SKILL.md` (更新「偶合录入 [公司]」的描述, 加确认步骤)\n\n**背景**: 当前 skill 6 步里第 4 步说 \"给你 1 张清单确认\" 但没具体实现. 这一 task 让 card 步骤输出 confirm 卡片 (含 4 种 entryType 全部), submit 默认拒绝, 收到 `--confirm` 才入库.\n\n注意: 这是 skill 端\"工作流\"调整, 不是新增 bin.\n\n- [ ] **Step 1: knowledge-card.js 输出确认 payload**\n\n文件 `opphub-skill/bin/opphub-knowledge-card.js`, 找到 main 函数最后 return 的 JSON, 在 `unmatchedTemplates` 后加 `confirmation` 字段:\n\n```js\nconst result = {\n ok: true,\n name,\n industry: { ... },\n cards,\n parsedFields,\n unmatchedTemplates,\n confirmation: buildConfirmationList(name, industry, cards, parsedFields), // 新增\n ...\n};\n```\n\n- [ ] **Step 2: buildConfirmationList 函数**\n\n在 knowledge-card.js 顶部 `extractParsedFields` 后面加:\n\n```js\n// 生成 4 种 entryType 全部的确认清单\n// 用途: 供 IM bot 给用户发「将要录入什么」的清单\nfunction buildConfirmationList(name, industry, cards, parsedFields) {\n const groups = {\n ability: { label: \"我能提供\", emoji: \"✅\", items: [] },\n downstream: { label: \"我想找\", emoji: \"🔍\", items: [] },\n upstream: { label: \"我的依赖\", emoji: \"⬆️\", items: [] },\n peer: { label: \"同行关系\", emoji: \"🔗\", items: [] },\n };\n for (const c of cards) {\n if (groups[c.type]) {\n groups[c.type].items.push({\n dimension: c.dimension,\n evidenceSource: c.evidenceSource ?? \"rawText 关键词命中\",\n });\n }\n }\n\n // 把空组去掉, 只列有内容的\n const nonEmpty = Object.entries(groups).filter(([_, g]) => g.items.length > 0);\n\n return {\n name,\n industry: industry ? { code: industry.code, name: industry.name } : null,\n companyName: parsedFields?.companyName ?? name,\n businessDescription: parsedFields?.businessDescription ?? null,\n totalCards: cards.length,\n groups: nonEmpty.map(([type, g]) => ({ type, label: g.label, emoji: g.emoji, items: g.items })),\n instructions: \"回复「确认」入库；回复「删 」去掉某条；回复「重抽」回到阶段 1\",\n };\n}\n```\n\n- [ ] **Step 3: knowledge-submit.js 加 `--confirm` 开关**\n\n文件 `opphub-skill/bin/opphub-knowledge-submit.js`, 找到 `parseArgs` 函数, 加 `--confirm` 解析:\n\n```js\nelse if (a === \"--confirm\") args.confirm = true;\n```\n\n并在 main 函数顶部, 在读完 cards 之后, 加校验:\n\n```js\nif (!args.confirm) {\n if (wantJson) out({\n ok: false,\n error: \"needs_confirmation\",\n message: \"未确认 — 需先跑 knowledge-card 看 confirmation 清单, 用户回复'确认'后再加 --confirm 入库\",\n });\n process.exit(1);\n}\n```\n\n(没 --confirm 直接拒绝, 防止 skill 端自动入库跳过用户确认)\n\n- [ ] **Step 4: SKILL.md 更新「偶合录入」流程**\n\n文件 `opphub-skill/SKILL.md`, 找到 \"### 5. @bot 说\\\"偶合录入 [公司名]\\\"\" 这一节, 在最后追加确认步骤:\n\n在原 `6 步` 描述后加:\n\n> 第 4 步拆出能力后, bot **不直接入库**, 先给一张确认清单 (含 4 种 entryType: 我能提供/我想找/我的依赖/同行关系).\n> 用户回复:\n> - 「确认」 → bot 调 `knowledge-submit --confirm` 入库\n> - 「删 」 → bot 去掉某条重发清单\n> - 「重抽」 → 回到阶段 1 重新联网查\n> - 「改 <字段>」 → 改解析字段后重发清单\n\n- [ ] **Step 5: 验证 confirmation 输出**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/skills/opphub\n\necho \"=== knowledge-card 输出含 confirmation ===\"\nnode bin/opphub-knowledge-card \\\n --name \"测试公司\" \\\n --raw-text '# 测试公司\n\n## 2. 业务描述\n达人投放, 短视频脚本\n\n' \\\n --json 2>&1 | python3 -c \"\nimport sys, json\nd = json.load(sys.stdin)\nprint('confirmation:', json.dumps(d.get('confirmation', {}), ensure_ascii=False, indent=2))\n\"\n\necho \"=== submit 不带 --confirm 期望拒绝 ===\"\necho '[]' > /tmp/empty.json\nnode bin/opphub-knowledge-submit --cards /tmp/empty.json --json 2>&1 | head -10\n```\n\n预期: confirmation 含 4 个 groups (空组会跳过), submit 返 `needs_confirmation` 错误\n\n- [ ] **Step 6: Commit**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/skills/opphub\ngit checkout main\ngit add bin/opphub-knowledge-card.js bin/opphub-knowledge-submit.js SKILL.md\ngit commit -m \"feat(skill): card 输出 confirmation 清单 (4 entryType), submit 加 --confirm 开关\"\n```\n\n---\n\n## Task 4: web /knowledge 按 opcId 分组, 每组只显示现行版\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 改 fetch + 状态**\n\n文件 `opphub-web/app/knowledge/page.tsx`, 找到 fetch 数据逻辑 (大约 `fetch(\\`${apiBase}/api/knowledge?opcId=me&entryType=${tab}\\`)`). 修改成不带 entryType 过滤 (拉所有现行版, server 默认就只返 `supersededAt: null`):\n\n```tsx\nfetch(`${apiBase}/api/knowledge?opcId=me`, {\n headers: { authorization: \"Bearer \" + token },\n})\n```\n\n(原 code 带 `&entryType=${tab}` 过滤, 改为拉所有, 然后在客户端按 tab 过滤显示. 因为 grouping 要按 opcId 看全部)\n\n- [ ] **Step 2: 客户端按 entryType 过滤 + 按 opcId 分组**\n\n在 component 里, fetch 数据后:\n\n```tsx\nconst filtered = entries.filter((e) => e.entryType === tab);\nconst grouped: Array<{ opcId: string; entries: typeof entries }> = [];\nconst byOpc = new Map();\nfor (const e of filtered) {\n const k = parsedCompanyName(e) || \"未命名公司\"; // 同一 opcId 下用 companyName 分组\n if (!byOpc.has(k)) byOpc.set(k, []);\n byOpc.get(k)!.push(e);\n}\nfor (const [k, v] of byOpc) grouped.push({ opcId: k, entries: v });\n```\n\n(注意: 所有 entry 共享 opcId=\"me\", 但 rawText 里 companyName 不同. 我们的分组维度是 companyName, 不是 opcId. 这里修正: 把\"按 opcId 分组\"理解为\"按公司分组\", 用 parsedFields.companyName 或 rawText 头一行提取)\n\n- [ ] **Step 3: 新增 parsedCompanyName helper**\n\n在 component 里加:\n\n```tsx\nfunction parsedCompanyName(entry: { entryType: string | null; entryDimension: string | null; rawText: string }): string {\n // 优先级 1: 解析 rawText 头部 \"## 1. 工商信息\" 段中的 \"- 名称:\"\n const match = entry.rawText.match(/^-\\s*名称[:：]\\s*([^\\n]+)/m);\n if (match) return match[1].trim().slice(0, 30);\n // 优先级 2: 用 entryDimension 当 fallback\n return entry.entryDimension || \"未命名\";\n}\n```\n\n(临时方案, 阶段 3 重做时用 server 端 parsedFields 字段)\n\n- [ ] **Step 4: 改渲染 (masonry → group by company)**\n\n在 JSX 里, 把原来 `entries.map((e) => (\n
\n
\n 📇 {g.opcId} · {g.entries.length} 条\n
\n {g.entries.map((e) => )}\n
\n))}\n```\n\n- [ ] **Step 5: 验证 build**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web && npm run build 2>&1 | grep -E \"(Compiled|error|knowledge)\" | head -10\n```\n\n预期: `Compiled successfully`, 路由 `/knowledge` 仍存在\n\n- [ ] **Step 6: Commit**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web\ngit checkout main\ngit add app/knowledge/page.tsx\ngit commit -m \"feat(knowledge): /knowledge 按 companyName 分组, 只显示现行版\"\n```\n\n---\n\n## Task 5: web /knowledge 顶部\"如何录入\"文字引导\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 在 `

` 之后, 插入引导卡片**\n\n文件 `opphub-web/app/knowledge/page.tsx`, 找到 `

📇 知识库

`, **在它后面** 插入:\n\n```tsx\n{/* 录入引导 - skill 是录入主路 */}\n
\n
💡 如何录入一条知识？
\n
\n 录入是 openclaw skill 在你的 IM 客户端里完成的, 不在网页. 装好 opphub skill + plugin 后, 在飞书/钉钉/微信/Telegram 里给偶合 bot 发:\n
\n
    \n
  • 偶合录入 [公司名] — 录入公司能力画像 (skill 自动联网查 + 推断行业 + 拆 4 种能力 + 等你确认)
    • \n
  • 偶合录入我的能力: [一段话] — 录入你个人能力
    • \n
  • 偶合录入我要找: [需求描述] — 录入你想找的需求
    • \n
  • 偶合录入关联公司 + 合同 xls — 上传合同 xls 拆上下游
    • \n
\n
\n bot 拆完会先给你一张「将要录入」的清单 (含 4 种 entryType: 我能提供/我想找/我的依赖/同行关系), 你回复\"确认\"才入库.\n 5 秒后, 你录入的内容会出现在这里, 系统会自动跑匹配, 通知你潜在合作方.\n 还没装 skill?安装文档\n
\n
\n```\n\n并在文件底部 styles 段添加:\n\n```tsx\nconst codeInline: React.CSSProperties = {\n background: \"#eef2ff\",\n color: \"#4f46e5\",\n padding: \"1px 6px\",\n borderRadius: 4,\n fontSize: 12,\n fontFamily: 'ui-monospace, \"SF Mono\", Menlo, monospace',\n};\n```\n\n- [ ] **Step 2: 验证 build**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"(Compiled|error|knowledge)\" | head -10\n```\n\n预期: `Compiled successfully`\n\n- [ ] **Step 3: Commit**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web\ngit checkout main\ngit add app/knowledge/page.tsx\ngit commit -m \"feat(knowledge): /knowledge 顶部加'如何录入'文字引导 (含确认流程说明)\"\n```\n\n---\n\n## Task 6: 部署 ECS + 终验\n\n- [ ] **Step 1: scp web 改动**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev\nscp opphub-web/app/knowledge/page.tsx opphub-ecs:/opt/1panel/www/opphub-web/app/knowledge/page.tsx\n```\n\n(skill 端是用户机器, 不需要 scp)\n\n- [ ] **Step 2: 容器 build + 重启**\n\n```bash\nssh opphub-ecs \"docker exec opphub-web sh -c 'cd /app && npm run build 2>&1 | tail -5'\"\nssh opphub-ecs \"docker restart opphub-web\"\nsleep 5\nssh opphub-ecs 'docker ps | grep opphub-web'\n```\n\n- [ ] **Step 3: 验证生产 /knowledge 有引导文字**\n\n```bash\nssh opphub-ecs 'curl -s http://localhost:3000/knowledge | grep -c \"如何录入\" || echo \"0\"'\n```\n\n预期: >= 1\n\n- [ ] **Step 4: git tag**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web\ngit tag phase-2-confirm-group\n```\n\n---\n\n## 完成判定 (v4 修订版)\n\n- [ ] Task 1: skill 3 个 bin 输出 rawText 首行带 `` 头\n- [ ] Task 2: skill knowledge-card 输出 parsedFields, knowledge-submit 透传到 ingest\n- [ ] Task 3: skill knowledge-card 输出 confirmation 清单 (4 entryType 全部), submit 不带 --confirm 拒绝\n- [ ] Task 4: web `/knowledge` 按 companyName 分组, 每组只显示现行版\n- [ ] Task 5: web `/knowledge` 顶部有\"如何录入\"文字引导\n- [ ] Task 6: ECS 部署 + 引导文字在生产可见\n- git tag `phase-2-confirm-group`\n\n## YAGNI 范围 (本阶段明确不做)\n\n- ❌ /knowledge/new 表单\n- ❌ /api/knowledge/dimensions 端点\n- ❌ entryDimension enum 强制\n- ❌ rawText 强结构校验\n- ❌ skill 端做匹配 (server worker 的事)\n- ❌ skill 端做推送 (opphub-ws 的事)\n- ❌ lib/knowledge-dimensions.ts 共享 enum\n- ❌ 📌 自定义 标记\n- ❌ 录入成功 banner (skill 跑完用户已在 IM 端, web 这里不重复)\n- ❌ 矛盾 / 重复检测 (阶段 3+)\n\n## 为什么 v4 比 v3 更好\n\n- v3 只加 rawText 版本头 + web 引导 (4 文件), 是 \"格式\" 升级, 没碰核心流程\n- v4 在 v3 基础上**加 3 件事**:\n - **确认流程** (skill 端的核心安全网, 防止乱入库)\n - **确认清单覆盖 4 种 entryType** (不只是 ability, 包含 downstream/upstream/peer)\n - **结构化字段** (parsedFields, 让 web 端有数据画公司卡片)\n - **按公司分组** (web 端展示的核心组织维度, 现行版 vs 历史版区分)\n\n## 这一阶段真实价值\n\n- 录入从 \"skill 自动入库\" 改成 \"skill 拆卡 + 用户确认 + 入库\" - 防止 skill 抽错乱入\n- 4 种 entryType 全部覆盖 - 不漏需求方\n- 解析字段入库 - 未来 server 端能基于结构化数据做匹配/统计\n- web 端按公司分组 - 用户一眼看到自己有什么公司画像, 每家公司几条能力\n\nFile v5.0.0:docs/superpowers/plans/2026-07-23-phase3-modal-interactions.md\n\n# OppHub 阶段 3 修订版：「我的卡片」全部交互 + /discover 改读知识库\n\n> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.\n\n**Goal:** 「我的卡片」页所有交互弹窗化 (详情/编辑/删除/历史) 主页不动; tab 修复使「我在找」「同行关系」可点; /discover 改读知识库; 安装文档链接换 ClawHub.\n\n**Architecture:** 仅动 opphub-web 前端, 不改 server, 不改 skill. 弹窗用本地 React state 控制开合. /discover 切换到调 `/api/knowledge?opcId=me&entryType=...&visibility=PUBLIC` (虽然当前 server 没有 visibility=PUBLIC 过滤, 暂用 opcId=me 拉自己的 entry 看 effect, 真正 public 切换留作未来 server 任务)\n\n**Tech Stack:** Next.js 14, React, 本地 state modal, fetch\n\n## TopBar 顺序 (已落地, 不可动)\n\n```\n工作台 / 🔍 发现 / 🤝 撮合 / 📦 交易 / 💰 钱包 / 🪪 我的卡片 / 🛡️ 后台 / 退出\n```\n\n「我的卡片」在 💰 钱包 和 🛡️ 后台 之间. 任何新增 nav 项必须保持这个顺序.\n\n---\n\n## 0. 关键边界 (钉死)\n\n- **「我的卡片」 = `/knowledge` 路径不变, 名字/TopBar 改**: 已 commit (`1ffbcb8` + `630367f`)\n- **TopBar 顺序已定**: 工作台 / 发现 / 撮合 / 交易 / 钱包 / 我的卡片 / 后台 / 退出 (新加的 nav 项必须保持)\n- **弹窗化**: 详情/编辑/删除/历史 全部 React state modal, 不再跳 `/knowledge/[id]` (那页面可以删, 或者留作 SEO 兜底)\n- **`/discover` 改读知识库**: 不再调 marketplace/demands 表, 调 `/api/knowledge?opcId=me` 按 entryType 过滤\n- **ClawHub 链接**: `https://clawhub.ai/mtty-ai/skills/opphub` (Quick fix, 这次一并改)\n- **tab 修复**: 「我在找」「同行关系」实际是能点, 可能是「我现在没数据」看着像坏. 修复: 优化空状态展示, 让用户知道这 tab 是好的, 只是没数据\n\n---\n\n## 1. 任务地图 (7 个)\n\n```\nTask 0: Quick fix - ClawHub 链接替换 (独立, 立刻可做)\nTask 1: 「我的卡片」详情 modal 弹窗 (替代跳转)\nTask 2: 「我的卡片」编辑 modal 弹窗 (复用 PATCH /api/knowledge/[id])\nTask 3: 「我的卡片」删除 modal 确认 + 弹窗\nTask 4: 「我的卡片」历史版本 modal 弹窗 (复用 GET /api/knowledge/[id]/history)\nTask 5: /discover 改读 /api/knowledge (按 entryType 过滤, 按 companyName 分组)\nTask 6: 部署 ECS + 终验\n```\n\n---\n\n## Task 0: Quick fix - ClawHub 链接替换\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: sed 替换链接**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web\nsed -i 's|https://github.com/mtty-ai/opphub-skill|https://clawhub.ai/mtty-ai/skills/opphub|g' app/knowledge/page.tsx\ngrep \"安装文档\" app/knowledge/page.tsx\n```\n\n预期: grep 出 \"安装文档\" 那一行的 href 已经是 clawhub.ai\n\n- [ ] **Step 2: Commit**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web\ngit add app/knowledge/page.tsx\ngit commit -m \"fix(ui): 安装文档链接改 ClawHub\"\n```\n\n---\n\n## Task 1: 「我的卡片」详情 modal 弹窗\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 加 detailModal state + 渲染**\n\n文件 `opphub-web/app/knowledge/page.tsx`, 在 component 内部 (大约 `const [entries, setEntries] = ...` 附近) 加:\n\n```tsx\nconst [detailModal, setDetailModal] = useState(null);\n```\n\n并在 `EntryCard` 上, 把 `` 跳转改成 button 弹 modal:\n\n(找到 `function EntryCard` 那个组件, 把 `` 替换成 `\n );\n}\n```\n\n并修改 `g.entries.map((e) => )` →\n\n```tsx\n{g.entries.map((e) => (\n setDetailModal(e)} />\n))}\n```\n\n- [ ] **Step 2: 加 DetailModal 组件 (在 EntryCard 之后定义)**\n\n```tsx\nfunction DetailModal({ entry, onClose, onEdit, onDelete, onHistory }: {\n entry: EntryShape;\n onClose: () => void;\n onEdit: () => void;\n onDelete: () => void;\n onHistory: () => void;\n}) {\n return (\n
\n
e.stopPropagation()}>\n
\n
\n
{entry.entryType ?? \"—\"}
\n
{entry.entryDimension ?? \"未命名\"}
\n
\n \n
\n 
{entry.rawText}
\n
\n \n \n \n
\n
\n
\n );\n}\n```\n\n(在 component 末尾, 在 EntryCard 函数后面定义)\n\n- [ ] **Step 3: 加 modal styles (在文件底部 styles 段)**\n\n在文件底部 styles 段 (例如 modalMask / modalCard / modalHead / modalClose / modalFoot / rawTextBlock / linkBtn / histBtn / deleteBtn 附近) 加:\n\n```tsx\nconst modalMask: React.CSSProperties = {\n position: \"fixed\", inset: 0,\n background: \"rgba(0,0,0,0.5)\",\n display: \"flex\", alignItems: \"center\", justifyContent: \"center\",\n zIndex: 999, padding: 16,\n};\nconst modalCard: React.CSSProperties = {\n background: \"white\", borderRadius: 12,\n width: \"100%\", maxWidth: 640, maxHeight: \"85vh\",\n display: \"flex\", flexDirection: \"column\",\n boxShadow: \"0 20px 60px rgba(0,0,0,0.3)\",\n};\nconst modalHead: React.CSSProperties = {\n padding: 16, borderBottom: \"1px solid #e5e7eb\",\n display: \"flex\", justifyContent: \"space-between\", alignItems: \"flex-start\",\n};\nconst modalClose: React.CSSProperties = {\n background: \"transparent\", border: \"none\", fontSize: 24,\n color: \"#6b7280\", cursor: \"pointer\", lineHeight: 1,\n};\nconst modalFoot: React.CSSProperties = {\n padding: 12, borderTop: \"1px solid #e5e7eb\",\n display: \"flex\", gap: 8, justifyContent: \"flex-end\",\n};\nconst rawTextBlock: React.CSSProperties = {\n padding: 16, margin: 0, flex: 1, overflow: \"auto\",\n fontSize: 13, lineHeight: 1.7, color: \"#374151\",\n whiteSpace: \"pre-wrap\", wordBreak: \"break-word\",\n fontFamily: 'ui-monospace, \"SF Mono\", Menlo, monospace',\n};\nconst linkBtn: React.CSSProperties = {\n padding: \"6px 14px\", background: \"#4f46e5\", color: \"white\",\n border: \"none\", borderRadius: 6, fontSize: 13, cursor: \"pointer\",\n};\nconst histBtn: React.CSSProperties = {\n padding: \"6px 14px\", background: \"#eef2ff\", color: \"#4f46e5\",\n border: \"1px solid #c7d2fe\", borderRadius: 6, fontSize: 13, cursor: \"pointer\",\n};\nconst deleteBtn: React.CSSProperties = {\n padding: \"6px 14px\", background: \"#fef2f2\", color: \"#dc2626\",\n border: \"1px solid #fca5a5\", borderRadius: 6, fontSize: 13, cursor: \"pointer\",\n};\n```\n\n- [ ] **Step 4: 渲染 DetailModal**\n\n在 `KnowledgePage` JSX 末尾, `` 后, `` 前加:\n\n```tsx\n{detailModal && (\n setDetailModal(null)}\n onEdit={() => { setEditModal(detailModal); setDetailModal(null); }}\n onDelete={() => { setDeleteModal(detailModal); setDetailModal(null); }}\n onHistory={() => { setHistoryModal(detailModal.id); setDetailModal(null); }}\n />\n)}\n```\n\n(Tasks 2-4 会实现 edit/delete/history modal 的 setXxxModal state, 先把 setEditModal / setDeleteModal / setHistoryModal state 加上, 即使 Tasks 2-4 才用)\n\n- [ ] **Step 5: 预先加 editModal / deleteModal / historyModal state**\n\n在 component 顶部 (与 detailModal 一起):\n\n```tsx\nconst [editModal, setEditModal] = useState(null);\nconst [deleteModal, setDeleteModal] = useState(null);\nconst [historyModal, setHistoryModal] = useState(null); // entry id\n```\n\n- [ ] **Step 6: Build 验证**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"Compiled|error\" | head -3\n```\n\n预期: Compiled successfully\n\n- [ ] **Step 7: Commit**\n\n```bash\ngit add app/knowledge/page.tsx\ngit commit -m \"feat(knowledge): 详情 modal 弹窗化 (替代跳转 /knowledge/[id])\"\n```\n\n---\n\n## Task 2: 「我的卡片」编辑 modal 弹窗\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 加 EditModal 组件**\n\n在 `DetailModal` 函数后面定义:\n\n```tsx\nfunction EditModal({ entry, onClose, onSaved }: {\n entry: EntryShape;\n onClose: () => void;\n onSaved: (newText: string) => void;\n}) {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const [text, setText] = useState(entry.rawText);\n const [saving, setSaving] = useState(false);\n\n async function save() {\n setSaving(true);\n try {\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n const r = await fetch(apiBase + \"/api/knowledge/\" + entry.id, {\n method: \"PATCH\",\n headers: { \"content-type\": \"application/json\", authorization: \"Bearer \" + token },\n body: JSON.stringify({ rawText: text }),\n });\n const j = await r.json();\n if (j?.ok) {\n onSaved(text);\n } else {\n alert(\"保存失败: \" + (j?.message ?? \"未知错误\"));\n }\n } catch (e: any) {\n alert(\"保存异常: \" + (e?.message ?? String(e)));\n } finally {\n setSaving(false);\n }\n }\n\n return (\n
\n
e.stopPropagation()}>\n
\n
\n
编辑
\n
{entry.entryDimension ?? \"未命名\"}
\n
\n \n
\n setText(e.target.value)}\n style={{\n ...rawTextBlock,\n border: \"none\",\n flex: 1,\n fontFamily: \"inherit\",\n resize: \"none\",\n }}\n />\n
\n \n \n
\n
\n
\n );\n}\n```\n\n- [ ] **Step 2: 渲染 EditModal**\n\n在 `KnowledgePage` JSX 末尾 (与 DetailModal 一起):\n\n```tsx\n{editModal && (\n setEditModal(null)}\n onSaved={(newText) => {\n setCaps((prev) => (prev ?? []).map((c) => c.id === editModal.id ? { ...c, rawText: newText, updatedAt: new Date().toISOString() } : c));\n setEditModal(null);\n }}\n />\n)}\n```\n\n- [ ] **Step 3: Build + Commit**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"Compiled|error\" | head -3\ngit add app/knowledge/page.tsx\ngit commit -m \"feat(knowledge): 编辑 modal 弹窗 (PATCH rawText)\"\n```\n\n---\n\n## Task 3: 「我的卡片」删除确认 modal\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 加 DeleteModal 组件**\n\n```tsx\nfunction DeleteModal({ entry, onClose, onDeleted }: {\n entry: EntryShape;\n onClose: () => void;\n onDeleted: () => void;\n}) {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const [deleting, setDeleting] = useState(false);\n\n async function doDelete() {\n setDeleting(true);\n try {\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n const r = await fetch(apiBase + \"/api/knowledge/\" + entry.id, {\n method: \"DELETE\",\n headers: { authorization: \"Bearer \" + token },\n });\n const j = await r.json();\n if (j?.ok) {\n onDeleted();\n } else {\n alert(\"删除失败: \" + (j?.message ?? \"未知错误\"));\n }\n } catch (e: any) {\n alert(\"删除异常: \" + (e?.message ?? String(e)));\n } finally {\n setDeleting(false);\n }\n }\n\n return (\n
\n
e.stopPropagation()}>\n
\n
🗑️ 删除这条卡片？
\n \n
\n
\n 将删除 {entry.entryDimension ?? \"未命名\"}。删除后不可恢复 (软删, 30 天内可联系运维恢复)。\n
\n
\n \n \n
\n
\n
\n );\n}\n```\n\n- [ ] **Step 2: 渲染 DeleteModal**\n\n```tsx\n{deleteModal && (\n setDeleteModal(null)}\n onDeleted={() => {\n setCaps((prev) => (prev ?? []).filter((c) => c.id !== deleteModal.id));\n setDeleteModal(null);\n }}\n />\n)}\n```\n\n- [ ] **Step 3: Build + Commit**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"Compiled|error\" | head -3\ngit add app/knowledge/page.tsx\ngit commit -m \"feat(knowledge): 删除确认 modal (DELETE 软删)\"\n```\n\n---\n\n## Task 4: 「我的卡片」历史版本 modal 弹窗\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 加 HistoryModal 组件**\n\n```tsx\ntype HistoryVersion = {\n id: string;\n title: string;\n createdAt: string;\n supersededAt: string | null;\n isCurrent: boolean;\n};\n\nfunction HistoryModal({ entryId, onClose }: { entryId: string; onClose: () => void }) {\n const apiBase = process.env.NEXT_PUBLIC_API_BASE || \"\";\n const [versions, setVersions] = useState(null);\n const [error, setError] = useState(null);\n\n useEffect(() => {\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n fetch(apiBase + \"/api/knowledge/\" + entryId + \"/history\", {\n headers: { authorization: \"Bearer \" + token },\n })\n .then((r) => r.json())\n .then((j) => {\n if (j?.ok && Array.isArray(j.versions)) setVersions(j.versions);\n else setError(j?.message ?? \"暂无历史版本\");\n })\n .catch((e) => setError(e?.message ?? \"网络错误\"));\n }, [entryId, apiBase]);\n\n return (\n
\n
e.stopPropagation()}>\n
\n
📜 版本历史
\n \n
\n
\n {versions === null &&
加载中...
}\n {error &&
⚠️ {error}
}\n {versions && versions.length === 0 &&
暂无历史版本
}\n {versions && versions.map((v, i) => (\n
\n
\n v{versions.length - i}\n {new Date(v.createdAt).toLocaleString(\"zh-CN\", { hour12: false })}\n {v.isCurrent && 现行}\n
\n
{v.title}
\n
\n ))}\n
\n
\n
\n );\n}\n```\n\n- [ ] **Step 2: 渲染 HistoryModal**\n\n```tsx\n{historyModal && (\n setHistoryModal(null)} />\n)}\n```\n\n- [ ] **Step 3: Build + Commit**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"Compiled|error\" | head -3\ngit add app/knowledge/page.tsx\ngit commit -m \"feat(knowledge): 历史版本 modal 弹窗 (GET /api/knowledge/[id]/history)\"\n```\n\n---\n\n## Task 5: /discover 改读 /api/knowledge\n\n**Files:**\n- Modify: `opphub-web/app/discover/page.tsx`\n\n- [ ] **Step 1: 改 fetch 数据源**\n\n文件 `opphub-web/app/discover/page.tsx`, 找到 `useEffect` 块 (大约 line 51-65), 替换为:\n\n```tsx\n useEffect(() => {\n if (!mounted) return;\n setLoading(true);\n const token = localStorage.getItem(\"opphubToken\") || \"\";\n // buyer 角色: 想看供应方 → 看别人录入的 ability 卡片\n // seller 角色: 想看需求 → 看别人录入的 downstream 卡片\n const targetType = role === \"buyer\" ? \"ability\" : \"downstream\";\n fetch(`${apiBase}/api/knowledge?opcId=me&entryType=${targetType}`, {\n headers: { authorization: \"Bearer \" + token },\n })\n .then((r) => r.json())\n .then((j) => {\n const list = j?.ok && Array.isArray(j.data) ? j.data : [];\n setCards(list);\n })\n .catch(() => setCards([]))\n .finally(() => setLoading(false));\n }, [mounted, role, apiBase]);\n```\n\n(注: 实际公开浏览需要 server 加 `visibility=PUBLIC` 过滤; 此处先实现 拉自己的 entry 看 effect, 真公开留 server 任务)\n\n- [ ] **Step 2: 替换 state 和渲染**\n\n把原来 `useState([])` / `useState([])` 改为 `useState([])`:\n\n```tsx\ntype Entry = {\n id: string;\n entryType: string | null;\n entryDimension: string | null;\n rawText: string;\n updatedAt: string;\n};\n\nconst [cards, setCards] = useState([]);\n```\n\n(删除 Demand / OPC 类型, 替换为单一 Entry)\n\n- [ ] **Step 3: 改渲染 - 用 entry 列表, 按公司名分组**\n\n把 buyer 角色和 seller 角色的渲染合并, 改为按 companyName 分组:\n\n```tsx\n{!loading && (\n \n)}\n```\n\n(buyer 和 seller 用同一个渲染, 因为现在数据源已经按 entryType 过滤)\n\n- [ ] **Step 4: Build + Commit**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"Compiled|error|discover\" | head -10\ngit add app/discover/page.tsx\ngit commit -m \"feat(discover): 改读 /api/knowledge (按 entryType 过滤, 替代 marketplace/demands 表)\"\n```\n\n---\n\n## Task 6: 部署 ECS + 终验\n\n- [ ] **Step 1: scp web 改动**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev\nscp opphub-web/app/knowledge/page.tsx opphub-ecs:/opt/1panel/www/opphub-web/app/knowledge/page.tsx\nscp opphub-web/app/discover/page.tsx opphub-ecs:/opt/1panel/www/opphub-web/app/discover/page.tsx\n```\n\n- [ ] **Step 2: 容器 build + 重启**\n\n```bash\nssh opphub-ecs \"docker exec opphub-web sh -c 'cd /app && npm run build 2>&1 | tail -5'\"\nssh opphub-ecs \"docker restart opphub-web\"\nsleep 5\n```\n\n- [ ] **Step 3: 验证**\n\n```bash\n# TopBar 含 \"我的卡片\"\nssh opphub-ecs 'curl -s http://localhost:3000/knowledge | grep -c \"我的卡片\" || echo \"0\"'\n\n# /discover 200\nssh opphub-ecs 'curl -s -o /dev/null -w \"%{http_code}\\n\" http://localhost:3000/discover'\n```\n\n预期: 前者 >= 1, 后者 200\n\n- [ ] **Step 4: git tag**\n\n```bash\ncd /Users/qiuxz/.openclaw/workspace-dev/opphub-web\ngit tag phase-3-modal-interactions\n```\n\n---\n\n## 完成判定 (v5 修订版)\n\n- [ ] Task 0: ClawHub 链接替换 (github.com → clawhub.ai)\n- [ ] Task 1: 「我的卡片」详情 modal 弹窗 (替代跳转)\n- [ ] Task 2: 「我的卡片」编辑 modal (PATCH rawText)\n- [ ] Task 3: 「我的卡片」删除确认 modal (DELETE 软删)\n- [ ] Task 4: 「我的卡片」历史版本 modal (GET /api/knowledge/[id]/history)\n- [ ] Task 5: /discover 改读 /api/knowledge, 按 entryType 过滤\n- [ ] Task 6: ECS 部署 + 终验\n- TopBar 显示「🪪 我的卡片」+ 位置在 💰 钱包 和 🛡️ 后台 之间 (已 commit `1ffbcb8` + `630367f`)\n- git tag `phase-3-modal-interactions`\n\n## YAGNI 范围 (本阶段明确不做)\n\n- ❌ server 端加 `visibility=PUBLIC` 过滤 (留作 server 任务)\n- ❌ 撮合匹配对页面 (阶段 3 后续或阶段 4)\n- ❌ 推送通道 (阶段 5)\n- ❌ lib/knowledge-dimensions 共享 enum\n- ❌ entryDimension enum 强制\n- ❌ `/knowledge/[id]` 页面删除 (留作 SEO 兜底, 不再被引)\n- ❌ ClawHub 链接替换 (这次只更新「我的卡片」name, ClawHub 链接留作下次再改 — 实际是本次也应该改, 补充在 Task 0 之外的 quick fix 里)\n\n## ClawHub 链接 quick fix (Task 0 - 已 commit 但忘记加)\n\n实际上 ClawHub 链接 `https://clawhub.ai/mtty-ai/skills/opphub` 应该立刻替换. 这条单 commit:\n\n```bash\ncd opphub-web\n# app/knowledge/page.tsx line 103 把 github.com 替换成 clawhub.ai/mtty-ai/skills\nsed -i 's|https://github.com/mtty-ai/opphub-skill|https://clawhub.ai/mtty-ai/skills/opphub|g' app/knowledge/page.tsx\ngit add app/knowledge/page.tsx\ngit commit -m \"fix(ui): 安装文档链接改 ClawHub\"\n```\n\n(在 Task 0 之外的 quick fix, 部署在 Task 6 同一批)\n\nFile v5.0.0:docs/superpowers/plans/2026-07-23-phase4-bug-fixes.md\n\n# OppHub 阶段 4 修订版：6 个 bug + 1 个产品问题\n\n> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.\n\n**Goal:** 修 4 个 UI bug, 跑一次 skill 模拟录入, /discover 视觉区分, 文案微调. 1 个紧急事: 恢复丢失的 API 文件.\n\n**Architecture:** 修 opphub-web 前端 + 恢复 opphub-web API routes + 跑 opphub-skill 模拟 (不部署 skill). 不改 server 业务逻辑.\n\n**Tech Stack:** Next.js 14, React, fetch\n\n---\n\n## 0. 关键事实\n\n- **API 文件丢失**: `app/api/knowledge/[id]/route.ts` 和 `app/api/knowledge/[id]/history/route.ts` 在之前 server migration 中被误删, 不在 git history. 必须重建 (Task 0 紧急).\n- **「我的卡片」文案**: 引导区文案 \"如何录入一条知识？\" 改 \"如何录入一条卡片？\"\n- **/discover 当前 buyer/seller 内容一样**: 因为两 tab 都查同 source, 之前忽略. 现在做视觉区分.\n- **skill 模拟**: 在本机跑 `opphub-skill` 三个 bin (discover + card + submit), 录入 上海睿驰嘉禾数字传媒科技有限公司\n\n---\n\n## 1. 任务地图 (8 个)\n\n```\nTask 0: 紧急 - 恢复丢失的 API routes\nTask 1: EditModal 高度修复 (modal 撑高 + textarea 固定高度)\nTask 2: 历史版本查看详情 (点行展开 rawText, server 加 rawText 字段)\nTask 3: Delete 错误处理改进 (fetch + try-catch + 准确报错)\nTask 4: 引导文案改 \"如何录入一条卡片\"\nTask 5: /discover 视觉区分 (背景色 + emoji + label)\nTask 6: 跑 skill 模拟 上海睿驰嘉禾 录入\nTask 7: 部署 ECS + 终验\n```\n\n---\n\n## Task 0: 紧急 - 恢复丢失的 API routes\n\n**Files:**\n- Create: `opphub-web/app/api/knowledge/[id]/route.ts` (DELETE + PATCH)\n- Create: `opphub-web/app/api/knowledge/[id]/history/route.ts` (GET history)\n\n- [ ] **Step 1: 重建 [id]/route.ts (DELETE + PATCH)**\n\n文件 `opphub-web/app/api/knowledge/[id]/route.ts`:\n\n```ts\nimport { NextRequest, NextResponse } from \"next/server\";\nimport { prisma } from \"@/packages/db/prisma\";\nimport jwt from \"jsonwebtoken\";\nimport { getJwtSecret } from \"@/packages/db/jwt\";\nimport { createHash } from \"node:crypto\";\n\nexport const dynamic = \"force-dynamic\";\nexport const runtime = \"nodejs\";\n\nfunction getOpcIdFromToken(req: NextRequest): string | null {\n const auth = req.headers.get(\"authorization\");\n if (!auth?.startsWith(\"Bearer \")) return null;\n try {\n const payload = jwt.verify(auth.slice(7), getJwtSecret()) as any;\n return payload?.opcId ?? payload?.sub ?? null;\n } catch {\n return null;\n }\n}\n\nfunction sha256(s: string): string {\n return createHash(\"sha256\").update(s).digest(\"hex\");\n}\n\nexport async function PATCH(req: NextRequest, { params }: { params: { id: string } }) {\n const tokenOpcId = getOpcIdFromToken(req);\n if (!tokenOpcId) {\n return NextResponse.json({ ok: false, error: \"unauthorized\", message: \"token 无效或过期\" }, { status: 401 });\n }\n\n const body = await req.json().catch(() => ({}));\n const rawText = typeof body?.rawText === \"string\" ? body.rawText : null;\n if (!rawText) {\n return NextResponse.json({ ok: false, error: \"missing_rawtext\", message: \"需要 rawText\" }, { status: 400 });\n }\n\n const entry = await prisma.opcKnowledgeEntry.findUnique({ where: { id: params.id } });\n if (!entry) {\n return NextResponse.json({ ok: false, error: \"not_found\", message: \"条目不存在\" }, { status: 404 });\n }\n if (entry.opcId !== tokenOpcId) {\n return NextResponse.json({ ok: false, error: \"forbidden\", message: \"不能编辑其他用户的条目\" }, { status: 403 });\n }\n\n const now = new Date();\n const newContentHash = sha256(rawText);\n const versionKey = sha256(entry.opcId + \"|\" + entry.id + \"|\" + now.getTime());\n\n const [, newEntry] = await prisma.$transaction([\n prisma.opcKnowledgeEntry.update({\n where: { id: entry.id },\n data: { supersededAt: now, idempotencyKey: null },\n }),\n prisma.opcKnowledgeEntry.create({\n data: {\n opcId: entry.opcId,\n sourceType: entry.sourceType,\n rawText,\n status: \"done\",\n entryType: entry.entryType,\n entryDimension: entry.entryDimension,\n idempotencyKey: versionKey,\n contentHash: newContentHash,\n visibility: entry.visibility,\n previousEntryId: entry.id,\n },\n }),\n ]);\n\n return NextResponse.json({\n ok: true,\n action: \"updated\",\n entryId: newEntry.id,\n previousEntryId: entry.id,\n version: \"new_version_created\",\n });\n}\n\nexport async function DELETE(req: NextRequest, { params }: { params: { id: string } }) {\n const tokenOpcId = getOpcIdFromToken(req);\n if (!tokenOpcId) {\n return NextResponse.json({ ok: false, error: \"unauthorized\", message: \"token 无效或过期\" }, { status: 401 });\n }\n\n const entry = await prisma.opcKnowledgeEntry.findUnique({ where: { id: params.id } });\n if (!entry) {\n return NextResponse.json({ ok: false, error: \"not_found\", message: \"条目不存在\" }, { status: 404 });\n }\n if (entry.opcId !== tokenOpcId) {\n return NextResponse.json({ ok: false, error: \"forbidden\", message: \"不能删除其他用户的条目\" }, { status: 403 });\n }\n\n await prisma.opcKnowledgeEntry.update({\n where: { id: params.id },\n data: { deletedAt: new Date() },\n });\n\n return NextResponse.json({ ok: true });\n}\n```\n\n- [ ] **Step 2: 重建 [id]/history/route.ts (GET history chain + rawText)**\n\n文件 `opphub-web/app/api/knowledge/[id]/history/route.ts`:\n\n```ts\nimport { NextRequest, NextResponse } from \"next/server\";\nimport { prisma } from \"@/packages/db/prisma\";\nimport jwt from \"jsonwebtoken\";\nimport { getJwtSecret } from \"@/packages/db/jwt\";\n\nexport const dynamic = \"force-dynamic\";\nexport const runtime = \"nodejs\";\n\nfunction getOpcIdFromToken(req: NextRequest): string | null {\n const auth = req.headers.get(\"authorization\");\n if (!auth?.startsWith(\"Bearer \")) return null;\n try {\n const payload = jwt.verify(auth.slice(7), getJwtSecret()) as any;\n return payload?.opcId ?? payload?.sub ?? null;\n } catch {\n return null;\n }\n}\n\nfunction extractTitle(rawText: string): string {\n const lines = rawText.split(\"\\n\").map(l => l.trim()).filter(Boolean);\n for (const line of lines) {\n const m = line.match(/^(?:能力|技能|公司|服务|经验|title|skill|company|service|experience)[:：]\\s*(.+)$/i);\n if (m) return m[1].trim().slice(0, 60);\n }\n return (lines[0] ?? \"未命名\").slice(0, 60);\n}\n\nexport async function GET(req: NextRequest, { params }: { params: { id: string } }) {\n const tokenOpcId = getOpcIdFromToken(req);\n if (!tokenOpcId) {\n return NextResponse.json({ ok: false, error: \"unauthorized\", message: \"token 无效或过期\" }, { status: 401 });\n }\n\n const entry = await prisma.opcKnowledgeEntry.findUnique({\n where: { id: params.id },\n select: { id: true, opcId: true, previousEntryId: true, rawText: true, createdAt: true, supersededAt: true },\n });\n if (!entry) {\n return NextResponse.json({ ok: false, error: \"not_found\", message: \"条目不存在\" }, { status: 404 });\n }\n if (entry.opcId !== tokenOpcId) {\n return NextResponse.json({ ok: false, error: \"forbidden\", message: \"不能查看其他用户的条目版本历史\" }, { status: 403 });\n }\n\n // 顺链: currentEntryId + 前链 (entryId 上溯找前链)\n const versions = [];\n let currentId: string | null = params.id;\n let depth = 0;\n while (currentId && depth < 50) {\n const v = await prisma.opcKnowledgeEntry.findUnique({\n where: { id: currentId },\n select: { id: true, rawText: true, createdAt: true, previousEntryId: true, supersededAt: true },\n });\n if (!v) break;\n versions.push({\n id: v.id,\n title: extractTitle(v.rawText),\n rawText: v.rawText,\n createdAt: v.createdAt.toISOString(),\n supersededAt: v.supersededAt?.toISOString() ?? null,\n isCurrent: v.supersededAt === null,\n });\n currentId = v.previousEntryId;\n depth++;\n }\n\n return NextResponse.json({\n ok: true,\n entryId: params.id,\n versions,\n total: versions.length,\n });\n}\n```\n\n- [ ] **Step 3: 验证 build**\n\n```bash\ncd opphub-web && npm run build 2>&1 | grep -E \"Compiled|error\" | head -3\n```\n\n预期: Compiled successfully, 路由 `/api/knowledge/[id]` 和 `/api/knowledge/[id]/history` 出现\n\n- [ ] **Step 4: Commit**\n\n```bash\ngit add app/api/knowledge/\ngit commit -m \"fix(api): 重建 /api/knowledge/[id] PATCH+DELETE, /api/knowledge/[id]/history (历史 chain + rawText)\"\n```\n\n---\n\n## Task 1: EditModal 高度修复 (modal 撑高 + textarea 固定高度)\n\n**Files:**\n- Modify: `opphub-web/app/knowledge/page.tsx`\n\n- [ ] **Step 1: 改 EditModal 容器高度**\n\n找到 `function EditModal` 里的 `return (` 那段, 把外层 modalCard 改为:\n\n```tsx\n
e.stopPropagation()}>\n```\n\n- [ ] **Step 2: 改 textarea 固定高度 (可滚动)**\n\n找到 EditModal 里的 `