codebee 0.1.65 → 0.1.67

package/app/core/flows.py

@@ -1,355 +1,401 @@
-# -*- coding: utf-8 -*-
-"""流程注册表重写：预置流程可编辑/可恢复默认 + 自定义流程极简配置。
-
-设计：
-  - 预置流程（code/novel/serial_novel/doc/...）**可编辑**（阈值/轮数/维度/章节数/
-    提示词），改动存 data/flows.json 的 overrides；「恢复默认」一键回滚；
-  - 自定义流程**只需名称 + 引擎**：其余全部有合理默认（产出文件名按 id 生成、
-    维度用引擎默认、阈值 7.0、轮数 2）；进阶项留空即默认，不阻塞创建。
-"""
-from __future__ import annotations
-
-import json
-import re
-import threading
-
-from . import paths, volumes
-
-_LOCK = threading.RLock()
-_FILE = paths.DATA_DIR / "flows.json"
-
-ENGINES = ("code", "review", "direct")
-
-# 引擎默认参数：自定义流程留空时的兜底
-ENGINE_DEFAULTS = {
-    "review": {"manuscript": "output.md", "rubric": ["内容", "结构", "表达"],
-               "threshold": 7.0, "rounds": 2, "best_of": 1},
-    "code": {"verify_command": ""},
-    # direct：无参数——目标+附件即全部输入，跑完即止
-}
-
-# icon 约定："i-*" = 前端精灵表单色线性图标（随日/夜主题黑白）；其他值（emoji）原样显示，
-# 供自定义流程兜底。预置图标不进 overrides（见 _EDITABLE），保证升级后老数据也拿到新图标。
-BUILTIN_FLOWS = [
-    {"id": "direct", "name": "直接执行", "icon": "i-chat", "engine": "direct", "builtin": True,
-     "goal_hint": "让 AI 直接做什么（一句话，可带附件）",
-     "note": "CodeBee 直连模型 API 干活（无 CLI 进程），无可用供应商时回退本机 CLI；无拆解/评审（快）"},
-    {"id": "code", "name": "代码", "icon": "i-code", "engine": "code", "builtin": True,
-     "goal_hint": "要实现/修复什么（一句话）",
-     "note": "按难度动态编排：简单任务走实现+验证短链，复杂任务增加规划、评审、修复与换将"},
-    {"id": "novel", "name": "小说", "icon": "i-book-open", "engine": "review", "builtin": True,
-     "manuscript": "manuscript.md",
-     "rubric": ["情节", "人物", "文笔", "节奏", "吸引力"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "写什么（题材 / 篇幅 / 风格）",
-     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "serial_novel", "name": "连载小说", "icon": "i-library", "engine": "review", "builtin": True,
-     "manuscript": "manuscript.md",
-     "rubric": ["情节", "人物", "文笔", "节奏", "吸引力"],
-     "threshold": 7.0, "rounds": 2,
-     "serial": {"chapters": 8, "words_per_chapter": 2500},
-     "goal_hint": "题材/受众/卖点 + 总字数（例：2 万字都市女频，可签约平台）",
-     "note": "大纲 → 逐章起草 → 每章多维评审修订 → 全局一致性评审 → 合并（可断点续跑）"},
-    {"id": "article", "name": "自媒体文章", "icon": "i-news", "engine": "review", "builtin": True,
-     "manuscript": "article.md",
-     "rubric": ["选题与标题", "开头吸引力", "结构节奏", "平台适配", "传播性"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "什么主题、投哪个平台（公众号/头条/知乎）、给谁看",
-     "note": "公众号/头条风格文章：起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "video_script", "name": "短视频脚本", "icon": "i-clapper", "engine": "review", "builtin": True,
-     "manuscript": "script.md",
-     "rubric": ["黄金3秒钩子", "节奏密度", "口播流畅", "画面可执行", "互动引导"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "什么主题、多长时长、投哪个平台（抖音/B站/视频号）",
-     "note": "分镜 + 口播脚本：起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "doc", "name": "文档", "icon": "i-file-text", "engine": "review", "builtin": True,
-     "manuscript": "document.md",
-     "rubric": ["准确性", "结构清晰", "表达流畅", "实用价值"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "要写什么文档、给谁看",
-     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "translation", "name": "翻译", "icon": "i-lang", "engine": "review", "builtin": True,
-     "manuscript": "translation.md",
-     "rubric": ["忠实度", "流畅度", "术语一致性", "风格贴合"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "翻译什么（源文本位置 / 目标语言 / 要求）",
-     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "rank_scan", "name": "扫榜选材", "icon": "i-chart", "engine": "direct", "builtin": True,
-     "goal_hint": "想写哪个方向（一句话，可留空默认分析总榜热门题材）",
-     "note": "抓取七猫+番茄排行榜公开数据 → AI 提炼跨平台热门题材/人设/差异化切入点（快档直出报告）"},
-    {"id": "research", "name": "调研报告", "icon": "i-file-search", "engine": "review", "builtin": True,
-     "manuscript": "report.md",
-     "rubric": ["全面性", "深度", "论据可靠", "可读性", "结论质量"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "调研什么问题、产出给谁用",
-     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "speech", "name": "演讲稿", "icon": "i-mic", "engine": "review", "builtin": True,
-     "manuscript": "speech.md",
-     "rubric": ["主题聚焦", "结构", "感染力", "语言风格"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "演讲场合 / 听众 / 时长 / 核心信息",
-     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "weekly_report", "name": "工作汇报", "icon": "i-clip-check", "engine": "review", "builtin": True,
-     "manuscript": "weekly.md",
-     "rubric": ["重点突出", "数据支撑", "结构清晰", "下一步可执行"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "周报/月报/述职？汇报给谁、做了什么（可贴流水账让小队提炼）",
-     "note": "周报/月报/述职材料：起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "email", "name": "商务邮件", "icon": "i-mail", "engine": "review", "builtin": True,
-     "manuscript": "email.md",
-     "rubric": ["目的明确", "语气得体", "信息完整", "简洁度"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "给谁写、要达成什么、对方背景与你们的关系",
-     "note": "商务/客户/跨部门邮件：起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "tech_proposal", "name": "技术方案", "icon": "i-blocks", "engine": "review", "builtin": True,
-     "manuscript": "proposal.md",
-     "rubric": ["可行性", "方案完整性", "风险识别", "成本与收益"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "要解决什么问题、约束条件（工期/技术栈/预算）、给谁评审",
-     "note": "技术选型/架构/实施方案：起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "resume", "name": "简历", "icon": "i-idcard", "engine": "review", "builtin": True,
-     "manuscript": "resume.md",
-     "rubric": ["真实可信", "岗位匹配", "成果量化", "关键词覆盖", "简洁度"],
-     "threshold": 7.0, "rounds": 2,
-     "goal_hint": "目标岗位 + 个人经历（可贴旧简历附件），几年经验投什么职级",
-     "note": "简历优化/定制：起草 → 多维评审 → 修订循环 → 发布门禁"},
-    {"id": "bid_doc", "name": "标书编制", "icon": "i-file-search", "engine": "review", "builtin": True,
-     "manuscript": "bid.md",
-     "rubric": ["应答完整性", "合规符合度", "方案针对性", "评分点覆盖", "商务清晰度"],
-     "threshold": 7.5, "rounds": 2,
-     "goal_hint": "招标文件（可作附件）+ 标段/包件 + 我方资质与业绩亮点",
-     "note": "标书/投标文件：起草 → 多维评审 → 修订循环 → 发布门禁"},
-]
-
-_ID_RE = re.compile(r"^[a-z][a-z0-9_-]{0,31}$")
-# 可编辑字段（预置流程的 overrides 与自定义流程共用同一套规范化）。
-# icon 刻意不进来：预置图标跟版本走（老 overrides 里钉的旧 emoji 会被忽略），
-# 自定义流程的 icon 在 upsert_flow 里直接落盘，不走 overrides。
-_EDITABLE = ("name", "goal_hint", "note", "manuscript", "rubric",
-             "threshold", "rounds", "serial", "draft_prompt", "critique_prompt",
-             "verify_command", "best_of")
-
-
-def _read():
-    try:
-        data = json.loads(_FILE.read_text(encoding="utf-8"))
-        return data if isinstance(data, dict) else {}
-    except Exception:
-        return {}
-
-
-def _write(data):
-    _FILE.parent.mkdir(parents=True, exist_ok=True)
-    tmp = _FILE.with_suffix(".tmp")
-    tmp.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
-    tmp.replace(_FILE)
-
-
-def _norm_serial(serial):
-    if not isinstance(serial, dict) or not serial.get("chapters"):
-        return None
-    try:
-        out = {"chapters": max(2, min(20, int(serial["chapters"]))),
-               "words_per_chapter": max(500, min(8000, int(serial.get("words_per_chapter") or 2500)))}
-    except Exception:
-        return None
-    # 分卷（可选）：显式卷表优先，其次每卷章数。流程模板里可以预置，
-    # 任务级 payload 仍可覆盖/关闭（0 = 不分卷）。
-    spec = volumes.norm_spec(serial.get("volumes"))
-    if spec:
-        out["volumes"] = spec
-    per = volumes.norm_per(serial.get("volume_chapters"))
-    if per:
-        out["volume_chapters"] = per
-    return out
-
-
-def _norm_rubric(rubric):
-    if isinstance(rubric, str):
-        rubric = re.split(r"[,，、\n]+", rubric)
-    if not isinstance(rubric, list):
-        return None
-    out = [str(d).strip() for d in rubric if str(d).strip()][:8]
-    return out or None
-
-
-def _apply_overrides(base, ov):
-    """把 overrides 合并到预置流程上（只认白名单字段）。"""
-    f = dict(base)
-    if not isinstance(ov, dict):
-        return f
-    for k in _EDITABLE:
-        if k not in ov:
-            continue
-        v = ov[k]
-        if k == "rubric":
-            v = _norm_rubric(v)
-            if v:
-                f["rubric"] = v
-        elif k == "serial":
-            v = _norm_serial(v)
-            if v:
-                f["serial"] = v
-            else:
-                f.pop("serial", None)
-        elif k == "threshold":
-            try:
-                f["threshold"] = max(1.0, min(10.0, float(v)))
-            except Exception:
-                pass
-        elif k == "rounds":
-            try:
-                f["rounds"] = max(1, min(5, int(v)))
-            except Exception:
-                pass
-        elif k == "best_of":
-            try:
-                f["best_of"] = max(1, min(3, int(v)))
-            except Exception:
-                pass
-        elif k in ("manuscript", "verify_command"):
-            s = re.sub(r"[\\/]+", "_", str(v or "")).strip()
-            s = re.sub(r"\.{2,}", "_", s).lstrip(".")
-            if s:
-                f[k] = s
-        elif k in ("draft_prompt", "critique_prompt"):
-            s = str(v or "").strip()
-            if s:
-                f[k] = s[:4000]
-            else:
-                f.pop(k, None)
-        else:
-            f[k] = str(v or "")[:80]
-    return f
-
-
-def _custom():
-    data = _read().get("flows")
-    return [f for f in data if isinstance(f, dict) and f.get("id")] if isinstance(data, list) else []
-
-
-def _overrides():
-    ov = _read().get("overrides")
-    return ov if isinstance(ov, dict) else {}
-
-
-def list_flows():
-    """自定义流程在前（便于看到新增），预置流程按定义顺序，均已套用 overrides。"""
-    with _LOCK:
-        custom, ov = _custom(), _overrides()
-    builtin = [_apply_overrides(b, ov.get(b["id"])) for b in BUILTIN_FLOWS]
-    for f in builtin:
-        f["edited"] = bool(ov.get(f["id"]))
-    return custom + builtin
-
-
-def get_flow(flow_id):
-    for f in list_flows():
-        if f["id"] == flow_id:
-            return f
-    return None
-
-
-def upsert_flow(payload):
-    """新增/更新流程。返回 (flow, 错误)。
-
-    - 预置流程 id：写入 overrides（可编辑、可恢复默认）；
-    - 新 id：创建自定义流程——**只需 name + engine**，其余字段留空走引擎默认。
-    """
-    fid = str(payload.get("id") or "").strip()
-    name = str(payload.get("name") or "").strip()
-    engine = str(payload.get("engine") or "review").strip()
-    if not _ID_RE.match(fid):
-        return None, "流程 ID 只能是小写字母开头的字母/数字/-/_（≤32 位）"
-    if engine not in ENGINES:
-        return None, "engine 必须是 code、review 或 direct"
-    builtin_ids = {b["id"] for b in BUILTIN_FLOWS}
-
-    if fid in builtin_ids:
-        base = next(b for b in BUILTIN_FLOWS if b["id"] == fid)
-        if name and name != base["name"]:
-            pass  # 允许改名
-        ov = dict(payload)
-        ov.pop("id", None)
-        ov["engine"] = base["engine"]   # 引擎不可改（避免语义错乱）
-        merged = _apply_overrides(base, ov)
-        with _LOCK:
-            data = _read()
-            ovs = data.get("overrides") if isinstance(data.get("overrides"), dict) else {}
-            ovs[fid] = {k: merged[k] for k in _EDITABLE if k in merged}
-            data["overrides"] = ovs
-            _write(data)
-        merged["edited"] = True
-        return merged, None
-
-    if not name:
-        return None, "名称不能为空"
-    flow = {"id": fid, "name": name[:20], "icon": str(payload.get("icon") or "✨")[:4],
-            "engine": engine, "builtin": False,
-            "goal_hint": str(payload.get("goal_hint") or "")[:80],
-            "note": str(payload.get("note") or "")[:80]}
-    if engine == "review":
-        dflt = ENGINE_DEFAULTS["review"]
-        ms = re.sub(r"[\\/]+", "_", str(payload.get("manuscript") or (fid + ".md"))).strip()
-        ms = re.sub(r"\.{2,}", "_", ms).lstrip(".") or (fid + ".md")
-        flow["manuscript"] = ms
-        flow["rubric"] = _norm_rubric(payload.get("rubric")) or list(dflt["rubric"])
-        try:
-            flow["threshold"] = max(1.0, min(10.0, float(payload.get("threshold") or dflt["threshold"])))
-        except Exception:
-            flow["threshold"] = dflt["threshold"]
-        try:
-            flow["rounds"] = max(1, min(5, int(payload.get("rounds") or dflt["rounds"])))
-        except Exception:
-            flow["rounds"] = dflt["rounds"]
-        try:
-            flow["best_of"] = max(1, min(3, int(payload.get("best_of") or dflt.get("best_of") or 1)))
-        except Exception:
-            flow["best_of"] = 1
-        serial = _norm_serial(payload.get("serial"))
-        if serial:
-            flow["serial"] = serial
-        for key in ("draft_prompt", "critique_prompt"):
-            v = str(payload.get(key) or "").strip()
-            if v:
-                flow[key] = v[:4000]
-    elif engine == "code":
-        flow["verify_command"] = str(payload.get("verify_command") or "")[:200]
-    # direct：无流程参数（目标+附件即全部输入）
-    with _LOCK:
-        data = _read()
-        flows = data.get("flows") if isinstance(data.get("flows"), list) else []
-        flows = [f for f in flows if isinstance(f, dict) and f.get("id") != fid]
-        flows.append(flow)
-        data["flows"] = flows
-        _write(data)
-    return flow, None
-
-
-def delete_flow(flow_id):
-    """删除自定义流程；预置流程只能「恢复默认」不能删除。"""
-    if flow_id in {b["id"] for b in BUILTIN_FLOWS}:
-        return "预置流程不可删除（可点「恢复默认」回滚改动）"
-    with _LOCK:
-        data = _read()
-        flows = data.get("flows") if isinstance(data.get("flows"), list) else []
-        kept = [f for f in flows if not (isinstance(f, dict) and f.get("id") == flow_id)]
-        if len(kept) == len(flows):
-            return "流程不存在"
-        data["flows"] = kept
-        _write(data)
-    return None
-
-
-def reset_flow(flow_id):
-    """预置流程恢复默认（清掉 overrides）。返回错误或 None。"""
-    if flow_id not in {b["id"] for b in BUILTIN_FLOWS}:
-        return "只有预置流程支持恢复默认"
-    with _LOCK:
-        data = _read()
-        ovs = data.get("overrides") if isinstance(data.get("overrides"), dict) else {}
-        if flow_id in ovs:
-            ovs.pop(flow_id, None)
-            data["overrides"] = ovs
-            _write(data)
-    return None
+# -*- coding: utf-8 -*-
+"""流程注册表重写：预置流程可编辑/可恢复默认 + 自定义流程极简配置。
+
+设计：
+  - 预置流程（code/novel/serial_novel/doc/...）**可编辑**（阈值/轮数/维度/章节数/
+    提示词），改动存 data/flows.json 的 overrides；「恢复默认」一键回滚；
+  - 自定义流程**只需名称 + 引擎**：其余全部有合理默认（产出文件名按 id 生成、
+    维度用引擎默认、阈值 7.0、轮数 2）；进阶项留空即默认，不阻塞创建。
+"""
+from __future__ import annotations
+
+import hashlib
+import json
+import re
+import threading
+
+from . import paths, volumes
+
+_LOCK = threading.RLock()
+_FILE = paths.DATA_DIR / "flows.json"
+
+ENGINES = ("code", "review", "direct")
+
+# 引擎默认参数：自定义流程留空时的兜底
+ENGINE_DEFAULTS = {
+    "review": {"manuscript": "output.md", "rubric": ["内容", "结构", "表达"],
+               "threshold": 7.0, "rounds": 2, "best_of": 1},
+    "code": {"verify_command": ""},
+    # direct：无参数——目标+附件即全部输入，跑完即止
+}
+
+# icon 约定："i-*" = 前端精灵表单色线性图标（随日/夜主题黑白）；其他值（emoji）原样显示，
+# 供自定义流程兜底。预置图标不进 overrides（见 _EDITABLE），保证升级后老数据也拿到新图标。
+BUILTIN_FLOWS = [
+    {"id": "direct", "name": "直接执行", "icon": "i-chat", "engine": "direct", "builtin": True,
+     "goal_hint": "让 AI 直接做什么（一句话，可带附件）",
+     "note": "CodeBee 直连模型 API 干活（无 CLI 进程），无可用供应商时回退本机 CLI；无拆解/评审（快）"},
+    {"id": "code", "name": "代码", "icon": "i-code", "engine": "code", "builtin": True,
+     "goal_hint": "要实现/修复什么（一句话）",
+     "note": "按难度动态编排：简单任务走实现+验证短链，复杂任务增加规划、评审、修复与换将"},
+    {"id": "novel", "name": "小说", "icon": "i-book-open", "engine": "review", "builtin": True,
+     "manuscript": "manuscript.md",
+     "rubric": ["情节", "人物", "文笔", "节奏", "吸引力"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "写什么（题材 / 篇幅 / 风格）",
+     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "serial_novel", "name": "连载小说", "icon": "i-library", "engine": "review", "builtin": True,
+     "manuscript": "manuscript.md",
+     "rubric": ["情节", "人物", "文笔", "节奏", "吸引力"],
+     "threshold": 7.0, "rounds": 2,
+     "serial": {"chapters": 8, "words_per_chapter": 2500},
+     "goal_hint": "题材/受众/卖点 + 总字数（例：2 万字都市女频，可签约平台）",
+     "note": "大纲 → 逐章起草 → 每章多维评审修订 → 全局一致性评审 → 合并（可断点续跑）"},
+    {"id": "article", "name": "自媒体文章", "icon": "i-news", "engine": "review", "builtin": True,
+     "manuscript": "article.md",
+     "rubric": ["选题与标题", "开头吸引力", "结构节奏", "平台适配", "传播性"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "什么主题、投哪个平台（公众号/头条/知乎）、给谁看",
+     "note": "公众号/头条风格文章：起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "video_script", "name": "短视频脚本", "icon": "i-clapper", "engine": "review", "builtin": True,
+     "manuscript": "script.md",
+     "rubric": ["黄金3秒钩子", "节奏密度", "口播流畅", "画面可执行", "互动引导"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "什么主题、多长时长、投哪个平台（抖音/B站/视频号）",
+     "note": "分镜 + 口播脚本：起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "doc", "name": "文档", "icon": "i-file-text", "engine": "review", "builtin": True,
+     "manuscript": "document.md",
+     "rubric": ["准确性", "结构清晰", "表达流畅", "实用价值"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "要写什么文档、给谁看",
+     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "translation", "name": "翻译", "icon": "i-lang", "engine": "review", "builtin": True,
+     "manuscript": "translation.md",
+     "rubric": ["忠实度", "流畅度", "术语一致性", "风格贴合"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "翻译什么（源文本位置 / 目标语言 / 要求）",
+     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "rank_scan", "name": "扫榜选材", "icon": "i-chart", "engine": "direct", "builtin": True,
+     "goal_hint": "想写哪个方向（一句话，可留空默认分析总榜热门题材）",
+     "note": "抓取七猫/番茄/起点/纵横四平台排行榜公开数据 → AI 提炼跨平台热门题材/人设/差异化切入点（快档直出报告）"},
+    {"id": "defect_retro", "name": "缺陷复盘", "icon": "i-clipboard", "engine": "direct", "builtin": True,
+     "goal_hint": "复盘哪个版本/时间段的缺陷（建议附带禅道或 Jira 导出 CSV）",
+     "note": "读取附带的缺陷导出数据 → AI 输出产品/开发/测试三视角复盘报告：整体画像、根因聚类、漏测分析、改进动作（快档直出报告）"},
+    {"id": "research", "name": "调研报告", "icon": "i-file-search", "engine": "review", "builtin": True,
+     "manuscript": "report.md",
+     "rubric": ["全面性", "深度", "论据可靠", "可读性", "结论质量"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "调研什么问题、产出给谁用",
+     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "speech", "name": "演讲稿", "icon": "i-mic", "engine": "review", "builtin": True,
+     "manuscript": "speech.md",
+     "rubric": ["主题聚焦", "结构", "感染力", "语言风格"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "演讲场合 / 听众 / 时长 / 核心信息",
+     "note": "起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "weekly_report", "name": "工作汇报", "icon": "i-clip-check", "engine": "review", "builtin": True,
+     "manuscript": "weekly.md",
+     "rubric": ["重点突出", "数据支撑", "结构清晰", "下一步可执行"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "周报/月报/述职？汇报给谁、做了什么（可贴流水账让小队提炼）",
+     "note": "周报/月报/述职材料：起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "email", "name": "商务邮件", "icon": "i-mail", "engine": "review", "builtin": True,
+     "manuscript": "email.md",
+     "rubric": ["目的明确", "语气得体", "信息完整", "简洁度"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "给谁写、要达成什么、对方背景与你们的关系",
+     "note": "商务/客户/跨部门邮件：起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "tech_proposal", "name": "技术方案", "icon": "i-blocks", "engine": "review", "builtin": True,
+     "manuscript": "proposal.md",
+     "rubric": ["可行性", "方案完整性", "风险识别", "成本与收益"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "要解决什么问题、约束条件（工期/技术栈/预算）、给谁评审",
+     "note": "技术选型/架构/实施方案：起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "resume", "name": "简历", "icon": "i-idcard", "engine": "review", "builtin": True,
+     "manuscript": "resume.md",
+     "rubric": ["真实可信", "岗位匹配", "成果量化", "关键词覆盖", "简洁度"],
+     "threshold": 7.0, "rounds": 2,
+     "goal_hint": "目标岗位 + 个人经历（可贴旧简历附件），几年经验投什么职级",
+     "note": "简历优化/定制：起草 → 多维评审 → 修订循环 → 发布门禁"},
+    {"id": "bid_doc", "name": "标书编制", "icon": "i-file-search", "engine": "review", "builtin": True,
+     "manuscript": "bid.md",
+     "rubric": ["应答完整性", "合规符合度", "方案针对性", "评分点覆盖", "商务清晰度"],
+     "threshold": 7.5, "rounds": 2,
+     "goal_hint": "招标文件（可作附件）+ 标段/包件 + 我方资质与业绩亮点",
+     "note": "标书/投标文件：起草 → 多维评审 → 修订循环 → 发布门禁"},
+]
+
+_ID_RE = re.compile(r"^[a-z][a-z0-9_-]{0,31}$")
+# 可编辑字段（预置流程的 overrides 与自定义流程共用同一套规范化）。
+# icon 刻意不进来：预置图标跟版本走（老 overrides 里钉的旧 emoji 会被忽略），
+# 自定义流程的 icon 在 upsert_flow 里直接落盘，不走 overrides。
+_EDITABLE = ("name", "goal_hint", "note", "manuscript", "rubric",
+             "threshold", "rounds", "serial", "draft_prompt", "critique_prompt",
+             "verify_command", "best_of")
+
+
+def _read():
+    try:
+        data = json.loads(_FILE.read_text(encoding="utf-8"))
+        return data if isinstance(data, dict) else {}
+    except Exception:
+        return {}
+
+
+def _write(data):
+    _FILE.parent.mkdir(parents=True, exist_ok=True)
+    tmp = _FILE.with_suffix(".tmp")
+    tmp.write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
+    tmp.replace(_FILE)
+
+
+def _norm_serial(serial):
+    if not isinstance(serial, dict) or not serial.get("chapters"):
+        return None
+    try:
+        out = {"chapters": max(2, min(20, int(serial["chapters"]))),
+               "words_per_chapter": max(500, min(8000, int(serial.get("words_per_chapter") or 2500)))}
+    except Exception:
+        return None
+    # 分卷（可选）：显式卷表优先，其次每卷章数。流程模板里可以预置，
+    # 任务级 payload 仍可覆盖/关闭（0 = 不分卷）。
+    spec = volumes.norm_spec(serial.get("volumes"))
+    if spec:
+        out["volumes"] = spec
+    per = volumes.norm_per(serial.get("volume_chapters"))
+    if per:
+        out["volume_chapters"] = per
+    return out
+
+
+def _norm_rubric(rubric):
+    if isinstance(rubric, str):
+        rubric = re.split(r"[,，、\n]+", rubric)
+    if not isinstance(rubric, list):
+        return None
+    out = [str(d).strip() for d in rubric if str(d).strip()][:8]
+    return out or None
+
+
+def _apply_overrides(base, ov):
+    """把 overrides 合并到预置流程上（只认白名单字段）。"""
+    f = dict(base)
+    if not isinstance(ov, dict):
+        return f
+    for k in _EDITABLE:
+        if k not in ov:
+            continue
+        v = ov[k]
+        if k == "rubric":
+            v = _norm_rubric(v)
+            if v:
+                f["rubric"] = v
+        elif k == "serial":
+            v = _norm_serial(v)
+            if v:
+                f["serial"] = v
+            else:
+                f.pop("serial", None)
+        elif k == "threshold":
+            try:
+                f["threshold"] = max(1.0, min(10.0, float(v)))
+            except Exception:
+                pass
+        elif k == "rounds":
+            try:
+                f["rounds"] = max(1, min(5, int(v)))
+            except Exception:
+                pass
+        elif k == "best_of":
+            try:
+                f["best_of"] = max(1, min(3, int(v)))
+            except Exception:
+                pass
+        elif k in ("manuscript", "verify_command"):
+            s = re.sub(r"[\\/]+", "_", str(v or "")).strip()
+            s = re.sub(r"\.{2,}", "_", s).lstrip(".")
+            if s:
+                f[k] = s
+        elif k in ("draft_prompt", "critique_prompt"):
+            s = str(v or "").strip()
+            if s:
+                f[k] = s[:4000]
+            else:
+                f.pop(k, None)
+        else:
+            f[k] = str(v or "")[:80]
+    return f
+
+
+def _custom():
+    data = _read().get("flows")
+    return [f for f in data if isinstance(f, dict) and f.get("id")] if isinstance(data, list) else []
+
+
+def _overrides():
+    ov = _read().get("overrides")
+    return ov if isinstance(ov, dict) else {}
+
+
+def list_flows():
+    """自定义流程在前（便于看到新增），预置流程按定义顺序，均已套用 overrides。"""
+    with _LOCK:
+        custom, ov = _custom(), _overrides()
+    builtin = [_apply_overrides(b, ov.get(b["id"])) for b in BUILTIN_FLOWS]
+    for f in builtin:
+        f["edited"] = bool(ov.get(f["id"]))
+    return custom + builtin
+
+
+def get_flow(flow_id):
+    for f in list_flows():
+        if f["id"] == flow_id:
+            return f
+    return None
+
+
+# 参与 digest 的语义字段：会进运行链的参数与提示词。icon/note/goal_hint/name 是
+# 纯展示位——措辞变化不影响执行语义，不应制造「修订漂移」噪音。
+_DIGEST_FIELDS = ("id", "engine", "manuscript", "rubric", "threshold", "rounds",
+                  "best_of", "serial", "draft_prompt", "critique_prompt",
+                  "verify_command")
+
+
+def flow_content_sha256(flow):
+    """Return the full semantic content hash for an immutable flow revision."""
+    if not isinstance(flow, dict):
+        return ""
+    basis = {k: flow.get(k) for k in _DIGEST_FIELDS}
+    blob = json.dumps(basis, ensure_ascii=False, sort_keys=True, default=str)
+    return hashlib.sha256(blob.encode("utf-8")).hexdigest()
+
+
+def flow_digest(flow):
+    """流程定义的内容指纹（sha256 前 16 位）——「这个任务当时用的是哪一版流程」
+    的对账真源（借鉴 WorkDSH ADR-0010 发布后不可变修订）。"""
+    return flow_content_sha256(flow)[:16]
+
+
+def flow_drift(task):
+    """任务流程修订漂移检测：任务创建时钉的 digest vs 当前流程定义。
+
+    任务参数在创建时已固化（改流程不影响固化字段），但 task_compile 对历史
+    任务缺失字段（threshold/rubric/manuscript）仍回落当前流程定义——续跑/
+    重试因此可能静默吃到新流程。这里只诊断不裁定：返回
+    {"flow", "pinned", "current"} 或 None（无快照/流程已删/无漂移）。"""
+    snap = (task or {}).get("flow_snapshot") or {}
+    pinned = str(snap.get("digest") or "")
+    if not pinned:
+        return None
+    cur = get_flow(str((task or {}).get("type") or ""))
+    if not cur:
+        return None
+    current = flow_digest(cur)
+    if not current or current == pinned:
+        return None
+    return {"flow": cur.get("id") or "", "pinned": pinned, "current": current}
+
+
+def upsert_flow(payload):
+    """新增/更新流程。返回 (flow, 错误)。
+
+    - 预置流程 id：写入 overrides（可编辑、可恢复默认）；
+    - 新 id：创建自定义流程——**只需 name + engine**，其余字段留空走引擎默认。
+    """
+    fid = str(payload.get("id") or "").strip()
+    name = str(payload.get("name") or "").strip()
+    engine = str(payload.get("engine") or "review").strip()
+    if not _ID_RE.match(fid):
+        return None, "流程 ID 只能是小写字母开头的字母/数字/-/_（≤32 位）"
+    if engine not in ENGINES:
+        return None, "engine 必须是 code、review 或 direct"
+    builtin_ids = {b["id"] for b in BUILTIN_FLOWS}
+
+    if fid in builtin_ids:
+        base = next(b for b in BUILTIN_FLOWS if b["id"] == fid)
+        if name and name != base["name"]:
+            pass  # 允许改名
+        ov = dict(payload)
+        ov.pop("id", None)
+        ov["engine"] = base["engine"]   # 引擎不可改（避免语义错乱）
+        merged = _apply_overrides(base, ov)
+        with _LOCK:
+            data = _read()
+            ovs = data.get("overrides") if isinstance(data.get("overrides"), dict) else {}
+            ovs[fid] = {k: merged[k] for k in _EDITABLE if k in merged}
+            data["overrides"] = ovs
+            _write(data)
+        merged["edited"] = True
+        return merged, None
+
+    if not name:
+        return None, "名称不能为空"
+    flow = {"id": fid, "name": name[:20], "icon": str(payload.get("icon") or "✨")[:4],
+            "engine": engine, "builtin": False,
+            "goal_hint": str(payload.get("goal_hint") or "")[:80],
+            "note": str(payload.get("note") or "")[:80]}
+    if engine == "review":
+        dflt = ENGINE_DEFAULTS["review"]
+        ms = re.sub(r"[\\/]+", "_", str(payload.get("manuscript") or (fid + ".md"))).strip()
+        ms = re.sub(r"\.{2,}", "_", ms).lstrip(".") or (fid + ".md")
+        flow["manuscript"] = ms
+        flow["rubric"] = _norm_rubric(payload.get("rubric")) or list(dflt["rubric"])
+        try:
+            flow["threshold"] = max(1.0, min(10.0, float(payload.get("threshold") or dflt["threshold"])))
+        except Exception:
+            flow["threshold"] = dflt["threshold"]
+        try:
+            flow["rounds"] = max(1, min(5, int(payload.get("rounds") or dflt["rounds"])))
+        except Exception:
+            flow["rounds"] = dflt["rounds"]
+        try:
+            flow["best_of"] = max(1, min(3, int(payload.get("best_of") or dflt.get("best_of") or 1)))
+        except Exception:
+            flow["best_of"] = 1
+        serial = _norm_serial(payload.get("serial"))
+        if serial:
+            flow["serial"] = serial
+        for key in ("draft_prompt", "critique_prompt"):
+            v = str(payload.get(key) or "").strip()
+            if v:
+                flow[key] = v[:4000]
+    elif engine == "code":
+        flow["verify_command"] = str(payload.get("verify_command") or "")[:200]
+    # direct：无流程参数（目标+附件即全部输入）
+    with _LOCK:
+        data = _read()
+        flows = data.get("flows") if isinstance(data.get("flows"), list) else []
+        flows = [f for f in flows if isinstance(f, dict) and f.get("id") != fid]
+        flows.append(flow)
+        data["flows"] = flows
+        _write(data)
+    return flow, None
+
+
+def delete_flow(flow_id):
+    """删除自定义流程；预置流程只能「恢复默认」不能删除。"""
+    if flow_id in {b["id"] for b in BUILTIN_FLOWS}:
+        return "预置流程不可删除（可点「恢复默认」回滚改动）"
+    with _LOCK:
+        data = _read()
+        flows = data.get("flows") if isinstance(data.get("flows"), list) else []
+        kept = [f for f in flows if not (isinstance(f, dict) and f.get("id") == flow_id)]
+        if len(kept) == len(flows):
+            return "流程不存在"
+        data["flows"] = kept
+        _write(data)
+    return None
+
+
+def reset_flow(flow_id):
+    """预置流程恢复默认（清掉 overrides）。返回错误或 None。"""
+    if flow_id not in {b["id"] for b in BUILTIN_FLOWS}:
+        return "只有预置流程支持恢复默认"
+    with _LOCK:
+        data = _read()
+        ovs = data.get("overrides") if isinstance(data.get("overrides"), dict) else {}
+        if flow_id in ovs:
+            ovs.pop(flow_id, None)
+            data["overrides"] = ovs
+            _write(data)
+    return None