codebee 0.1.69 → 0.1.71

package/app/core/notify.py

@@ -1,106 +1,201 @@
-# -*- coding: utf-8 -*-
-"""运行结果群推送（借鉴 agency-orchestrator 的 --notify）：任务跑完把结果摘要
-推到钉钉/飞书/企业微信群机器人，配合定时自动化就是「AI 团队每天定点交活」。
-
-webhook 一个地址全包——按域名自动适配三种机器人格式（同 agency-orchestrator
-思路）；也接受任意 https 地址（按钉钉 text 形状发，自建 n8n 等自选）。
-推送在后台线程、失败只记日志——通知永远不影响任务本身。
-"""
-from __future__ import annotations
-
-import logging
-
-log = logging.getLogger(__name__)
-
-
-def _payload_for(webhook, text):
-    """按 webhook 域名适配群机器人消息体。返回 None = 无法识别的地址。"""
-    from urllib.parse import urlparse
-    host = (urlparse(webhook).hostname or "").lower()
-    if "dingtalk" in host:
-        return {"msgtype": "text", "text": {"content": text}}
-    if "feishu" in host or "larksuite" in host:
-        return {"msg_type": "text", "content": {"text": text}}
-    if "weixin" in host or "wechat" in host or "work.weixin" in host:
-        return {"msgtype": "text", "text": {"content": text}}
-    if host:
-        return {"msgtype": "text", "text": {"content": text}}   # 未知域名按钉钉形状
-    return None
-
-
-def _webhook():
-    try:
-        from . import settings
-        return str(settings.load().get("notify_webhook") or "").strip()
-    except Exception:
-        return ""
-
-
-def _post(hook, text):
-    """curl 子进程 POST（Python 不经手响应体）；网络失败只记日志。"""
-    from . import runner
-    import json as _json
-    body = _json.dumps(_payload_for(hook, text), ensure_ascii=False)
-    r = runner.run_process(
-        argv=["curl", "-sS", "--max-time", "20",
-              "-H", "Content-Type: application/json",
-              "-d", body, hook],
-        timeout=30)
-    return bool(r["ok"])
-
-
-def push_text(text):
-    """推一条文本到群。配置了 webhook 才推；失败返回 False 不抛错。"""
-    hook = _webhook()
-    if not hook:
-        return False
-    if not hook.startswith("https://"):
-        log.warning("notify webhook 必须是 https")
-        return False
-    try:
-        return _post(hook, text)
-    except Exception as e:
-        log.warning("notify push failed: %s", e)
-        return False
-
-
-def push_run_async(run_id):
-    """任务收尾后异步推送结果摘要（jobs 层调用；绝不阻塞/影响任务）。"""
-    threading.Thread(target=push_run, daemon=True,
-                     name="notify-%s" % run_id, args=(run_id,)).start()
-
-
-def push_run(run_id):
-    """组装 run 结果摘要并推送（webhook 未配置时静默跳过）。
-
-    摘要末尾附分享页路径（notify_base_url 设置非空时拼完整链接，否则只给
-    run_id 供在本机 CodeBee 界面查找）。"""
-    hook = _webhook()
-    if not hook:
-        return False
-    from . import store
-    run = store.get_run(run_id) or {}
-    if not run:
-        return False
-    status = str(run.get("status") or "")
-    mark = {"done": "✅", "failed": "❌", "cancelled": "⚪"}.get(status, "🔔")
-    lines = ["%s CodeBee 任务%s" % (mark, {"done": "完成", "failed": "失败",
-                                           "cancelled": "已取消"}.get(status, status))]
-    lines.append("任务：%s" % (run.get("title") or run_id))
-    if run.get("error"):
-        lines.append("错误：%s" % str(run["error"])[:200])
-    v = run.get("verdict") or {}
-    if v.get("overall") is not None:
-        lines.append("综合评分 %.1f（%s）" % (
-            float(v["overall"]), "达标" if v.get("publishable") else "未达标"))
-    try:
-        from .settings import load as _sload
-        base = str(_sload().get("notify_base_url") or "").strip().rstrip("/")
-        if base:
-            lines.append("📄 详情：%s/api/runs/%s/share" % (base, run_id))
-    except Exception:
-        pass
-    return push_text("\n".join(lines))
-
-
-import threading  # noqa: E402  （push_run_async 依赖；置底避免顶部循环导入）
+# -*- coding: utf-8 -*-
+"""运行结果推送（借鉴 agency-orchestrator 的 --notify）：任务跑完把结果摘要
+推到钉钉/飞书/企业微信群机器人，配合定时自动化就是「AI 团队每天定点交活」。
+
+两类通道：
+  - 群机器人 webhook 一个地址全包——按域名自动适配三种机器人格式；也接受
+    任意 https 地址（按钉钉 text 形状发，自建 n8n 等自选）。
+  - 个人推送通道（Bark / ntfy / Server酱 / Telegram Bot）：推到私人手机，
+    无人值守场景「半夜失败/待裁决」的刚需；群机器人是给群里看的，这是给你看的。
+推送在后台线程、失败只记日志——通知永远不影响任务本身。
+"""
+from __future__ import annotations
+
+import logging
+
+log = logging.getLogger(__name__)
+
+
+def _payload_for(webhook, text):
+    """按 webhook 域名适配群机器人消息体。返回 None = 无法识别的地址。"""
+    from urllib.parse import urlparse
+    host = (urlparse(webhook).hostname or "").lower()
+    if "dingtalk" in host:
+        return {"msgtype": "text", "text": {"content": text}}
+    if "feishu" in host or "larksuite" in host:
+        return {"msg_type": "text", "content": {"text": text}}
+    if "weixin" in host or "wechat" in host or "work.weixin" in host:
+        return {"msgtype": "text", "text": {"content": text}}
+    if host:
+        return {"msgtype": "text", "text": {"content": text}}   # 未知域名按钉钉形状
+    return None
+
+
+def _webhook():
+    try:
+        from . import settings
+        return str(settings.load().get("notify_webhook") or "").strip()
+    except Exception:
+        return ""
+
+
+def _send(url, json_body=None, form=None):
+    """curl 子进程发送（Python 不经手响应体）；网络失败只记日志。返回 bool。
+
+    json_body=JSON 体；form=表单体（值需 UTF-8 百分号编码，urlencode 负责）。
+    中文一律走请求体（Server酱表单 / ntfy JSON publish），不走 HTTP 头——
+    头里的非 ASCII 会被网关拒收（历史坑）。
+    """
+    from . import runner
+    import json as _json
+    argv = ["curl", "-sS", "--max-time", "20"]
+    if json_body is not None:
+        argv += ["-H", "Content-Type: application/json",
+                 "-d", _json.dumps(json_body, ensure_ascii=False)]
+    elif form is not None:
+        from urllib.parse import urlencode
+        argv += ["-H", "Content-Type: application/x-www-form-urlencoded",
+                 "-d", urlencode(form)]
+    argv.append(url)
+    r = runner.run_process(argv=argv, timeout=30)
+    return bool(r["ok"])
+
+
+def _post(hook, text):
+    return _send(hook, json_body=_payload_for(hook, text))
+
+
+def _personal_channels():
+    """读设置里的个人推送配置，返回 [(通道名, 发送函数(title, body)), ...]。
+
+    只收录配置完整的通道；键名与 settings.DEFAULTS 对齐。
+    """
+    from . import settings
+    s = settings.load()
+    chs = []
+
+    bark_key = str(s.get("notify_bark_key") or "").strip()
+    if bark_key:
+        server = (str(s.get("notify_bark_server") or "").strip().rstrip("/")
+                  or "https://api.day.app")
+        chs.append(("bark", lambda t, b: _send(
+            server + "/push",
+            json_body={"device_key": bark_key, "title": t, "body": b})))
+
+    ntfy = str(s.get("notify_ntfy_topic") or "").strip()
+    if ntfy:
+        from urllib.parse import urlparse
+        u = urlparse(ntfy if "://" in ntfy else "https://ntfy.sh/" + ntfy.lstrip("/"))
+        topic = u.path.strip("/")
+        if topic:
+            server = "%s://%s" % (u.scheme, u.netloc)
+            chs.append(("ntfy", lambda t, b: _send(
+                server + "/",
+                json_body={"topic": topic, "title": t, "message": b})))
+
+    sc_key = str(s.get("notify_serverchan_key") or "").strip()
+    if sc_key:
+        chs.append(("serverchan", lambda t, b: _send(
+            "https://sctapi.ftqq.com/%s.send" % sc_key,
+            form={"title": t, "desp": b})))
+
+    tg_token = str(s.get("notify_telegram_token") or "").strip()
+    tg_chat = str(s.get("notify_telegram_chat_id") or "").strip()
+    if tg_token and tg_chat:
+        chs.append(("telegram", lambda t, b: _send(
+            "https://api.telegram.org/bot%s/sendMessage" % tg_token,
+            json_body={"chat_id": tg_chat, "text": (t + "\n" + b).strip()})))
+    return chs
+
+
+def _split_title(text):
+    """首行作标题（≤40 字），其余作正文；单行文本标题正文同文。"""
+    lines = [ln for ln in str(text).splitlines() if ln.strip()]
+    title = (lines[0].strip() if lines else "CodeBee")[:40]
+    body = "\n".join(ln.strip() for ln in lines[1:]).strip() or title
+    return title, body
+
+
+def push_text_ex(text):
+    """推一条文本到全部已配置通道，返回 {通道名: 是否成功}。
+
+    未配置任何通道返回 {}；单通道失败不影响其他通道。
+    """
+    results = {}
+    hook = _webhook()
+    if hook:
+        if not hook.startswith("https://"):
+            log.warning("notify webhook 必须是 https")
+            results["webhook"] = False
+        else:
+            try:
+                results["webhook"] = _post(hook, text)
+            except Exception as e:
+                log.warning("notify push failed: %s", e)
+                results["webhook"] = False
+    try:
+        title, body = _split_title(text)
+    except Exception:
+        title, body = "CodeBee", str(text)
+    for name, fn in _personal_channels():
+        try:
+            results[name] = bool(fn(title, body))
+        except Exception as e:
+            log.warning("notify %s push failed: %s", name, e)
+            results[name] = False
+    return results
+
+
+def push_text(text):
+    """推一条文本；任一通道成功即 True；全未配置/全失败返回 False，不抛错。"""
+    try:
+        return any(push_text_ex(text).values())
+    except Exception as e:
+        log.warning("notify push failed: %s", e)
+        return False
+
+
+def push_run_async(run_id):
+    """任务收尾后异步推送结果摘要（jobs 层调用；绝不阻塞/影响任务）。"""
+    threading.Thread(target=push_run, daemon=True,
+                     name="notify-%s" % run_id, args=(run_id,)).start()
+
+
+def push_run(run_id):
+    """组装 run 结果摘要并推送（群 webhook 或任一个人通道配置了才推）。
+
+    摘要末尾附分享页路径（notify_base_url 设置非空时拼完整链接，否则只给
+    run_id 供在本机 CodeBee 界面查找）。"""
+    from . import store
+    run = store.get_run(run_id) or {}
+    if not run:
+        return False
+    status = str(run.get("status") or "")
+    mark = {"done": "✅", "failed": "❌", "cancelled": "⚪"}.get(status, "🔔")
+    lines = ["%s CodeBee 任务%s" % (mark, {"done": "完成", "failed": "失败",
+                                           "cancelled": "已取消"}.get(status, status))]
+    lines.append("任务：%s" % (run.get("title") or run_id))
+    if run.get("error"):
+        lines.append("错误：%s" % str(run["error"])[:200])
+    v = run.get("verdict") or {}
+    if v.get("overall") is not None:
+        lines.append("综合评分 %.1f（%s）" % (
+            float(v["overall"]), "达标" if v.get("publishable") else "未达标"))
+    try:
+        from .settings import load as _sload
+        base = str(_sload().get("notify_base_url") or "").strip().rstrip("/")
+        if base:
+            lines.append("📄 详情：%s/api/runs/%s/share" % (base, run_id))
+    except Exception:
+        pass
+    # 分支待裁决：任务跑完有改动等人拍板时点名提醒（手机端看到即可去详情处理）
+    try:
+        task = store.get_task(run.get("task_id")) if run.get("task_id") else None
+        if task and task.get("git_state") == "isolated":
+            lines.append("⚠ 任务分支待裁决（合并/丢弃）")
+    except Exception:
+        pass
+    return push_text("\n".join(lines))
+
+
+import threading  # noqa: E402  （push_run_async 依赖；置底避免顶部循环导入）

package/app/core/novel_quality.py

@@ -0,0 +1,189 @@
+# -*- coding: utf-8 -*-
+"""Deterministic quality checks for signing-oriented Chinese web fiction.
+
+The LLM remains responsible for literary judgment.  This module supplies
+stable rubric defaults and small text signals so reviewers can cite evidence
+for the common rejection reasons instead of relying on vague impressions.
+"""
+from __future__ import annotations
+
+import re
+
+
+SIGNING_RUBRIC = [
+    "开篇吸引力",
+    "情节推进",
+    "文风统一",
+    "情感细腻度",
+    "节奏控制",
+]
+
+OPENING_DIM = "开篇吸引力"
+SIGNING_CHECKPOINT_WORDS = 20000   # 平台口径：前 2 万字用于签约评估
+
+_SENTENCE_RE = re.compile(r"[^。！？!?；;\n]+[。！？!?；;]?")
+_DIALOGUE_RE = re.compile(r"[“‘\"「『](.*?)[”’\"」』]", re.S)
+_EMOTION_RE = re.compile(r"哭|泪|笑|怒|怕|颤|抖|咬|攥|握紧|心跳|呼吸|喉咙|指甲|沉默")
+
+
+def rubric_for(task):
+    """Return the configured rubric, or the signing rubric for serial novels."""
+    task = task if isinstance(task, dict) else {}
+    raw = task.get("rubric")
+    if isinstance(raw, str):
+        raw = [x.strip() for x in raw.replace("，", ",").split(",") if x.strip()]
+    if isinstance(raw, (list, tuple)):
+        values = [str(x).strip() for x in raw if str(x).strip()]
+        if values:
+            return values[:8]
+    if str(task.get("type") or "").lower() == "serial_novel":
+        return list(SIGNING_RUBRIC)
+    return []
+
+
+def opening_requirements(chapter, target_words=None):
+    """Return concise, chapter-aware acceptance requirements for prompts."""
+    try:
+        chapter = int(chapter)
+    except (TypeError, ValueError):
+        chapter = 1
+    if chapter == 1:
+        try:
+            short = int(target_words or 0) < 1200
+        except (TypeError, ValueError):
+            short = False
+        core = ("前 1200 字" if short else "前 2000 字")
+        return ("黄金一章门禁：前 100 字让读者知道主角是谁；前 200 字出现异常、欲望或冲突；"
+                "前 600 字把困境具体化；%s 展示核心卖点；结尾必须留下可兑现的钩子。" % core
+                + "禁止用世界观、环境或人物关系长铺垫开场。")
+    return "本章门禁：开头尽快承接上一章并制造新变化；结尾必须留下下一步行动或可兑现的章末钩子。"
+
+
+def signal_summary(text, chapter=None):
+    """Extract explainable text signals; no claim is made that they are scores."""
+    text = str(text or "").replace("\r\n", "\n").strip()
+    paragraphs = [p.strip() for p in re.split(r"\n\s*\n", text) if p.strip()]
+    sentences = [s.strip() for s in _SENTENCE_RE.findall(text) if s.strip()]
+    sentence_lengths = [len(re.sub(r"\s+", "", s)) for s in sentences]
+    dialogue_chars = sum(len(x) for x in _DIALOGUE_RE.findall(text))
+    emotion_hits = len(_EMOTION_RE.findall(text))
+    chars = len(re.sub(r"\s+", "", text))
+    avg = round(sum(sentence_lengths) / len(sentence_lengths), 1) if sentence_lengths else 0.0
+    long_sentences = sum(1 for n in sentence_lengths if n >= 55)
+    hints = []
+    if chapter == 1 and chars and chars < 600:
+        hints.append("首章正文不足 600 字，无法证明前 600 字已进入具体困境")
+    if avg >= 35 or long_sentences:
+        hints.append("存在偏长句，复查长定语、重复说明与动作拆分")
+    if len(paragraphs) >= 4 and chars / len(paragraphs) > 220:
+        hints.append("段落偏长，复查手机阅读下的节奏与信息密度")
+    if text and dialogue_chars == 0:
+        hints.append("缺少对白证据，复查是否用场景/行动推进冲突")
+    if text and emotion_hits == 0:
+        hints.append("缺少可见的情绪动作或生理反应，复查复杂情感是否写得过于概括")
+    return {
+        "chapter": chapter,
+        "chars": chars,
+        "paragraphs": len(paragraphs),
+        "sentences": len(sentences),
+        "avg_sentence_chars": avg,
+        "long_sentences": long_sentences,
+        "dialogue_chars": dialogue_chars,
+        "emotion_action_hits": emotion_hits,
+        "hints": hints[:6],
+    }
+
+
+def format_signal_summary(summary):
+    """Render signal_summary output as a compact prompt block."""
+    if not isinstance(summary, dict):
+        return ""
+    parts = [
+        "文本信号（仅作证据，不替代文学判断）：约 %s 字；%s 段；%s 句；平均句长 %s 字；长句 %s；对白 %s 字；情绪动作线索 %s 处。"
+        % (summary.get("chars", 0), summary.get("paragraphs", 0),
+           summary.get("sentences", 0), summary.get("avg_sentence_chars", 0),
+           summary.get("long_sentences", 0), summary.get("dialogue_chars", 0),
+           summary.get("emotion_action_hits", 0))]
+    hints = summary.get("hints") or []
+    if hints:
+        parts.append("确定性复查提示：" + "；".join(str(x) for x in hints))
+    return "\n".join(parts)
+
+
+def style_anchor_block(anchor):
+    """Render the book-level style anchor as a prompt block; empty when absent."""
+    anchor = str(anchor or "").strip()
+    if not anchor:
+        return ""
+    return ("## 全书文风锚（大纲裁定，所有章节起草/修订/评审必须遵守；"
+            "与其冲突处以本锚为准）\n%s\n" % anchor[:400])
+
+
+def fingerprint(summary):
+    """Normalize a signal_summary into per-1000-char style metrics."""
+    if not isinstance(summary, dict):
+        return {}
+    chars = max(1, int(summary.get("chars") or 0))
+    return {
+        "avg_sentence_chars": float(summary.get("avg_sentence_chars") or 0.0),
+        "dialogue_per_k": round(1000.0 * float(summary.get("dialogue_chars") or 0) / chars, 1),
+        "emotion_per_k": round(1000.0 * float(summary.get("emotion_action_hits") or 0) / chars, 1),
+    }
+
+
+def _median(values):
+    xs = sorted(float(v) for v in values if isinstance(v, (int, float)))
+    if not xs:
+        return 0.0
+    mid = len(xs) // 2
+    return xs[mid] if len(xs) % 2 else round((xs[mid - 1] + xs[mid]) / 2.0, 1)
+
+
+def style_deviation_notes(baselines, summary, min_baseline=2):
+    """Cross-chapter style drift warnings from deterministic fingerprints.
+
+    baselines: signal_summary dicts of the book's previous chapters.  Needs at
+    least ``min_baseline`` chapters before speaking (开篇几章没有稳定基线)；
+    all notes are evidence hints for reviewers/revisers, never scores.
+    """
+    notes = []
+    base = [fingerprint(b) for b in (baselines or []) if isinstance(b, dict)]
+    base = [b for b in base if b]
+    if len(base) < min_baseline:
+        return notes
+    cur = fingerprint(summary)
+    if not cur:
+        return notes
+    avg_base = _median([b["avg_sentence_chars"] for b in base])
+    if avg_base >= 12 and cur["avg_sentence_chars"] > avg_base * 1.45:
+        notes.append("本章平均句长 %s 字，此前各章中位 %s 字——句式明显变长，"
+                     "复查长定语、说明性插叙与节奏拖沓"
+                     % (cur["avg_sentence_chars"], avg_base))
+    elif avg_base >= 12 and cur["avg_sentence_chars"] < avg_base * 0.62:
+        notes.append("本章平均句长 %s 字，此前各章中位 %s 字——句式明显变碎，"
+                     "复查是否为凑字数拆散动作与因果"
+                     % (cur["avg_sentence_chars"], avg_base))
+    dial_base = _median([b["dialogue_per_k"] for b in base])
+    if dial_base >= 40 and cur["dialogue_per_k"] < dial_base * 0.45:
+        notes.append("对白占比骤降（本章 %s/千字，此前中位 %s/千字）——"
+                     "复查是否换成了大段叙述，文风与推进方式是否漂移"
+                     % (cur["dialogue_per_k"], dial_base))
+    emo_base = _median([b["emotion_per_k"] for b in base])
+    if emo_base >= 1.5 and cur["emotion_per_k"] < emo_base * 0.4:
+        notes.append("情绪动作密度骤降（本章 %s/千字，此前中位 %s/千字）——"
+                     "复查复杂情感是否退回「很伤心/很愤怒」式概括"
+                     % (cur["emotion_per_k"], emo_base))
+    return notes
+
+
+def opening_below(means, threshold):
+    """True when the signing opening dimension is scored below the line."""
+    try:
+        value = float((means or {}).get(OPENING_DIM))
+    except (TypeError, ValueError):
+        return False
+    try:
+        line = float(threshold)
+    except (TypeError, ValueError):
+        return False
+    return value < line