codebee 0.1.103 → 0.1.105

package/CHANGELOG.md

@@ -4,6 +4,18 @@ CodeBee 的用户可感知变更记录。发布新版时：最上面加一节，
 `<!-- relnotes:start -->…<!-- relnotes:end -->` 段（那段会被 `npm view` 的
 README 元数据带回，供老版本在「发现新版本」时展示新版更新内容）。
 
+## v0.1.105（2026-10-12）
+
+- **claude CLI 输出上限 16000→32000**：全局评审读 40 万字级书稿（约 28 万 token 输入）时按旧值反复续烧 23 分钟后报「response exceeded the 16000 output token maximum」三连挂（实案：单次 $3 零产出）；32000=claude CLI 官方默认档，`setdefault` 语义保留——用户显式配置不覆盖
+- **治理契约：外部人工评审请求通道**：新增 append-only 评审请求台账（requested→assigned→approved/changes_requested/dismissed/expired 幂等流转）与 `/api/reviews`、`/api/reviews/<id>/transition` 路由，外部智能体可按契约提交待人工评审的产物；配套 `/api/capabilities` 能力清单与 `/api/tasks/<id>/memory/preview` 任务记忆注入预览接口
+- **项目记忆注入预算封顶 20000**：注入预算配置值超过 20000 时按 20000 收敛，防止过大预算挤占提示词（含回归测试锁定）
+- **蜂巢 3D 蜜蜂翅膀精修**：翅膀更纤薄透明、减淡青雾、脉络几何强化，助手蜂手臂/键盘手部姿态与腕端高光对齐，视觉契约测试全程护航（约 40 笔打磨）
+- 英文界面补齐「会话续跑」能力标签词条
+
+## v0.1.104（2026-10-11）
+
+- **连载「一直重写不达标」根因防线（马甲指纹三道防线）**：「落笔入册」「整段连排」类跨章复用此前只能靠评审判死，而重写用的是同一作者模型、长出来的还是同族指纹——循环烧钱永不达标。现把防线前移到生成侧：①**章稿落盘确定性查重闸**：整段（≥40 字）/整句（≥12 字）/完整对白（≥6 字）逐字重复检测，阈值按真实事故章校准，白描短句复沓不误伤；带重复的章稿直接打回，同作者重试与换将的提示词自动追加消重指令（同提示重跑只会长出同族事故）；②**起草负样例注入**：近 4 章实际开笔/收束摘录随提示词下发（`__REUSE_GUARD__`），跨章句式同款从评审判据前移成起草约束；③**大纲轮换纪律**：开笔方式与收束形态批内硬性轮换。查重证据自动进评审面可见。生效范围：发版后的新批连载
+
 ## v0.1.103（2026-10-11）
 
 - **设置页「模型评测」子页顶栏标题修复**：标题表漏 `evalbench` 键，点入「模型评测」子页后顶栏回退显示「设置」；现正确显示「模型评测」（英文界面 Model Bench），并补静态断言回归防回流

package/LICENSE

@@ -1,21 +1,21 @@
-MIT License
-
-Copyright (c) 2026 scwuxinping
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
+MIT License
+
+Copyright (c) 2026 scwuxinping
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -25,10 +25,12 @@ Trae Agent、OpenClaw……），提供
 ---
 
 <!-- relnotes:start -->
-### 最新版更新内容（v0.1.103）
+### 最新版更新内容（v0.1.105）
 
-- 设置页「模型评测」子页顶栏标题修复：点入不再回退显示「设置」，英文界面正确显示 Model Bench
-- 文案巡检勘误与死项清理：扫榜四平台口径、缺陷复盘类型标注等全部对齐实际行为，零逻辑改动
+- 修复 40 万字级书稿全局评审被 claude 输出上限 16000 顶爆三连挂：上限提到 32000（用户显式配置不覆盖）
+- 新增治理契约：外部人工评审请求通道（append-only 台账+幂等流转）+ 能力清单/记忆注入预览接口
+- 蜂巢 3D 蜜蜂翅膀精修：更纤薄透明、脉络清晰、手部姿态自然
+- 项目记忆注入预算封顶 20000 防挤占
 <!-- relnotes:end -->
 
 

package/app/core/acceptance.py

@@ -1,100 +1,100 @@
-# -*- coding: utf-8 -*-
-"""Run acceptance matrix and evidence normalization.
-
-The matrix keeps correctness, performance, boundary behavior, exception
-handling, readability and security as separate conclusions.  Missing evidence
-is reported as ``not_evaluated``/``unknown`` rather than being inferred from a
-successful process exit.
-"""
-from __future__ import annotations
-
-import time
-
-CASES = ("correctness", "performance", "boundary", "exception",
-         "readability", "security")
-
-
-def _case(name, status, evidence=None, reason=""):
-    return {"case": name, "status": status,
-            "evidence": [evidence] if isinstance(evidence, dict) else [],
-            "reason": str(reason or "")[:300]}
-
-
-def _quality_verdict(run):
-    verdict = run.get("verdict") or {}
-    for key in ("pass", "publishable", "verify_pass"):
-        if isinstance(verdict.get(key), bool):
-            return verdict.get(key), key
-    return None, ""
-
-
-def evaluate_run(run, task=None):
-    """Build a deterministic acceptance result from persisted run evidence."""
-    run = run or {}
-    task = task or {}
-    status = str(run.get("status") or "")
-    verdict = run.get("verdict") or {}
-    out = []
-
-    correct, source = _quality_verdict(run)
-    if status in ("failed", "timeout", "cancelled"):
-        out.append(_case("correctness", "unknown",
-                         {"source": "run.status", "value": status},
-                         "任务终态未形成可判定的质量结论"))
-    elif correct is not None:
-        out.append(_case("correctness", "passed" if correct else "failed",
-                         {"source": "run.verdict", "field": source,
-                          "value": bool(correct)}))
-    else:
-        out.append(_case("correctness", "not_evaluated", reason="缺少验证或质量门证据"))
-
-    duration = run.get("duration_s")
-    if duration is None and run.get("started_at") and run.get("ended_at"):
-        duration = "recorded"
-    has_duration = isinstance(duration, (int, float)) and duration >= 0
-    has_usage = any(isinstance(run.get(key), (int, float)) and run.get(key) > 0
-                    for key in ("tokens", "cost_usd"))
-    if has_duration or has_usage:
-        out.append(_case("performance", "passed", {
-            "source": "run.metrics", "duration_s": duration,
-            "tokens": run.get("tokens", 0), "cost_usd": run.get("cost_usd", 0),
-        }, "仅表示性能数据已记录，不代表达到业务阈值"))
-    else:
-        out.append(_case("performance", "not_evaluated", reason="缺少耗时或用量证据"))
-
-    boundary = verdict.get("boundary_checks")
-    if isinstance(boundary, list) and boundary and all(
-            isinstance(x, dict) and isinstance(x.get("passed"), bool) for x in boundary):
-        failed = [x for x in boundary if x.get("passed") is False]
-        out.append(_case("boundary", "failed" if failed else "passed",
-                         {"source": "run.verdict.boundary_checks", "count": len(boundary),
-                          "failed": len(failed)}))
-    else:
-        out.append(_case("boundary", "not_evaluated", reason="当前任务未声明边界检查证据"))
-
-    exception = verdict.get("exception")
-    if isinstance(exception, bool):
-        out.append(_case("exception", "passed" if exception else "failed",
-                         {"source": "run.verdict.exception", "value": exception}))
-    elif status in ("failed", "timeout"):
-        out.append(_case("exception", "failed" if status in ("failed", "timeout") else "unknown",
-                         {"source": "run.error", "status": status,
-                          "error": str(run.get("error") or "")[:300]}))
-    elif status == "cancelled":
-        out.append(_case("exception", "unknown",
-                         {"source": "run.status", "value": status}))
-    else:
-        out.append(_case("exception", "not_evaluated", reason="缺少独立异常处理证据"))
-
-    for name, key in (("readability", "readability"), ("security", "security")):
-        value = verdict.get(key)
-        if isinstance(value, bool):
-            out.append(_case(name, "passed" if value else "failed",
-                             {"source": "run.verdict.%s" % key, "value": value}))
-        else:
-            out.append(_case(name, "not_evaluated", reason="未提供独立 %s 证据" % name))
-
-    # 总体通过要求六个维度都有明确通过证据；缺证据不能被两个局部绿灯掩盖。
-    return {"version": 1, "evaluated_at": time.time(),
-            "passed": bool(out and all(x["status"] == "passed" for x in out)),
-            "cases": out}
+# -*- coding: utf-8 -*-
+"""Run acceptance matrix and evidence normalization.
+
+The matrix keeps correctness, performance, boundary behavior, exception
+handling, readability and security as separate conclusions.  Missing evidence
+is reported as ``not_evaluated``/``unknown`` rather than being inferred from a
+successful process exit.
+"""
+from __future__ import annotations
+
+import time
+
+CASES = ("correctness", "performance", "boundary", "exception",
+         "readability", "security")
+
+
+def _case(name, status, evidence=None, reason=""):
+    return {"case": name, "status": status,
+            "evidence": [evidence] if isinstance(evidence, dict) else [],
+            "reason": str(reason or "")[:300]}
+
+
+def _quality_verdict(run):
+    verdict = run.get("verdict") or {}
+    for key in ("pass", "publishable", "verify_pass"):
+        if isinstance(verdict.get(key), bool):
+            return verdict.get(key), key
+    return None, ""
+
+
+def evaluate_run(run, task=None):
+    """Build a deterministic acceptance result from persisted run evidence."""
+    run = run or {}
+    task = task or {}
+    status = str(run.get("status") or "")
+    verdict = run.get("verdict") or {}
+    out = []
+
+    correct, source = _quality_verdict(run)
+    if status in ("failed", "timeout", "cancelled"):
+        out.append(_case("correctness", "unknown",
+                         {"source": "run.status", "value": status},
+                         "任务终态未形成可判定的质量结论"))
+    elif correct is not None:
+        out.append(_case("correctness", "passed" if correct else "failed",
+                         {"source": "run.verdict", "field": source,
+                          "value": bool(correct)}))
+    else:
+        out.append(_case("correctness", "not_evaluated", reason="缺少验证或质量门证据"))
+
+    duration = run.get("duration_s")
+    if duration is None and run.get("started_at") and run.get("ended_at"):
+        duration = "recorded"
+    has_duration = isinstance(duration, (int, float)) and duration >= 0
+    has_usage = any(isinstance(run.get(key), (int, float)) and run.get(key) > 0
+                    for key in ("tokens", "cost_usd"))
+    if has_duration or has_usage:
+        out.append(_case("performance", "passed", {
+            "source": "run.metrics", "duration_s": duration,
+            "tokens": run.get("tokens", 0), "cost_usd": run.get("cost_usd", 0),
+        }, "仅表示性能数据已记录，不代表达到业务阈值"))
+    else:
+        out.append(_case("performance", "not_evaluated", reason="缺少耗时或用量证据"))
+
+    boundary = verdict.get("boundary_checks")
+    if isinstance(boundary, list) and boundary and all(
+            isinstance(x, dict) and isinstance(x.get("passed"), bool) for x in boundary):
+        failed = [x for x in boundary if x.get("passed") is False]
+        out.append(_case("boundary", "failed" if failed else "passed",
+                         {"source": "run.verdict.boundary_checks", "count": len(boundary),
+                          "failed": len(failed)}))
+    else:
+        out.append(_case("boundary", "not_evaluated", reason="当前任务未声明边界检查证据"))
+
+    exception = verdict.get("exception")
+    if isinstance(exception, bool):
+        out.append(_case("exception", "passed" if exception else "failed",
+                         {"source": "run.verdict.exception", "value": exception}))
+    elif status in ("failed", "timeout"):
+        out.append(_case("exception", "failed" if status in ("failed", "timeout") else "unknown",
+                         {"source": "run.error", "status": status,
+                          "error": str(run.get("error") or "")[:300]}))
+    elif status == "cancelled":
+        out.append(_case("exception", "unknown",
+                         {"source": "run.status", "value": status}))
+    else:
+        out.append(_case("exception", "not_evaluated", reason="缺少独立异常处理证据"))
+
+    for name, key in (("readability", "readability"), ("security", "security")):
+        value = verdict.get(key)
+        if isinstance(value, bool):
+            out.append(_case(name, "passed" if value else "failed",
+                             {"source": "run.verdict.%s" % key, "value": value}))
+        else:
+            out.append(_case(name, "not_evaluated", reason="未提供独立 %s 证据" % name))
+
+    # 总体通过要求六个维度都有明确通过证据；缺证据不能被两个局部绿灯掩盖。
+    return {"version": 1, "evaluated_at": time.time(),
+            "passed": bool(out and all(x["status"] == "passed" for x in out)),
+            "cases": out}