codebee 0.1.101 → 0.1.103

package/app/core/defectretro.py

@@ -1,43 +1,43 @@
-# -*- coding: utf-8 -*-
-"""缺陷复盘报告（defect_retro，借鉴 zl2237/test-defect-retrospective）。
-
-禅道/Jira 缺陷导出 → 产品/开发/测试三视角复盘报告。
-数据不走网络：缺陷导出（CSV）作为任务附件注入上下文，这里只负责
-组装分析框架提示词。无附件时提示词会点名要求补附件——报告框架
-照常输出，不留静默空转。
-
-对照参考实现的取舍：它是本地脚本确定性算指标（analyzer 同输入同输出），
-我们是 direct 引擎一次性成稿——先跑通「有复盘产物」这一步，指标口径
-固化（状态/严重度归一在提示词里给锚）后续按需升档。
-"""
-from __future__ import annotations
-
-
-def retro_prompt(goal):
-    """组装缺陷复盘三视角提示词。纯框架无抓取，恒有返回值。"""
-    return (
-        "你是资深质量效能专家，负责输出一份缺陷复盘报告。\n\n"
-        "## 输入\n"
-        "用户会提供缺陷平台导出数据（推荐禅道/Jira 导出 CSV 作为任务附件；"
-        "Excel 二进制读不了，请导出 CSV）。字段口径以「标题、严重程度、"
-        "状态、创建日期、所属模块、解决方案、关闭原因、责任人」为锚，"
-        "缺失字段按实际有的算并在报告中声明口径。\n\n"
-        "## 用户关注点\n%s\n\n"
-        "## 你的产出（Markdown 报告）\n"
-        "1. **整体画像**：缺陷总量、严重度分布、状态分布、时间走势"
-        "（按创建日期聚合）、模块 Top5 集中度——只用表格+一句话结论，不复述原始数据\n"
-        "2. **产品视角**：需求类缺陷（描述不清/变更频繁/逻辑矛盾）占比与典型例；"
-        "类需求缺陷（被当 Bug 提的优化）点名\n"
-        "3. **开发视角**：根因聚类（编码疏忽/设计缺陷/环境/数据/回归）占比；"
-        "重复回归的模块点名；修复后又被激活的缺陷单列\n"
-        "4. **测试视角**：发现阶段分布（提测后/上线后/用户反馈）；漏测推测"
-        "（上线后才暴露的缺陷缺什么用例）；严重度与发现时长的错配\n"
-        "5. **改进动作**：每个视角 2-3 条，带建议责任角色与可验证的完成标准，"
-        "拒绝「加强沟通」类空话\n\n"
-        "## 纪律\n"
-        "- 所有数字必须从提供的导出数据里数出来，数不出的指标写「数据未提供」"
-        "而不是编造\n"
-        "- 若上下文里没有任何缺陷数据，直接说明需要附带导出 CSV 并给出导出步骤"
-        "（禅道：产品-缺陷-导出-CSV），不要硬写报告\n"
-        "- 直接输出报告，不要复述本提示词"
-        % (goal or "（用户未指定，按全量缺陷复盘）"))
+# -*- coding: utf-8 -*-
+"""缺陷复盘报告（defect_retro，借鉴 zl2237/test-defect-retrospective）。
+
+禅道/Jira 缺陷导出 → 产品/开发/测试三视角复盘报告。
+数据不走网络：缺陷导出（CSV）作为任务附件注入上下文，这里只负责
+组装分析框架提示词。无附件时提示词会点名要求补附件——报告框架
+照常输出，不留静默空转。
+
+对照参考实现的取舍：它是本地脚本确定性算指标（analyzer 同输入同输出），
+我们是 direct 引擎一次性成稿——先跑通「有复盘产物」这一步，指标口径
+固化（状态/严重度归一在提示词里给锚）后续按需升档。
+"""
+from __future__ import annotations
+
+
+def retro_prompt(goal):
+    """组装缺陷复盘三视角提示词。纯框架无抓取，恒有返回值。"""
+    return (
+        "你是资深质量效能专家，负责输出一份缺陷复盘报告。\n\n"
+        "## 输入\n"
+        "用户会提供缺陷平台导出数据（推荐禅道/Jira 导出 CSV 作为任务附件；"
+        "Excel 二进制读不了，请导出 CSV）。字段口径以「标题、严重程度、"
+        "状态、创建日期、所属模块、解决方案、关闭原因、责任人」为锚，"
+        "缺失字段按实际有的算并在报告中声明口径。\n\n"
+        "## 用户关注点\n%s\n\n"
+        "## 你的产出（Markdown 报告）\n"
+        "1. **整体画像**：缺陷总量、严重度分布、状态分布、时间走势"
+        "（按创建日期聚合）、模块 Top5 集中度——只用表格+一句话结论，不复述原始数据\n"
+        "2. **产品视角**：需求类缺陷（描述不清/变更频繁/逻辑矛盾）占比与典型例；"
+        "类需求缺陷（被当 Bug 提的优化）点名\n"
+        "3. **开发视角**：根因聚类（编码疏忽/设计缺陷/环境/数据/回归）占比；"
+        "重复回归的模块点名；修复后又被激活的缺陷单列\n"
+        "4. **测试视角**：发现阶段分布（提测后/上线后/用户反馈）；漏测推测"
+        "（上线后才暴露的缺陷缺什么用例）；严重度与发现时长的错配\n"
+        "5. **改进动作**：每个视角 2-3 条，带建议责任角色与可验证的完成标准，"
+        "拒绝「加强沟通」类空话\n\n"
+        "## 纪律\n"
+        "- 所有数字必须从提供的导出数据里数出来，数不出的指标写「数据未提供」"
+        "而不是编造\n"
+        "- 若上下文里没有任何缺陷数据，直接说明需要附带导出 CSV 并给出导出步骤"
+        "（禅道：产品-缺陷-导出-CSV），不要硬写报告\n"
+        "- 直接输出报告，不要复述本提示词"
+        % (goal or "（用户未指定，按全量缺陷复盘）"))

package/app/core/diagnostics.py

@@ -1,115 +1,115 @@
-# -*- coding: utf-8 -*-
-"""运行时不变量注册表：质量闸门从 pytest 测试期提升到运行时。
-
-设计稿：docs/migration/01-defense-patterns.md §5F。
-参考 dsh packages/runtime-diagnostics/invariants/src/index.ts：
-  enabled/package_allowlist/package_blocklist 三过滤；失败抛 InvariantError 含 packageName。
-
-CodeBee 实施简化为单文件 + pytest 钩子（设计稿说明）；register(name, fn) 注册断言，
-run_for(source, ctx) 在指定 source 下跑所有断言并返回失败列表。
-"""
-from __future__ import annotations
-
-import logging
-import threading
-from typing import Optional
-
-log = logging.getLogger(__name__)
-
-
-class InvariantError(Exception):
-    def __init__(self, source: str, check_name: str, msg: str):
-        super().__init__(f"[{source}/{check_name}] {msg}")
-        self.source = source
-        self.check_name = check_name
-
-
-class InvariantRegistry:
-    """按 source 分组的运行时断言注册表。
-
-    source 是来源标识（如 "pipeline"/"store"），name 是断言名；
-    fn(ctx) -> Optional[str] 返回 None 表示通过，返回字符串即失败消息。
-    """
-
-    def __init__(self):
-        self._lock = threading.Lock()
-        self._checks: dict = {}  # source -> {name: fn}
-
-    def register(self, source: str, name: str, fn):
-        with self._lock:
-            self._checks.setdefault(source, {})[name] = fn
-            log.debug("invariant registered: %s/%s", source, name)
-
-    def unregister(self, source: str, name: str = None):
-        """移除断言；name=None 时清空整个 source。"""
-        with self._lock:
-            if source not in self._checks:
-                return
-            if name is None:
-                del self._checks[source]
-            else:
-                self._checks[source].pop(name, None)
-
-    def run_for(self, source: str, ctx: dict, *, fail_fast: bool = False):
-        """在 source 下跑所有断言。
-
-        Returns:
-            [(source, name, message), ...] 失败列表（空 = 全过）。
-
-        Raises:
-            InvariantError: fail_fast=True 且首个失败。
-        """
-        fails = []
-        with self._lock:
-            checks = dict(self._checks.get(source, {}))
-        for name, fn in checks.items():
-            try:
-                msg = fn(ctx)
-            except Exception as e:
-                msg = f"check raised: {e!r}"
-            if msg:
-                fails.append((source, name, msg))
-                log.warning("invariant FAIL: %s/%s: %s", source, name, msg)
-                if fail_fast:
-                    raise InvariantError(source, name, msg)
-        return fails
-
-    def list(self, source: Optional[str] = None):
-        """列出已注册的 source/name；source=None 时全部。"""
-        with self._lock:
-            if source is None:
-                return [(s, n) for s, m in self._checks.items() for n in m]
-            return [(source, n) for n in self._checks.get(source, {})]
-
-
-invariants = InvariantRegistry()  # 单例
-
-
-# === 默认 invariants ===
-def register_default_checks():
-    """注册 CodeBee 默认运行时断言。"""
-
-    def review_no_all_fail_zero(ctx):
-        """评审全部 0 分但 run 标 success → 不允许（已锁 test_quality_gates:62-75）。"""
-        if (ctx.get("review_scores")
-                and all(s == 0 for s in ctx["review_scores"])
-                and ctx.get("final_status") == "success"):
-            return "评审全部 0 分但 run 标 success"
-        return None
-
-    def step_count_consistency(ctx):
-        """实际 step 数与预期不一致。"""
-        if ctx.get("actual_steps") != ctx.get("expected_steps"):
-            return f"step count {ctx.get('actual_steps')} != expected {ctx.get('expected_steps')}"
-        return None
-
-    def ok_but_error_code(ctx):
-        """step 级：ok=True 却带非空 error_code —— 解析降级（如 codex 空输出
-        回落 stdout 尾部）被当成功，值得告警复核。"""
-        if ctx.get("ok") and ctx.get("error_code"):
-            return "step 标记成功但携带 error_code=%s（疑似降级输出）" % ctx["error_code"]
-        return None
-
-    invariants.register("pipeline", "review_no_all_fail_zero", review_no_all_fail_zero)
-    invariants.register("pipeline", "step_count_consistency", step_count_consistency)
+# -*- coding: utf-8 -*-
+"""运行时不变量注册表：质量闸门从 pytest 测试期提升到运行时。
+
+设计稿：docs/migration/01-defense-patterns.md §5F。
+参考 dsh packages/runtime-diagnostics/invariants/src/index.ts：
+  enabled/package_allowlist/package_blocklist 三过滤；失败抛 InvariantError 含 packageName。
+
+CodeBee 实施简化为单文件 + pytest 钩子（设计稿说明）；register(name, fn) 注册断言，
+run_for(source, ctx) 在指定 source 下跑所有断言并返回失败列表。
+"""
+from __future__ import annotations
+
+import logging
+import threading
+from typing import Optional
+
+log = logging.getLogger(__name__)
+
+
+class InvariantError(Exception):
+    def __init__(self, source: str, check_name: str, msg: str):
+        super().__init__(f"[{source}/{check_name}] {msg}")
+        self.source = source
+        self.check_name = check_name
+
+
+class InvariantRegistry:
+    """按 source 分组的运行时断言注册表。
+
+    source 是来源标识（如 "pipeline"/"store"），name 是断言名；
+    fn(ctx) -> Optional[str] 返回 None 表示通过，返回字符串即失败消息。
+    """
+
+    def __init__(self):
+        self._lock = threading.Lock()
+        self._checks: dict = {}  # source -> {name: fn}
+
+    def register(self, source: str, name: str, fn):
+        with self._lock:
+            self._checks.setdefault(source, {})[name] = fn
+            log.debug("invariant registered: %s/%s", source, name)
+
+    def unregister(self, source: str, name: str = None):
+        """移除断言；name=None 时清空整个 source。"""
+        with self._lock:
+            if source not in self._checks:
+                return
+            if name is None:
+                del self._checks[source]
+            else:
+                self._checks[source].pop(name, None)
+
+    def run_for(self, source: str, ctx: dict, *, fail_fast: bool = False):
+        """在 source 下跑所有断言。
+
+        Returns:
+            [(source, name, message), ...] 失败列表（空 = 全过）。
+
+        Raises:
+            InvariantError: fail_fast=True 且首个失败。
+        """
+        fails = []
+        with self._lock:
+            checks = dict(self._checks.get(source, {}))
+        for name, fn in checks.items():
+            try:
+                msg = fn(ctx)
+            except Exception as e:
+                msg = f"check raised: {e!r}"
+            if msg:
+                fails.append((source, name, msg))
+                log.warning("invariant FAIL: %s/%s: %s", source, name, msg)
+                if fail_fast:
+                    raise InvariantError(source, name, msg)
+        return fails
+
+    def list(self, source: Optional[str] = None):
+        """列出已注册的 source/name；source=None 时全部。"""
+        with self._lock:
+            if source is None:
+                return [(s, n) for s, m in self._checks.items() for n in m]
+            return [(source, n) for n in self._checks.get(source, {})]
+
+
+invariants = InvariantRegistry()  # 单例
+
+
+# === 默认 invariants ===
+def register_default_checks():
+    """注册 CodeBee 默认运行时断言。"""
+
+    def review_no_all_fail_zero(ctx):
+        """评审全部 0 分但 run 标 success → 不允许（已锁 test_quality_gates:62-75）。"""
+        if (ctx.get("review_scores")
+                and all(s == 0 for s in ctx["review_scores"])
+                and ctx.get("final_status") == "success"):
+            return "评审全部 0 分但 run 标 success"
+        return None
+
+    def step_count_consistency(ctx):
+        """实际 step 数与预期不一致。"""
+        if ctx.get("actual_steps") != ctx.get("expected_steps"):
+            return f"step count {ctx.get('actual_steps')} != expected {ctx.get('expected_steps')}"
+        return None
+
+    def ok_but_error_code(ctx):
+        """step 级：ok=True 却带非空 error_code —— 解析降级（如 codex 空输出
+        回落 stdout 尾部）被当成功，值得告警复核。"""
+        if ctx.get("ok") and ctx.get("error_code"):
+            return "step 标记成功但携带 error_code=%s（疑似降级输出）" % ctx["error_code"]
+        return None
+
+    invariants.register("pipeline", "review_no_all_fail_zero", review_no_all_fail_zero)
+    invariants.register("pipeline", "step_count_consistency", step_count_consistency)
     invariants.register("step", "ok_but_error_code", ok_but_error_code)

package/app/core/dispatch.py

@@ -27,7 +27,7 @@ TYPE_DIMENSIONS = {
     "tech_proposal": "reasoning",
     "resume": "writing",
     "bid_doc": "writing",   # 标书：正文应答写作量大（合规判断由契约规则约束）
-    "defect_retro": "reasoning",   # 禅道工单：排查定责推理向
+    "defect_retro": "reasoning",   # 缺陷复盘：排查定责推理向
 }
 
 _KIND_AFFINITY = {

package/app/core/dispatch_log.py

@@ -1,128 +1,128 @@
-# -*- coding: utf-8 -*-
-"""调度事件台账：记录脱敏的候选、决策和结果，支持按运行回放。"""
-from __future__ import annotations
-
-import json
-import hashlib
-import threading
-import time
-
-from . import paths
-from .redact import scrub_text
-
-LOCK = threading.RLock()
-
-
-def _dispatch_dir():
-    # 读取动态 DATA_DIR，测试和多实例运行可在启动后重定向数据目录。
-    return paths.DATA_DIR / "dispatch"
-
-
-def _month_file(day):
-    return _dispatch_dir() / ("dispatch-%s.jsonl" % day[:7].replace("-", ""))
-
-
-def _text(value, limit=240):
-    value = str(value or "").replace("\r", " ").replace("\n", " ")
-    return scrub_text(value, limit=limit)
-
-
-def _candidate(row):
-    if not isinstance(row, dict):
-        return {}
-    return {"agent_id": _text(row.get("agent_id"), 64),
-            "label": _text(row.get("label"), 80),
-            "kind": _text(row.get("kind"), 32),
-            "score": row.get("score", 0),
-            "reason": _text(row.get("reason"), 240),
-            "order": row.get("order", 0)}
-
-
-def record_event(run_id="", task_id="", task_type="", difficulty="", role="",
-                 phase="selected", selected="", participants=(), candidates=(),
-                 fallback=(), selection_reason="", result="", verify_pass=None,
-                 review_pass=None, trace_id="", span_id="", parent_event_id=""):
-    """追加一条事件；仅保存路由元数据，不保存正文、提示词、密钥或文件内容。"""
-    try:
-        day = time.strftime("%Y-%m-%d")
-        event = {
-            "ts": time.strftime("%Y-%m-%d %H:%M:%S"),
-            "run_id": _text(run_id, 64), "task_id": _text(task_id, 64),
-            "task_type": _text(task_type, 32), "difficulty": _text(difficulty, 16),
-            "role": _text(role, 40), "phase": _text(phase, 16),
-            "selected": _text(selected, 64),
-            "participants": [_text(x.get("id") if isinstance(x, dict) else x, 64)
-                             for x in (participants or ())],
-            "candidates": [_candidate(x) for x in (candidates or ()) if isinstance(x, dict)],
-            "fallback": [_text(x, 64) for x in (fallback or ())],
-            "selection_reason": _text(selection_reason, 400),
-            "result": _text(result, 32),
-        }
-        # Trace identifiers are routing metadata only.  They deliberately do
-        # not carry prompt/output contents and remain optional for old callers.
-        if run_id:
-            from . import tracing
-            event["trace_id"] = _text(trace_id or tracing.trace_id(run_id), 64)
-        if span_id:
-            event["span_id"] = _text(span_id, 32)
-        if parent_event_id:
-            event["parent_event_id"] = _text(parent_event_id, 64)
-        event["event_id"] = hashlib.sha256(
-            ("%s|%s|%s|%s|%s|%s" % (
-                time.time_ns(), run_id, phase, role, selected, result
-            )).encode("utf-8")
-        ).hexdigest()[:24]
-        if verify_pass is not None:
-            event["verify_pass"] = bool(verify_pass)
-        if review_pass is not None:
-            event["review_pass"] = bool(review_pass)
-        with LOCK:
-            _dispatch_dir().mkdir(parents=True, exist_ok=True)
-            with open(_month_file(day), "a", encoding="utf-8") as fh:
-                fh.write(json.dumps(event, ensure_ascii=False) + "\n")
-        return event
-    except Exception:
-        return None
-
-
-def _iter_events(run_id="", task_type="", limit=100):
-    """从新到旧读取，满足 limit 即停，避免回放请求解析全部历史。"""
-    try:
-        directory = _dispatch_dir()
-        files = sorted(directory.glob("dispatch-*.jsonl"), reverse=True) \
-            if directory.is_dir() else []
-    except Exception:
-        return []
-    events = []
-    for path in files:
-        try:
-            lines = path.read_text(encoding="utf-8", errors="replace").splitlines()
-            for line in reversed(lines):
-                if not line.strip().startswith("{"):
-                    continue
-                try:
-                    item = json.loads(line)
-                except Exception:
-                    continue
-                if not isinstance(item, dict):
-                    continue
-                if run_id and item.get("run_id") != run_id:
-                    continue
-                if task_type and item.get("task_type") != task_type:
-                    continue
-                events.append(item)
-                if len(events) >= limit:
-                    return events
-        except Exception:
-            continue
-    return events
-
-
-def replay(run_id="", task_type="", limit=100):
-    """返回最新的脱敏调度事件，支持按 run_id/task_type 筛选。"""
-    try:
-        limit = max(1, min(1000, int(limit)))
-    except (TypeError, ValueError):
-        limit = 100
-    rid, ttype = _text(run_id, 64), _text(task_type, 32)
-    return _iter_events(rid, ttype, limit)
+# -*- coding: utf-8 -*-
+"""调度事件台账：记录脱敏的候选、决策和结果，支持按运行回放。"""
+from __future__ import annotations
+
+import json
+import hashlib
+import threading
+import time
+
+from . import paths
+from .redact import scrub_text
+
+LOCK = threading.RLock()
+
+
+def _dispatch_dir():
+    # 读取动态 DATA_DIR，测试和多实例运行可在启动后重定向数据目录。
+    return paths.DATA_DIR / "dispatch"
+
+
+def _month_file(day):
+    return _dispatch_dir() / ("dispatch-%s.jsonl" % day[:7].replace("-", ""))
+
+
+def _text(value, limit=240):
+    value = str(value or "").replace("\r", " ").replace("\n", " ")
+    return scrub_text(value, limit=limit)
+
+
+def _candidate(row):
+    if not isinstance(row, dict):
+        return {}
+    return {"agent_id": _text(row.get("agent_id"), 64),
+            "label": _text(row.get("label"), 80),
+            "kind": _text(row.get("kind"), 32),
+            "score": row.get("score", 0),
+            "reason": _text(row.get("reason"), 240),
+            "order": row.get("order", 0)}
+
+
+def record_event(run_id="", task_id="", task_type="", difficulty="", role="",
+                 phase="selected", selected="", participants=(), candidates=(),
+                 fallback=(), selection_reason="", result="", verify_pass=None,
+                 review_pass=None, trace_id="", span_id="", parent_event_id=""):
+    """追加一条事件；仅保存路由元数据，不保存正文、提示词、密钥或文件内容。"""
+    try:
+        day = time.strftime("%Y-%m-%d")
+        event = {
+            "ts": time.strftime("%Y-%m-%d %H:%M:%S"),
+            "run_id": _text(run_id, 64), "task_id": _text(task_id, 64),
+            "task_type": _text(task_type, 32), "difficulty": _text(difficulty, 16),
+            "role": _text(role, 40), "phase": _text(phase, 16),
+            "selected": _text(selected, 64),
+            "participants": [_text(x.get("id") if isinstance(x, dict) else x, 64)
+                             for x in (participants or ())],
+            "candidates": [_candidate(x) for x in (candidates or ()) if isinstance(x, dict)],
+            "fallback": [_text(x, 64) for x in (fallback or ())],
+            "selection_reason": _text(selection_reason, 400),
+            "result": _text(result, 32),
+        }
+        # Trace identifiers are routing metadata only.  They deliberately do
+        # not carry prompt/output contents and remain optional for old callers.
+        if run_id:
+            from . import tracing
+            event["trace_id"] = _text(trace_id or tracing.trace_id(run_id), 64)
+        if span_id:
+            event["span_id"] = _text(span_id, 32)
+        if parent_event_id:
+            event["parent_event_id"] = _text(parent_event_id, 64)
+        event["event_id"] = hashlib.sha256(
+            ("%s|%s|%s|%s|%s|%s" % (
+                time.time_ns(), run_id, phase, role, selected, result
+            )).encode("utf-8")
+        ).hexdigest()[:24]
+        if verify_pass is not None:
+            event["verify_pass"] = bool(verify_pass)
+        if review_pass is not None:
+            event["review_pass"] = bool(review_pass)
+        with LOCK:
+            _dispatch_dir().mkdir(parents=True, exist_ok=True)
+            with open(_month_file(day), "a", encoding="utf-8") as fh:
+                fh.write(json.dumps(event, ensure_ascii=False) + "\n")
+        return event
+    except Exception:
+        return None
+
+
+def _iter_events(run_id="", task_type="", limit=100):
+    """从新到旧读取，满足 limit 即停，避免回放请求解析全部历史。"""
+    try:
+        directory = _dispatch_dir()
+        files = sorted(directory.glob("dispatch-*.jsonl"), reverse=True) \
+            if directory.is_dir() else []
+    except Exception:
+        return []
+    events = []
+    for path in files:
+        try:
+            lines = path.read_text(encoding="utf-8", errors="replace").splitlines()
+            for line in reversed(lines):
+                if not line.strip().startswith("{"):
+                    continue
+                try:
+                    item = json.loads(line)
+                except Exception:
+                    continue
+                if not isinstance(item, dict):
+                    continue
+                if run_id and item.get("run_id") != run_id:
+                    continue
+                if task_type and item.get("task_type") != task_type:
+                    continue
+                events.append(item)
+                if len(events) >= limit:
+                    return events
+        except Exception:
+            continue
+    return events
+
+
+def replay(run_id="", task_type="", limit=100):
+    """返回最新的脱敏调度事件，支持按 run_id/task_type 筛选。"""
+    try:
+        limit = max(1, min(1000, int(limit)))
+    except (TypeError, ValueError):
+        limit = 100
+    rid, ttype = _text(run_id, 64), _text(task_type, 32)
+    return _iter_events(rid, ttype, limit)