codebee 0.1.66 → 0.1.68

package/app/core/store.py

@@ -18,7 +18,7 @@ import threading
 import time
 from pathlib import Path
 
-from . import artifacts, paths, runner, volumes
+from . import artifacts, paths, runner, tracing, volumes
 
 LOCK = threading.RLock()
 _TASKS = {}
@@ -30,9 +30,12 @@ _STREAM_LOCK = threading.Lock()
 _STREAM_TS = {}
 LIVE_TEXT_MAX = 20000   # 单步骤实时思考/正文的保留上限（与 builtin_agent 同口径）
 
-DEFAULT_TASK_TIMEOUT_S = 3600
-MIN_TASK_TIMEOUT_S = 1
-MAX_TASK_TIMEOUT_S = 7 * 24 * 3600
+DEFAULT_TASK_TIMEOUT_S = 3600
+MIN_TASK_TIMEOUT_S = 1
+MAX_TASK_TIMEOUT_S = 7 * 24 * 3600
+REVIEW_TASK_TIMEOUT_S = 7200
+LONG_TASK_TIMEOUT_S = 14400
+SERIAL_EXTRA_CHAPTER_S = 1800
 
 # ---------------------------------------------------------------- 状态版本（SSE 事件驱动）
 # 任何落盘写都算状态变化：SSE 连接等版本号变化才构建/推送全量状态，
@@ -83,7 +86,7 @@ def _safe_name(s):
     return re.sub(r"[^0-9A-Za-z_-]+", "-", str(s))[:40].strip("-") or "x"
 
 
-def _task_timeout_s(value=None):
+def _task_timeout_s(value=None):
     """Resolve one bounded wall-clock budget for a task."""
     raw = value
     if raw in (None, ""):
@@ -92,7 +95,27 @@ def _task_timeout_s(value=None):
         seconds = int(float(raw))
     except (TypeError, ValueError):
         seconds = DEFAULT_TASK_TIMEOUT_S
-    return max(MIN_TASK_TIMEOUT_S, min(MAX_TASK_TIMEOUT_S, seconds))
+    return max(MIN_TASK_TIMEOUT_S, min(MAX_TASK_TIMEOUT_S, seconds))
+
+
+def _default_timeout_s(task):
+    """Choose a bounded default from the amount of work, not one hour for all."""
+    env_value = os.environ.get("TUTTI_TASK_TIMEOUT_S")
+    if env_value not in (None, ""):
+        return _task_timeout_s(env_value)
+    serial = task.get("serial")
+    if isinstance(serial, dict) and serial.get("chapters"):
+        try:
+            chapters = max(1, min(20, int(serial["chapters"])))
+        except (TypeError, ValueError):
+            chapters = 8
+        return min(MAX_TASK_TIMEOUT_S,
+                   LONG_TASK_TIMEOUT_S + max(0, chapters - 8) * SERIAL_EXTRA_CHAPTER_S)
+    if task.get("engine") == "code":
+        return LONG_TASK_TIMEOUT_S
+    if task.get("engine") == "review":
+        return REVIEW_TASK_TIMEOUT_S
+    return DEFAULT_TASK_TIMEOUT_S
 
 
 # ---------------------------------------------------------------- 任务
@@ -159,7 +182,19 @@ def create_task(payload):
     }
     thinking = str(payload.get("thinking") or "standard").strip().lower()
     task["thinking"] = thinking if thinking in ("auto", "low", "standard", "high") else "auto"
-    task["timeout_s"] = _task_timeout_s(payload.get("timeout_s"))
+    # 流程修订出处（借鉴 WorkDSH ADR-0010）：创建时钉住当时流程定义的指纹。
+    # 固化参数已保证「改流程不影响本任务」，digest 再补上可对账的一面——
+    # 续跑/重试时若流程已改，flows.flow_drift 能给出明确诊断而非静默换新。
+    flow_content_sha = flows_mod.flow_content_sha256(flow)
+    task["flow_snapshot"] = {"id": flow["id"], "name": (flow.get("name") or "")[:40],
+                             "engine": flow["engine"],
+                             "edited": bool(flow.get("edited")),
+                             "digest": flow_content_sha[:16],
+                             "revision_id": "flowrev-" + flow_content_sha[:16],
+                             "content_sha256": flow_content_sha}
+    # 参数修订号（借鉴 WorkDSH 契约纪律 expectedRevision）：写接口据此做过期
+    # 写入检测。老任务无此键按 1 处理；不带 expected_rev 的调用不受影响。
+    task["rev"] = 1
     if flow["engine"] == "direct":
         task["direct_provider_id"] = _text(payload.get("direct_provider_id"),
                                             "direct_provider_id")[:80]
@@ -221,11 +256,11 @@ def create_task(payload):
         if serial_value is None or (isinstance(serial_value, dict) and
                                     serial_value.get("enabled") is False):
             serial = None
-        elif isinstance(serial_value, dict):
-            serial = serial_value
-        else:
-            serial = flow.get("serial")
-        if isinstance(serial, dict) and serial.get("chapters"):
+        elif isinstance(serial_value, dict):
+            serial = serial_value
+        else:
+            serial = flow.get("serial")
+        if isinstance(serial, dict) and serial.get("chapters"):
             try:
                 s = {
                     # 续写批次允许只续 1 章，下限放宽到 1（全新连载仍由前端约束 ≥2）
@@ -277,10 +312,13 @@ def create_task(payload):
                 if vol_spec:
                     s["volumes"] = vol_spec
                 vper = volumes.norm_per(serial.get("volume_chapters"))
-                if vper:
-                    s["volume_chapters"] = vper
-                task["serial"] = s
-    critics = payload.get("critics")
+                if vper:
+                    s["volume_chapters"] = vper
+                task["serial"] = s
+    raw_timeout = payload.get("timeout_s")
+    task["timeout_s"] = (_task_timeout_s(raw_timeout)
+                         if raw_timeout not in (None, "") else _default_timeout_s(task))
+    critics = payload.get("critics")
     if isinstance(critics, list) and critics:
         task["critics"] = [str(c) for c in critics]
     # 初始故事圣经必须在任务入队前落盘，保证首个章节步骤就能读到设定。
@@ -312,8 +350,20 @@ def create_task(payload):
     att_ids = payload.get("attachments")
     if isinstance(att_ids, list) and att_ids:
         from . import attachments as att_mod
-        items = [a for a in att_ids if isinstance(a, dict) and a.get("path")]
-        if not items:  # 纯 id 形态 → 从待提交区移入工作目录
+        dict_items = [a for a in att_ids if isinstance(a, dict) and a.get("path")]
+        if dict_items:
+            root = Path(str(wd)).resolve()
+            items = []
+            for raw in dict_items[:att_mod.MAX_FILES]:
+                rel = att_mod.norm_rel(raw)
+                fp = (root / rel).resolve() if rel else root / "__invalid__"
+                if rel and root in fp.parents and fp.is_file():
+                    item = dict(raw)
+                    item["path"] = rel
+                    items.append(item)
+            if not items:
+                raise ValueError("附件路径必须位于工作目录的 _attachments/ 内")
+        else:  # 纯 id 形态 → 从待提交区移入工作目录
             try:
                 items = att_mod.commit_to_workdir(
                     str(wd), [str(x) for x in att_ids][:att_mod.MAX_FILES])
@@ -321,6 +371,18 @@ def create_task(payload):
                 raise ValueError("附件落盘失败: %s" % e)
         if items:
             task["attachments"] = items
+            task["asset_refs"] = att_mod.asset_refs(str(wd), items)
+            try:
+                from . import assets
+                for ref in task["asset_refs"]:
+                    saved = assets.register(str(wd), ref.get("path"),
+                                            owner_type="task", owner_id=task["id"],
+                                            name=ref.get("name"), mime=ref.get("mime"))
+                    if saved:
+                        ref.update({k: saved[k] for k in ("asset_id", "revision_id",
+                                                          "content_sha256")})
+            except Exception:
+                pass
             task["context"] = att_mod.merge_context(
                 task["context"], items, workdir=str(wd))
     with LOCK:
@@ -569,12 +631,106 @@ def set_run_auditor(auditor):
     _RUN_AUDITOR = auditor if callable(auditor) else None
 
 
+def _execution_snapshot(task):
+    """Capture the redacted execution inputs at run creation time.
+
+    This is intentionally a snapshot of identities and revisions, never API
+    keys or task bodies.  Later binding/skill/knowledge changes therefore do
+    not rewrite the explanation of an existing run.
+    """
+    task = task or {}
+    flow_snapshot = dict(task.get("flow_snapshot") or {})
+    flow_digest = str(flow_snapshot.get("digest") or "")
+    flow_content_sha = str(flow_snapshot.get("content_sha256") or "")
+    flow_revision_id = str(flow_snapshot.get("revision_id") or "")
+    if not flow_content_sha and len(flow_digest) == 64:
+        flow_content_sha = flow_digest
+    if not flow_revision_id and flow_digest:
+        flow_revision_id = "flowrev-" + flow_digest[:16]
+
+    def _redact_command(command):
+        value = str(command or "")[:300]
+        if not value:
+            return ""
+        pattern = re.compile(
+            r"(?i)(api[_-]?key|access[_-]?token|auth(?:orization)?|password|passwd|secret|token)"
+            r"(\s*[:=]\s*|\s+)([^\s]+)")
+        return pattern.sub(lambda m: m.group(1) + m.group(2) + "<redacted>", value)
+
+    snapshot = {
+        "captured_at": time.strftime("%Y-%m-%d %H:%M:%S"),
+        "task_id": task.get("id") or "",
+        "flow_revision": flow_digest,
+        "flow_revision_id": flow_revision_id,
+        "flow_content_sha256": flow_content_sha,
+        "flow_snapshot": flow_snapshot,
+        "workdir": str(task.get("workdir") or "")[:500],
+        "git_rev": str(task.get("git_rev") or "")[:160],
+        "verify_command": _redact_command(task.get("verify_command") or ""),
+        "quality_gate": {
+            "threshold": task.get("threshold"),
+            "rubric": list(task.get("rubric") or [])[:8],
+            "rounds": task.get("rounds"),
+            "best_of": task.get("best_of"),
+        },
+    }
+    try:
+        from . import knowledge, modelhub, skills
+        snapshot["skills"] = skills.revision_for_task(task)
+        snapshot["knowledge"] = knowledge.revision_for_task(task)
+        providers = {str(p.get("id") or ""): p for p in modelhub.providers()}
+        bindings = {}
+        for agent_id, binding in (modelhub.bindings() or {}).items():
+            chain = []
+            for item in (binding or {}).get("chain") or []:
+                pid = str(item.get("provider_id") or "")
+                prov = providers.get(pid) or {}
+                try:
+                    key_ids = [str(k.get("id") or "") for k in modelhub._provider_keys(prov)
+                               if k.get("id")]
+                except Exception:
+                    key_ids = []
+                entry = {
+                    "provider_id": pid,
+                    "model": str(item.get("model") or "")[:160],
+                    "key_ids": key_ids[:8],
+                }
+                if pid:
+                    entry.update(modelhub.connection_identity(pid, key_ids[0] if key_ids else ""))
+                chain.append(entry)
+            bindings[str(agent_id)[:80]] = {
+                "difficulty_routing": bool((binding or {}).get("difficulty_routing")),
+                "chain": chain,
+            }
+        snapshot["agent_bindings"] = bindings
+        if task.get("direct_provider_id") or task.get("direct_model"):
+            direct = {
+                "provider_id": str(task.get("direct_provider_id") or "")[:80],
+                "model": str(task.get("direct_model") or "")[:160],
+                "agent": str(task.get("direct_agent") or "")[:80],
+            }
+            direct_pid = str(task.get("direct_provider_id") or "")
+            direct_prov = providers.get(direct_pid) or {}
+            direct_keys = modelhub._provider_keys(direct_prov)
+            if direct_pid:
+                direct.update(modelhub.connection_identity(
+                    direct_pid, str((direct_keys[0] if direct_keys else {}).get("id") or "")))
+            snapshot["direct"] = direct
+    except Exception:
+        # A snapshot is diagnostic metadata.  A broken optional catalog must
+        # never prevent the task from being created.
+        snapshot.setdefault("skills", {})
+        snapshot.setdefault("knowledge", {})
+        snapshot.setdefault("agent_bindings", {})
+    return snapshot
+
+
 # run 终态集合：与 CAS 收口、状态回填共用同一份定义
 TERMINAL_STATUSES = ("done", "failed", "cancelled", "timeout")
 
 
-def create_run(kind, title, task_id=None, entry_id=None, op=None):
-    run = {
+def create_run(kind, title, task_id=None, entry_id=None, op=None):
+    run = {
         "id": _new_id("r" if kind == "orchestration" else "m"),
         "kind": kind,  # orchestration | mgmt
         "title": title,
@@ -584,11 +740,20 @@ def create_run(kind, title, task_id=None, entry_id=None, op=None):
         "started_at": None, "ended_at": None,
         "deadline_at": None,
         "cost_usd": 0.0, "tokens": 0, "error": "",
-        "verdict": None, "summary": "",
-    }
-    if kind == "orchestration" and task_id:
-        task = get_task(task_id) or {}
-        run["deadline_at"] = time.time() + _task_timeout_s(task.get("timeout_s"))
+        "verdict": None, "summary": "",
+    }
+    # Persist trace identity at run creation so replay remains stable across
+    # process restarts and legacy readers can still use the normal run shape.
+    run["trace_schema_version"] = tracing.SCHEMA_VERSION
+    run["trace_id"] = tracing.trace_id(run["id"])
+    run["root_span_id"] = tracing.root_span_id(run["id"])
+    if kind == "orchestration" and task_id:
+        task = get_task(task_id) or {}
+        run["timeout_budget_s"] = _task_timeout_s(task.get("timeout_s"))
+        # Provisional for direct callers; jobs resets it at worker claim so
+        # queued/backoff time does not shorten the actual execution budget.
+        run["deadline_at"] = time.time() + run["timeout_budget_s"]
+        run["execution_snapshot"] = _execution_snapshot(task)
         try:
             eta = _RUN_ESTIMATOR(
                 task_type=task.get("type") or "", mode=task.get("mode") or "auto",
@@ -816,7 +981,6 @@ def update_run(run_id, expected_status=None, **fields):
                         s["summary"] = ("步骤未正常收尾（任务总时限已到）"
                                          if st == "timeout" else
                                          "步骤未正常收尾（终态自动恢复）")
-        _save_json(paths.RUNS_DIR / run_id / "run.json", run)
         # 状态回填：起跑与结束都同步任务状态。只回填终态的话，run 在跑、
         # 任务永远显示「排队中」（2026-09-18 实案：「# 重写·续」run 已 running
         # 写了一小时，任务卡 queued，用户以为卡死连点重试）。
@@ -826,6 +990,16 @@ def update_run(run_id, expected_status=None, **fields):
             if task:
                 task["status"] = st
                 _save_json(paths.TASKS_DIR / (tid + ".json"), task)
+        if st in TERMINAL_STATUSES:
+            try:
+                from . import acceptance
+                run["acceptance"] = acceptance.evaluate_run(
+                    run, (_TASKS.get(run.get("task_id")) if run.get("task_id") else None) or {})
+            except Exception:
+                # Acceptance reporting is evidence enrichment; it must not
+                # prevent the primary run from reaching a terminal state.
+                pass
+        _save_json(paths.RUNS_DIR / run_id / "run.json", run)
         if st in TERMINAL_STATUSES and prev_status not in TERMINAL_STATUSES:
             # 事后对账：只有「非终态 → 终态」这一次翻转才记账，重复写终态
             # （UI 补写、迟到写手）不得二次入账。快照在锁内浅拷，锁外调用
@@ -1038,6 +1212,29 @@ def run_artifacts(run_id, limit=200, cache=False):
                               running=running, cache=cache)
 
 
+def run_asset_refs(run_id, limit=200):
+    """Return content-addressed references for this run's output files."""
+    wd, files = run_artifacts(run_id, limit=limit, cache=False)
+    if not wd:
+        return []
+    from . import attachments
+    refs = attachments.asset_refs(wd, [{"path": f.get("name") or f.get("path") or ""}
+                                       for f in files if isinstance(f, dict)])
+    try:
+        from . import assets
+        run = get_run(run_id) or {}
+        for ref in refs:
+            saved = assets.register(wd, ref.get("path"), owner_type="run",
+                                    owner_id=run_id, name=ref.get("name"),
+                                    mime=ref.get("mime"))
+            if saved:
+                ref.update({k: saved[k] for k in ("asset_id", "revision_id",
+                                                  "content_sha256")})
+    except Exception:
+        pass
+    return refs
+
+
 def task_step_count(task_id):
     """该任务所有 run 的步骤总数（含进行中）。
 
@@ -1529,30 +1726,47 @@ def rename_task(task_id, title):
 
 
 def update_task_params(task_id, patch):
-    """更新任务编排参数（对话条三件套：mode/thinking/direct_provider_id/direct_model）。
-    只在任务未在跑时生效——运行中改参数不会影响当前执行，静默跳过避免误导。"""
+    """更新任务编排参数（对话条三件套：mode/thinking/direct_provider_id/direct_model，
+    另有 timeout_s 任务总时限）。
+    只在任务未在跑时生效——运行中改参数不会影响当前执行，静默跳过避免误导。
+
+    修订 CAS（借鉴 WorkDSH 契约纪律 expectedRevision）：patch 携带整数
+    expected_rev 时与任务当前 rev 比对，不符=别处已先改——拒绝写入并说明，
+    不做过期覆盖；不带 expected_rev 的旧调用保持原行为。写成功 rev+1，
+    调用方（路由）从 get_task 回读新 rev 交前端续用。"""
     if not _valid_id(task_id):
         return False, "非法的任务 ID"
+    patch = dict(patch) if isinstance(patch, dict) else {}
+    expected_rev = patch.pop("expected_rev", None)
     with LOCK:
         task = _TASKS.get(task_id)
         if not task:
             return False, "任务不存在"
         if task.get("status") in ("queued", "running"):
             return False, "任务运行中，参数不可改"
+        if expected_rev is not None:
+            try:
+                want_rev = int(expected_rev)
+                cur_rev = int(task.get("rev") or 1)
+            except (TypeError, ValueError):
+                return False, "expected_rev 必须是整数"
+            if want_rev != cur_rev:
+                return False, ("任务参数已被其他窗口修改（当前修订 %d），"
+                               "已拒绝过期写入，界面稍后自动同步最新值" % cur_rev)
         changed = False
-        mode = str((patch or {}).get("mode") or "").strip()
+        mode = str(patch.get("mode") or "").strip()
         if mode in ("auto", "fast", "expert", "manual"):
             if task.get("mode") != mode:
                 task["mode"] = mode
                 changed = True
-        thinking = str((patch or {}).get("thinking") or "").strip()
+        thinking = str(patch.get("thinking") or "").strip()
         if thinking in ("auto", "low", "standard", "high"):
             if task.get("thinking") != thinking:
                 task["thinking"] = thinking
                 changed = True
         if task.get("engine") == "direct":
-            pv = str((patch or {}).get("direct_provider_id") or "").strip()[:80]
-            mv = str((patch or {}).get("direct_model") or "").strip()[:160]
+            pv = str(patch.get("direct_provider_id") or "").strip()[:80]
+            mv = str(patch.get("direct_model") or "").strip()[:160]
             if mv and not pv:
                 return False, "指定对话模型时必须同时指定厂商"
             if task.get("direct_provider_id") != pv or task.get("direct_model") != mv:
@@ -1561,30 +1775,49 @@ def update_task_params(task_id, patch):
                 changed = True
             # 执行 CLI 可改/可清（清空 = 回内置智能体或模型绑定）；与模型绑定
             # 允许共存，运行时 CLI 优先
-            av = str((patch or {}).get("direct_agent") or "").strip()[:80]
+            av = str(patch.get("direct_agent") or "").strip()[:80]
             if task.get("direct_agent") != av and (av or task.get("direct_agent")):
                 task["direct_agent"] = av
                 changed = True
+        # 任务总时限（秒）：建任务时按工作量给默认值，此处支持空闲时显式改。
+        # 下次重试/续跑按新值重新给预算；已在跑的 run 保持原截止时间。
+        # 统一钳制 [1s, 7天]，脏值回落基础默认。
+        raw_ts = patch.get("timeout_s")
+        if raw_ts not in (None, ""):
+            new_ts = _task_timeout_s(raw_ts)
+            if task.get("timeout_s") != new_ts:
+                task["timeout_s"] = new_ts
+                changed = True
         if changed:
+            task["rev"] = int(task.get("rev") or 1) + 1
             _save_json(paths.TASKS_DIR / (task_id + ".json"), task)
             bump_state()
     return True, ""
 
 
-def add_step(run_id, role, agent_id, agent_label, note="", model=""):
+def add_step(run_id, role, agent_id, agent_label, note="", model="", provider=""):
     with LOCK:
         run = _RUNS.get(run_id)
         if not run:
             return None, None
         n = len(run["steps"]) + 1
-        step = {
+        step = {
             "n": n, "role": role, "agent": agent_id, "agent_label": agent_label,
             "note": note,
             # 起跑即记解析到的模型：运行中蜂巢卡/步骤列表就能看到「在用哪个模型」，
             # 不必等 finish_step 用实际结果覆盖（换将成功后以实际为准）
             "model": str(model or "")[:80],
-            "status": "running", "started_at": time.strftime("%H:%M:%S"),
-            "ended_at": None, "duration_s": None, "exit_code": None,
+            # 绑定速记（借鉴 WorkDSH ResolvedExecutionBinding）：本步实际用的
+            # provider 链与上游 host——烧错配额/模型漂移直接对账步骤行。
+            # 只记身份不记密钥；老数据无此键，读取一律 .get
+            "provider": str(provider or "")[:120],
+            "status": "running", "started_at": time.strftime("%H:%M:%S"),
+            "started_at_epoch": time.time(),
+            "trace_id": str(run.get("trace_id") or tracing.trace_id(run_id)),
+            "span_id": tracing.span_id(run_id, n),
+            "parent_span_id": str(run.get("root_span_id") or
+                                   tracing.root_span_id(run_id)),
+            "ended_at": None, "duration_s": None, "exit_code": None,
             "summary": "", "log": None, "cost_usd": 0.0, "tokens": 0,
         }
         run["steps"].append(step)
@@ -1654,8 +1887,9 @@ def finish_step(run_id, n, status, summary="", exit_code=None,
             return
         for s in run["steps"]:
             if s["n"] == n:
-                s["status"] = status
-                s["ended_at"] = time.strftime("%H:%M:%S")
+                s["status"] = status
+                s["ended_at"] = time.strftime("%H:%M:%S")
+                s["ended_at_epoch"] = time.time()
                 # 步骤摘要是 CLI 文本的汇聚点（失败时=错误尾巴，成功时=智能体结论）：
                 # 落盘前统一清洗，避免 ANSI/覆写/乱码墙进 UI 与报告
                 s["summary"] = runner.clean_cli_text(summary)
@@ -1664,8 +1898,11 @@ def finish_step(run_id, n, status, summary="", exit_code=None,
                 s["tokens"] = tokens
                 if model:
                     s["model"] = str(model)[:80]
-                if duration_s is not None:
-                    s["duration_s"] = round(duration_s, 1)
+                if duration_s is not None:
+                    s["duration_s"] = round(duration_s, 1)
+                elif s.get("started_at_epoch") is not None:
+                    s["duration_s"] = round(max(
+                        0.0, s["ended_at_epoch"] - s["started_at_epoch"]), 1)
                 if output is not None:
                     # output 是智能体正文（对话气泡直读）：只剥 ANSI，不做噪声折叠
                     # ——正文里的装饰性长串是作者写的，不能替它省略