briefloop 0.31.6__tar.gz → 0.31.7__tar.gz

{briefloop-0.31.6 → briefloop-0.31.7}/CHANGELOG.md

@@ -1,5 +1,13 @@
 # 变更记录
 
+## 0.31.7
+
+- 修复研究工具参数错误被误认成搜索服务故障，作者可按回执纠正参数继续检索。
+- 修复多行原文读取和历史日期过滤，直接研究无需派发 Scout 也能正常交接写稿。
+- 失败事实核查可在原预算内重新尝试，保留旧结果并拒绝旧任务的迟到提交。
+- 默认直写作者支持上下文连续压缩，长研究不再因漏接压缩而中止。
+- Mac 发行接入 Developer ID 签名和 Apple 公证；正式通过状态以发行验收记录为准。
+
 ## 0.31.6
 
 - 收敛新报告入口：按任务自动选择所需流程，保留材料范围、研究投入和篇幅；原有任务配置不改。

{briefloop-0.31.6 → briefloop-0.31.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: briefloop
-Version: 0.31.6
+Version: 0.31.7
 Summary: Local multi-runtime agent workbench for sourced briefings, independent review and skill learning
 Author: Stahl-G
 License-Expression: MIT
@@ -52,7 +52,7 @@ Dynamic: license-file
 
 报告可选中文或英文，英文写作规范、词数和导出版式贯穿写稿与审阅；新增 Excel 报表导出，未安装 OfficeCLI 也能使用。OpenCode 1.x 与 2.x 同时支持，普通审阅和严格审阅分别展示实际能力，重大事实错误与证据缺口仍阻止正式交付。
 
-本轮提供 macOS Apple Silicon、Python/CLI/PyPI 与 GitHub 源码；Windows 安装包沿用既有稳定版，待原生验收后单独更新。本轮 Mac 包未做 Developer ID 签名或 Apple 公证，平台验证与工件哈希见发行页。完整变更见 [CHANGELOG](CHANGELOG.md)。
+提供 macOS Apple Silicon、Windows 和 Python/CLI 安装入口。Mac 签名、公证、各平台验收状态及工件哈希见对应发行页。完整变更见 [CHANGELOG](CHANGELOG.md)。
 
 新工作区默认优先 Tavily，同时允许宿主自带搜索补查；可主动启用博查、智谱搜索或 DuckDuckGo，按跨市场覆盖或具体缺口补查。受控渠道共用本轮预算，搜索失败、实际发现渠道与原文分别保留。Tavily 需要 API Key，额度以服务账号为准；宿主自带搜索取决于实际工具与权限。
 

{briefloop-0.31.6 → briefloop-0.31.7}/README.md

@@ -12,7 +12,7 @@
 
 报告可选中文或英文，英文写作规范、词数和导出版式贯穿写稿与审阅；新增 Excel 报表导出，未安装 OfficeCLI 也能使用。OpenCode 1.x 与 2.x 同时支持，普通审阅和严格审阅分别展示实际能力，重大事实错误与证据缺口仍阻止正式交付。
 
-本轮提供 macOS Apple Silicon、Python/CLI/PyPI 与 GitHub 源码；Windows 安装包沿用既有稳定版，待原生验收后单独更新。本轮 Mac 包未做 Developer ID 签名或 Apple 公证，平台验证与工件哈希见发行页。完整变更见 [CHANGELOG](CHANGELOG.md)。
+提供 macOS Apple Silicon、Windows 和 Python/CLI 安装入口。Mac 签名、公证、各平台验收状态及工件哈希见对应发行页。完整变更见 [CHANGELOG](CHANGELOG.md)。
 
 新工作区默认优先 Tavily，同时允许宿主自带搜索补查；可主动启用博查、智谱搜索或 DuckDuckGo，按跨市场覆盖或具体缺口补查。受控渠道共用本轮预算，搜索失败、实际发现渠道与原文分别保留。Tavily 需要 API Key，额度以服务账号为准；宿主自带搜索取决于实际工具与权限。
 

briefloop-0.31.7/VERSION

@@ -0,0 +1 @@
+0.31.7

{briefloop-0.31.6 → briefloop-0.31.7}/desktop/electron/package-lock.json

@@ -1,12 +1,12 @@
 {
   "name": "briefloop-desktop",
-  "version": "0.31.6",
+  "version": "0.31.7",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "briefloop-desktop",
-      "version": "0.31.6",
+      "version": "0.31.7",
       "hasInstallScript": true,
       "license": "MIT",
       "dependencies": {

{briefloop-0.31.6 → briefloop-0.31.7}/desktop/electron/package.json

@@ -1,6 +1,6 @@
 {
   "name": "briefloop-desktop",
-  "version": "0.31.6",
+  "version": "0.31.7",
   "private": true,
   "description": "BriefLoop local desktop workspace",
   "author": "Stahl-G",

{briefloop-0.31.6 → briefloop-0.31.7}/docs/default-behavior.md

@@ -12,10 +12,10 @@
 | 报告反馈继续补查 | Native 报告聊天与其他宿主一样，在该消息获准联网后提供实际搜索、网页读取工具 | 绑定本消息、原稿版本与剩余预算；普通聊天、只读讨论、离线反馈不会取得该权限，完成或中断后失效 |
 | 直写稿独立事实核查 | 已有依据补全调用选择最多 12 条公开事实候选，定位到正文与原件后进入默认核查流程 | 候选不是正确性判定；联网核查及独立审阅仍各自执行。只处理可确认的匿名公开 HTTP 原件，混有未确认材料时不自动提取 |
 | 核查缺口显示 | 检查栏显示当前版本未进入核查的原因 | 旧稿原因不覆盖新稿，真实结果和正在执行的阶段优先；不往报告正文追加流程声明 |
-| 失败检查恢复 | 绑定原稿的检查可按当前模型新建重试，保留失败记录 | 只重做检查，正文不变；实际模型和思考档位单独记录 |
+| 失败检查恢复 | 绑定原稿的检查可恢复或按当前模型重试，保留失败记录与已用额度 | 工具或抓取失败后的核查可重试；只重做检查，实际模型和思考档位单独记录 |
 
 公开原件准入当前不包含上传文件、连接器或通用提供方响应，也不接受含凭据、查询参数或片段的地址。
-空内容的失败公开抓取不会阻断同报告其他有效原件，但不能作为候选依据。已有标准流程及手动登记的主张维持原有行为。
+失败或空正文来源也必须通过公开资料许可检查；不能因抓取失败而放行其 URL。混有未确认来源时，自动候选提取与外部事实核查均停止；已有标准流程及手动登记的主张维持原有行为。
 候选选择仍会遗漏事实；没有候选、搜索额度不足或检查未完成都不能显示为整稿核实。
 自动派发和实际执行前会复核公开资料范围，失效的旧稿绑定不进入新版核查。
 事实核查查询可以查历史背景，不被本期新闻窗口覆盖；普通 Tavily 研究按 UTC 日期扩大到覆盖边界，报告自身时间范围保持原值。

{briefloop-0.31.6 → briefloop-0.31.7}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "briefloop"
-version = "0.31.6"
+version = "0.31.7"
 description = "Local multi-runtime agent workbench for sourced briefings, independent review and skill learning"
 readme = "README.md"
 requires-python = ">=3.11"

briefloop-0.31.7/src/briefloop/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.31.7"

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/cli.py

@@ -15,11 +15,14 @@ from . import __version__
 
 def _search_failure(operation,exc,provider='tavily'):
     """A structured, redacted failure so the Scout's path-switching can act on it."""
-    return {'provider':getattr(exc,'provider',None) or provider,'operation':operation,'status':'failed',
-            'failure_kind':getattr(exc,'failure_kind','provider_error'),
-            'http_status':getattr(exc,'status',None),
-            'error':str(exc),
-            'request_record_path':getattr(exc,'request_record_path',None)}
+    result = {'provider':getattr(exc,'provider',None) or provider,'operation':operation,'status':'failed',
+              'failure_kind':getattr(exc,'failure_kind','provider_error'),
+              'http_status':getattr(exc,'status',None),
+              'error':str(exc),
+              'request_record_path':getattr(exc,'request_record_path',None)}
+    if result['failure_kind'] == 'invalid_arguments':
+        result.update(recoverable=True, recovery_action='correct_arguments')
+    return result
 
 
 def main():

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/draft_completion.py

@@ -5,7 +5,7 @@ an independent fact check, and completing this job does not create a release.
 """
 import hashlib
 import json
-from .store import dump, Conflict
+from .store import dump, Conflict, uid, now
 
 
 def source_metadata(store, source_id):
@@ -298,28 +298,41 @@ def require_fact_check_settled(store, job):
         raise ValueError('本版本候选核查尚未完成（'+str(stage.get('status') or '未接纳')+'）；请先处理核查预算或中断状态，再恢复检查。')
 
 
-def resume_cancelled_stage(connection, job):
-    """Explicit parent resume restores only its own cancelled stage/budget.
+def resume_fact_check_stage(connection, job):
+    """Resume an owned cancelled check or retry a failed one with the same allowance.
 
-    Keep the cancellation history and request ledger. No new allowance or stage
-    is created, and a budget-exhausted/completed stage is never reopened here.
+    Failed attempts retain their own stage identity, rejecting late results.
+    Request charges, source bindings and the original budget remain unchanged.
     """
     payload=json.loads(job['payload'])
     if not payload.get('continuation_of'):return
     from .research_plan import _read_plan, _save_plan
     plan=_read_plan(connection,payload['run_id']) or {}
     stage=plan.get('fact_check') or {}
-    if stage.get('status')!='cancelled':return
+    if stage.get('status') not in ('cancelled', 'failed'):return
     versions = _fact_check_versions(None, job, payload, connection=connection)
     owners = [attempt['id'] for attempt in _check_attempts(None, job, payload, connection=connection)]
-    owned=connection.execute("SELECT id FROM jobs WHERE kind='fact_check' AND status='cancelled' "
+    statuses = ('cancelled',) if stage['status'] == 'cancelled' else ('complete', 'failed', 'interrupted', 'cancelled')
+    owned=connection.execute("SELECT id FROM jobs WHERE kind='fact_check' AND status IN ("
+        + ','.join('?' for _ in statuses) + ') '
+
         "AND json_extract(payload,'$.parent_job_id') IN (" + ','.join('?' for _ in owners) + ') '
         "AND json_extract(payload,'$.run_id')=? "
         "AND json_extract(payload,'$.version_id') IN (" + ','.join('?' for _ in versions) + ') '
-        "AND json_extract(payload,'$.stage_id')=?", (*owners,payload['run_id'],*versions,stage.get('stage_id'))).fetchone()
+        "AND json_extract(payload,'$.stage_id')=?", (*statuses,*owners,payload['run_id'],*versions,stage.get('stage_id'))).fetchone()
     if not owned:return
-    stage.setdefault('interrupted_closures',[]).append({key:stage.get(key) for key in ('status','closed','outcome')})
-    stage.update(status='active',closed=None,outcome=None)
+    if stage['status'] == 'failed':
+        from .research_budget import restore_fact_check_budget_offset
+        retry = {**stage, 'status': 'active', 'closed': None, 'outcome': None}
+        # Restore old published offsets against the old request ledger before
+        # assigning a new identity. This cannot replenish consumed allowance.
+        restore_fact_check_budget_offset(connection, payload['run_id'], {**plan, 'fact_check': retry})
+        retry.update(stage_id=uid('fchk'), created=now(), retry_of_stage_id=stage['stage_id'])
+        plan.setdefault('fact_check_history', []).append(stage)
+        plan['fact_check'] = retry
+    else:
+        stage.setdefault('interrupted_closures', []).append({key: stage.get(key) for key in ('status', 'closed', 'outcome')})
+        stage.update(status='active', closed=None, outcome=None)
     _save_plan(connection,payload['run_id'],plan)
 
 

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/native_orchestrator.py

@@ -315,9 +315,9 @@ def _refresh_research(store, config):
     else:
         from .research_plan import frozen
         from .scout_coverage import view
-        # A new goal run can research directly. Do not fabricate Scout receipts
-        # or hide promised workers when creating the writer's input container.
-        if not (frozen(store, config['run_id']) or {}).get('goal_contract') or view(store, config['run_id'])['scout_execution']:
+        # Any frozen strategy can research directly without inventing Scout
+        # receipts. Promised workers still need their normal join/closeout.
+        if not frozen(store, config['run_id']) or view(store, config['run_id'])['scout_execution']:
             return
         research = {'sources': [], 'gaps': []}
     research = current_research(store, config['run_id'], research, register=True)

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/native_roles.py

@@ -481,6 +481,9 @@ def source_read(store, config, args):
     limit = min(int(args.get('max_chars') or READ_CHARS), READ_CHARS)
     try:
         offset = args.get('start_char')
+        # Zero offset does not conflict with an explicit multi-line window.
+        if type(offset) is int and offset == 0 and end is not None and end != (start or 1):
+            offset = None
         if offset is not None:
             lines = store.source_text(sid).splitlines(); number = start or 1
             if (type(offset) is not int or offset < 0 or number < 1 or number > len(lines)
@@ -628,7 +631,8 @@ SCOUT_READ_TOOLS = [
      'parameters': {'type': 'object', 'required': ['source_id'], 'additionalProperties': False,
                     'properties': {'source_id': {'type': 'string'},
                                    'start_line': {'type': 'integer', 'minimum': 1},
-                                   'start_char': {'type': 'integer', 'minimum': 0, 'description': '超长单行的字符偏移，从 0 开始；只读取 start_line 这一行'},
+                                   'start_char': {'type': 'integer', 'minimum': 0,
+                                                  'description': '单行字符偏移，从 0 开始。非零偏移只读 start_line，end_line 须相同；0 可用于正常多行读取。'},
                                    'end_line': {'type': 'integer', 'minimum': 1},
                                    'max_chars': {'type': 'integer', 'minimum': 1}}},
      'handler': source_read},
@@ -654,7 +658,8 @@ def _scout_web_tools(channels):
                                        'query': {'type': 'string'},
                                        'purpose': {'type': 'string', 'enum': ['primary', 'coverage_probe', 'gap_repair']},
                                        'reason': {'type': 'string', 'description': '这次想多知道什么'},
-                                       'gap_id': {'type': 'string'},
+                                       'gap_id': {'type': 'string',
+                                                  'description': '可省略。仅引用现有计划已登记的真实缺口 ID；gap_repair 先取得 ID，primary/coverage_probe 不要编造。'},
                                        'topic': {'type': 'string', 'enum': ['general', 'news']},
                                        'time_range': {'type': 'string', 'enum': ['day', 'week', 'month', 'year']},
                                        'start_date': _DATE, 'end_date': _DATE,

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/runtime.py

@@ -725,12 +725,12 @@ class Worker:
                 if job['kind']=='learn':
                     self._check_feedback_owner(c,jid,payload)
                 if job['kind']=='assess' and payload.get('continuation_of'):
-                    from .draft_completion import resume_cancelled_stage
+                    from .draft_completion import resume_fact_check_stage
                     if c.execute("SELECT id FROM jobs WHERE id<>? AND status IN ('queued','running') AND json_extract(payload,'$.continuation_of')=?",(jid,payload['continuation_of'])).fetchone():
                         raise ValueError('本报告另一个检查任务仍在执行，请先等待或停止它')
                     if any(child['id'] in self._review_jobs for child in c.execute("SELECT id FROM jobs WHERE json_extract(payload,'$.parent_job_id')=?",(jid,))):
                         raise ValueError('关联核查仍在停止，请稍后恢复')
-                    resume_cancelled_stage(c,job)
+                    resume_fact_check_stage(c,job)
                 # Resume keeps the frozen configuration and original task identity.
                 payload['attempt']=int(payload.get('attempt',1))+1
                 c.execute("UPDATE jobs SET payload=?,status='queued',error=NULL,updated=? WHERE id=?",(dump(payload),now(),jid))
@@ -824,8 +824,8 @@ class Worker:
                     frozen['checks_identity']=hashlib.sha256(dump([
                         original['continuation_of'],payload['checks_input'],identity_payload]).encode()).hexdigest()
                     connection.execute('UPDATE jobs SET payload=? WHERE id=?',(dump(frozen),new_id))
-                    from .draft_completion import resume_cancelled_stage
-                    resume_cancelled_stage(connection, {'id': new_id, 'payload': dump(frozen)})
+                    from .draft_completion import resume_fact_check_stage
+                    resume_fact_check_stage(connection, {'id': new_id, 'payload': dump(frozen)})
                     return
                 if job['kind']!='learn':return
                 self._check_feedback_owner(connection,jid,original)

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/static/index.html

@@ -79,7 +79,7 @@
 <div class="settings-page-header"><div><p class="eyebrow">BRIEFLOOP · 设置</p><h1 id="settings-title">模型与工作设置</h1><p class="help">配置执行引擎、模型与工作习惯。</p></div><button id="settings-close" type="button" class="outline">返回 ↗</button></div>
 <div class="settings-layout"><nav class="settings-nav" aria-label="设置分类"><button type="button" data-settings-view="models" class="active">模型与提供商</button><button type="button" data-settings-view="connectors">数据连接器（MCP）</button><button type="button" data-settings-view="execution">运行与搜索</button><button type="button" data-settings-view="learning">反馈与学习</button><button type="button" data-settings-view="workspaces">工作区</button><button type="button" data-settings-view="updates">版本与更新</button></nav>
 <div class="settings-content"><section id="settings-view-models"><div class="settings-segments" role="tablist" aria-label="模型接入方式"><button id="settings-tab-api" role="tab" aria-selected="false" aria-controls="settings-api" type="button" title="基于 Pi SDK">BriefLoop Agent</button><button id="settings-tab-cli" role="tab" aria-selected="true" aria-controls="settings-cli" type="button">Agent CLI</button></div><div id="settings-session-note" class="settings-session-note help" hidden></div><section id="settings-cli" role="tabpanel" aria-labelledby="settings-tab-cli"><div id="settings-runtime-list"></div><div id="settings-model-block"></div></section><section id="settings-api" role="tabpanel" aria-labelledby="settings-tab-api" hidden><div id="settings-native-runtime"></div></section></section>
-<section id="settings-view-updates" data-web-version="0.31.6" hidden>
+<section id="settings-view-updates" data-web-version="0.31.7" hidden>
 <h2>版本与更新</h2><p id="app-update-version">正在读取实际运行版本…</p><p id="app-update-latest-version" class="help">尚未检查可用版本</p><pre id="app-update-command" hidden></pre>
 <p id="app-update-status" role="status" aria-live="polite">尚未检查更新</p>
 <p id="app-update-test-source" class="help" hidden>本地测试更新源 · 仅验证流程，不代表官方发布</p>

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/static/native-engine.mjs

@@ -2763,7 +2763,7 @@ Options:
 4. \u5173\u952E\u7ED3\u8BBA\u5BF9\u5E94\u7684\u539F\u6587\u4F4D\u7F6E\uFF0C\u4EE5\u53CA\u6539\u53D8\u884C\u52A8\u7684\u524D\u63D0\u3001\u8303\u56F4\u3001\u65E5\u671F\u3001\u53E3\u5F84\u3001\u53CD\u4F8B\u3001\u66F4\u6B63/\u53D6\u4EE3\u5173\u7CFB\uFF1B\u6765\u6E90\u4E8B\u5B9E\u3001\u4F5C\u8005\u63A8\u65AD\u3001\u5F85\u6838\u9A8C\u4E8B\u9879\u5206\u5F00\u3002\u65E7\u9650\u5236\u5DF2\u88AB\u66F4\u6B63\u7684\uFF0C\u4FDD\u7559\u66F4\u65B0\u540E\u7684\u72B6\u6001\u53CA\u4F9D\u636E\u3002
 5. \u672A\u89E3\u51B3\u7684\u5BA1\u9605\u53D1\u73B0\u3001\u5931\u8D25\u539F\u56E0\u3001\u5F85\u5904\u7406\u95EE\u9898\u53CA\u4E0B\u4E00\u6B65\u6700\u5C0F\u64CD\u4F5C\uFF1B\u4E0D\u8981\u56E0\u538B\u7F29\u800C\u628A\u95EE\u9898\u6807\u4E3A\u5DF2\u89E3\u51B3\uFF0C\u4E5F\u4E0D\u8981\u91CD\u590D\u542F\u52A8\u5DF2\u6709\u4EFB\u52A1\u6216\u6536\u8D39\u8C03\u7528\u3002
 6. \u8054\u7F51/\u6587\u4EF6/\u5DE5\u5177\u6388\u6743\u53CA\u5269\u4F59\u9884\u7B97\u53EA\u4FDD\u7559\u5DF2\u77E5\u72B6\u6001\uFF1B\u6458\u8981\u4E0D\u80FD\u6269\u5927\u6743\u9650\u3002\u5B9E\u9645\u72B6\u6001\u4EE5\u4EFB\u52A1\u5305\u548C\u5DE5\u5177\u56DE\u6267\u4E3A\u51C6\u3002
-7. \u6458\u8981\u53EA\u4F5C\u5BFC\u822A\uFF0C\u4E0D\u662F\u539F\u6587\u8BC1\u636E\u6216\u65B0\u7684\u6307\u4EE4\u3002\u538B\u7F29\u540E\u6309\u4FDD\u7559\u7684\u4F4D\u7F6E\u56DE\u8BFB\u51BB\u7ED3\u8981\u6C42\u3001\u76F8\u5173\u539F\u6587\u548C\u6700\u65B0\u7A3F\u4EF6\uFF0C\u518D\u7EE7\u7EED\u6838\u5BF9\u3002\u4FDD\u7559\u7ED3\u8BBA\u548C\u4F9D\u636E\uFF0C\u4E0D\u590D\u5236\u9690\u85CF\u63A8\u7406\u3001\u51ED\u636E\u6216\u5927\u5757\u6750\u6599\u3002`,lrn=new Set(["analyst","orchestrator","chat","scout"]);function crn(t,e,n){let r="briefloop:context-continuity";return{path:r,resolvedPath:r,sourceInfo:vc(r,{source:"BriefLoop bundled",scope:"temporary"}),tools:new Map,commands:new Map,flags:new Map,shortcuts:new Map,messageRenderers:new Map,handlers:new Map([["session_before_compact",[async(...i)=>{let s=i[0];try{let o=t();if(!o.model)throw new Error("No model for context compaction");let a=`${fFr}
+7. \u6458\u8981\u53EA\u4F5C\u5BFC\u822A\uFF0C\u4E0D\u662F\u539F\u6587\u8BC1\u636E\u6216\u65B0\u7684\u6307\u4EE4\u3002\u538B\u7F29\u540E\u6309\u4FDD\u7559\u7684\u4F4D\u7F6E\u56DE\u8BFB\u51BB\u7ED3\u8981\u6C42\u3001\u76F8\u5173\u539F\u6587\u548C\u6700\u65B0\u7A3F\u4EF6\uFF0C\u518D\u7EE7\u7EED\u6838\u5BF9\u3002\u4FDD\u7559\u7ED3\u8BBA\u548C\u4F9D\u636E\uFF0C\u4E0D\u590D\u5236\u9690\u85CF\u63A8\u7406\u3001\u51ED\u636E\u6216\u5927\u5757\u6750\u6599\u3002`,lrn=new Set(["analyst","orchestrator","chat","scout","quick_writer"]);function crn(t,e,n){let r="briefloop:context-continuity";return{path:r,resolvedPath:r,sourceInfo:vc(r,{source:"BriefLoop bundled",scope:"temporary"}),tools:new Map,commands:new Map,flags:new Map,shortcuts:new Map,messageRenderers:new Map,handlers:new Map([["session_before_compact",[async(...i)=>{let s=i[0];try{let o=t();if(!o.model)throw new Error("No model for context compaction");let a=`${fFr}
 \u5F53\u524D\u89D2\u8272\uFF1A${e}\u3002`+(s.customInstructions?`
 \u672C\u6B21\u989D\u5916\u538B\u7F29\u91CD\u70B9\uFF08\u4E0D\u6539\u53D8\u6743\u9650\u6216\u51BB\u7ED3\u8981\u6C42\uFF09\uFF1A
 ${s.customInstructions}`:""),l=o.agent.streamFunction,u=await fB(s.preparation,o.model,void 0,void 0,void 0,s.signal,o.thinkingLevel,(c,p,d)=>l(c,{...p,messages:[p.messages[0],{role:"system",content:a,timestamp:0},...p.messages.slice(1)]},d),void 0,o.settingsManager.getRetrySettings());return{compaction:{...u,details:{...u.details||{},briefloop_policy:h8}}}}catch{return s.signal.aborted||n(),{cancel:!0}}}]]])}}$l();_v();mk();import{createHash as gFr}from"node:crypto";import{realpathSync as mrn,readFileSync as m8,readdirSync as yFr,statSync as v3e}from"node:fs";import{extname as urn,isAbsolute as frn,resolve as k3e,sep as I3e}from"node:path";import{Script as AFr}from"node:vm";var f8=6e4,prn=20*1024*1024,_Fr=20*1024*1024,w3e=160,Jue={".png":"image/png",".jpg":"image/jpeg",".jpeg":"image/jpeg",".gif":"image/gif",".webp":"image/webp"},drn=new Set([".pdf",".bin",".xlsx",".xls",".docx",".doc",".zip",".pptx",...Object.keys(Jue)]);function ZF(t,e){if(!e||frn(e)||e.includes("\0"))throw new Error("packet paths are relative to the packet root");let n=k3e(t,e),r=t.endsWith(I3e)?t:t+I3e;if(n!==t&&!n.startsWith(r))throw new Error("packet tools only resolve inside the review packet");let i=mrn(n);if(i!==n&&!i.startsWith(r))throw new Error("packet tools refuse symlink escapes");if(!v3e(i).isFile())throw new Error(`not a packet file: ${e}`);return i}function R3e(t,e,n){for(let r of yFr(t,{withFileTypes:!0})){if(r.isSymbolicLink())continue;let i=k3e(t,r.name);r.isDirectory()?R3e(i,e,n):r.isFile()&&n.push(i.slice(e.length+1).split(I3e).join("/"))}}function IC(t,e={}){return{content:[{type:"text",text:t}],details:e}}function hrn(t,e=f8){return t.length<=e?t:t.slice(0,e)+`

briefloop-0.31.7/src/briefloop/static/release-notes.json

@@ -0,0 +1,4 @@
+{
+  "version": "0.31.7",
+  "notes": "- 修复研究工具参数错误被误认成搜索服务故障，作者可按回执纠正参数继续检索。\n- 修复多行原文读取和历史日期过滤，直接研究无需派发 Scout 也能正常交接写稿。\n- 失败事实核查可在原预算内重新尝试，保留旧结果并拒绝旧任务的迟到提交。\n- 默认直写作者支持上下文连续压缩，长研究不再因漏接压缩而中止。\n- Mac 发行接入 Developer ID 签名和 Apple 公证；正式通过状态以发行验收记录为准。"
+}

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop/websearch.py

@@ -15,7 +15,8 @@ PROVIDER_LABELS={'tavily':'Tavily','duckduckgo':'DuckDuckGo','bocha':'博查','z
 
 # Stable, redacted classification for the request envelope. Do not infer a
 # provider's billing outcome from these; they only say why an attempt stopped.
-FAILURE_KINDS=('auth','quota','rate_limit','timeout','tls','network','invalid_response','too_large','provider_error','budget','local_error')
+FAILURE_KINDS=('auth','quota','rate_limit','timeout','tls','network','invalid_response','too_large',
+               'provider_error','budget','local_error','invalid_arguments','policy')
 
 
 class SearchError(ValueError):
@@ -88,10 +89,14 @@ def check_run(store,run_id,provider):
     fail=getattr(provider_module(provider),'ProviderError',SearchError)
     run=store.one('runs',run_id)
     if not json.loads(run['requirements']).get('allow_web'):
-        error=fail('本轮未允许联网搜索');error.provider=provider;raise error
+        error=marked('本轮未允许联网搜索','policy',cls=fail)
+        error.provider=provider
+        raise error
     from .search_policy import for_run,allowed
     if provider not in allowed(for_run(store,run_id)):
-        error=fail('本轮搜索服务不是 '+PROVIDER_LABELS[provider]+'，请在设置中选择后发起新任务');error.provider=provider;raise error
+        error=marked('本轮搜索服务不是 '+PROVIDER_LABELS[provider]+'，请在设置中选择后发起新任务','policy',cls=fail)
+        error.provider=provider
+        raise error
     return run
 
 
@@ -124,11 +129,11 @@ def search(query,*,provider=None,topic='general',time_range=None,start_date=None
            key_file=None,store=None,run_id=None,purpose='primary',reason='',gap_id=None):
     """One metered search request; provider stays swappable behind one envelope."""
     if provider is None:
-        if store is None or run_id is None:raise SearchError('未指定搜索 provider')
+        if store is None or run_id is None:raise marked('未指定搜索 provider','invalid_arguments')
         provider=provider_for_run(store,run_id)
     else:provider=normalize_provider(provider)
-    if purpose not in ('primary','coverage_probe','gap_repair'):raise SearchError('无效搜索用途')
-    if not isinstance(reason,str) or len(reason)>1000:raise SearchError('搜索原因过长')
+    if purpose not in ('primary','coverage_probe','gap_repair'):raise marked('无效搜索用途','invalid_arguments')
+    if not isinstance(reason,str) or len(reason)>1000:raise marked('搜索原因过长','invalid_arguments')
     module=provider_module(provider)
     if store is not None and run_id and provider == 'tavily':
         import json
@@ -148,13 +153,18 @@ def search(query,*,provider=None,topic='general',time_range=None,start_date=None
             time_range = None
         # A fact-check query may concern historical/background statements.
         # Its explicit dates (if any) survive instead of inheriting the news window.
+    if provider == 'tavily' and (start_date or end_date):
+        time_range = None
     options={'topic':topic,'time_range':time_range,'start_date':start_date,'end_date':end_date,
              'include_domains':list(include_domains or []),'exclude_domains':list(exclude_domains or []),
              'max_results':max_results,'search_depth':search_depth}
     if provider=='zhipu':
         from .search_policy import for_run
         options['search_engine']=for_run(store,run_id)['zhipu_engine'] if store is not None and run_id else 'search_std'
-    parameters=module.validate_search(query,options)
+    try:
+        parameters=module.validate_search(query,options)
+    except SearchError as exc:
+        raise marked(str(exc),'invalid_arguments',cls=type(exc)) from exc
     reservation=None;local_id=None;round_id=None;claim=None
     if store is not None and run_id is not None:
         check_run(store,run_id,provider)
@@ -162,7 +172,9 @@ def search(query,*,provider=None,topic='general',time_range=None,start_date=None
             from .research_plan import frozen
             plan=frozen(store,run_id) or {}
             gaps=[g for info in plan.get('rounds',{}).values() for g in info.get('gaps',[])]
-            if not any(g.get('id')==gap_id for g in gaps):raise SearchError('缺口不属于本轮计划')
+            if not any(g.get('id')==gap_id for g in gaps):
+                raise marked('gap_id 不属于本轮已登记缺口；普通搜索可省略，补缺口时先取得计划中的真实 ID',
+                             'invalid_arguments')
         from . import research_budget as budget
         # The identity uses validated options; the shared budget caps admitted
         # URLs, not provider results. Different key files must not join one
@@ -317,7 +329,7 @@ def extract(store,urls,*,run_id=None,provider=None,extract_depth='basic',key_fil
     """Provider-facing extraction entry: Tavily uses its API; keyless providers
     register pages through the same metered direct-fetch path as add-url."""
     if provider is None:
-        if run_id is None:raise SearchError('未指定搜索 provider')
+        if run_id is None:raise marked('未指定搜索 provider','invalid_arguments')
         provider=provider_for_run(store,run_id)
     else:provider=normalize_provider(provider)
     return provider_module(provider).extract(store,urls,run_id=run_id,extract_depth=extract_depth,key_file=key_file)

{briefloop-0.31.6 → briefloop-0.31.7}/src/briefloop.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: briefloop
-Version: 0.31.6
+Version: 0.31.7
 Summary: Local multi-runtime agent workbench for sourced briefings, independent review and skill learning
 Author: Stahl-G
 License-Expression: MIT
@@ -52,7 +52,7 @@ Dynamic: license-file
 
 报告可选中文或英文，英文写作规范、词数和导出版式贯穿写稿与审阅；新增 Excel 报表导出，未安装 OfficeCLI 也能使用。OpenCode 1.x 与 2.x 同时支持，普通审阅和严格审阅分别展示实际能力，重大事实错误与证据缺口仍阻止正式交付。
 
-本轮提供 macOS Apple Silicon、Python/CLI/PyPI 与 GitHub 源码；Windows 安装包沿用既有稳定版，待原生验收后单独更新。本轮 Mac 包未做 Developer ID 签名或 Apple 公证，平台验证与工件哈希见发行页。完整变更见 [CHANGELOG](CHANGELOG.md)。
+提供 macOS Apple Silicon、Windows 和 Python/CLI 安装入口。Mac 签名、公证、各平台验收状态及工件哈希见对应发行页。完整变更见 [CHANGELOG](CHANGELOG.md)。
 
 新工作区默认优先 Tavily，同时允许宿主自带搜索补查；可主动启用博查、智谱搜索或 DuckDuckGo，按跨市场覆盖或具体缺口补查。受控渠道共用本轮预算，搜索失败、实际发现渠道与原文分别保留。Tavily 需要 API Key，额度以服务账号为准；宿主自带搜索取决于实际工具与权限。
 

{briefloop-0.31.6 → briefloop-0.31.7}/tests/test_draft_completion.py

@@ -94,7 +94,7 @@ def finish(store,job,worker):
 
 def test_evidence_child_fact_check_sources_are_admitted_without_accepting_unrelated_sources(tmp_path, monkeypatch):
     from briefloop import research_plan, tavily
-    from briefloop.draft_completion import binding, accept_stage_sources, require_fact_check_settled, resume_cancelled_stage
+    from briefloop.draft_completion import binding, accept_stage_sources, require_fact_check_settled, resume_fact_check_stage
     store = Store(tmp_path)
     run = store.create_run({'title': 'R', 'objective': 'Public update', 'allow_web': True, 'fact_check': True,
         'completion_mode': 'direct', 'search_policy': {'primary_provider': 'tavily', 'coverage_mode': 'primary_only'},
@@ -134,7 +134,7 @@ def test_evidence_child_fact_check_sources_are_admitted_without_accepting_unrela
     retry = worker.retry_with_current_model(parent['id'])
     assert research_plan.frozen(store, run['id'])['fact_check']['status'] == 'active'
     with store.tx() as connection:
-        resume_cancelled_stage(connection, parent)
+        resume_fact_check_stage(connection, parent)
     resumed = research_plan.frozen(store, run['id'])['fact_check']
     assert resumed['status'] == 'active'
     assert resumed['stage_id'] == before_resume['stage_id']
@@ -143,7 +143,7 @@ def test_evidence_child_fact_check_sources_are_admitted_without_accepting_unrela
     assert unchanged(resumed) == unchanged(before_resume)
     assert resumed['interrupted_closures'][-1]['status'] == 'cancelled'
     with store.tx() as connection:
-        resume_cancelled_stage(connection, parent)
+        resume_fact_check_stage(connection, parent)
     assert research_plan.frozen(store, run['id'])['fact_check'] == resumed
     store.update_job(check['id'], 'complete')
     research_plan.finish_fact_check(store, run['id'], status='completed')

{briefloop-0.31.6 → briefloop-0.31.7}/tests/test_fact_check_orchestration.py

@@ -321,3 +321,43 @@ def test_pending_grant_receipt_does_not_add_again_after_the_stage_consumes_it(tm
     assert len(before['fact_check']['initial_grants']) == 1
     assert research_plan.pending_fact_check_grant(store, run['id']) is None
     assert store.rows("SELECT id FROM jobs WHERE kind='fact_check'") == []
+
+
+def test_parent_retry_replaces_failed_check_stage_without_resetting_allowance(tmp_path):
+    from briefloop.runtime import Worker
+    from briefloop.draft_completion import binding, require_fact_check_settled
+    store, run, _ = _world(tmp_path)
+    brief, claim, _ = _claim_version(store, run)
+    research_plan.finish_round(store, run['id'])
+    parent = store.enqueue('assess', {'run_id': run['id'], 'version_id': brief['id'],
+        'runtime': {'model': 'test'}, 'continuation_of': 'job_original', 'checks_input': binding(store, brief['id'])})
+    worker = Worker(store)
+    first = worker._admit_fact_check(parent, brief)
+    old = store.rows("SELECT * FROM jobs WHERE kind='fact_check'")[0]
+    research_plan.add_fact_check_grant(store, run['id'], SHARE)
+    request = budget.reserve_search(store, run['id'], 1)
+    before = budget.snapshot(store, run['id'])
+    failed_result = {'version_id': brief['id'], 'stage_id': first['stage_id'],
+        'selection': {'claim_ids': [claim['id']]}, 'execution': {'status': 'failed', 'summary': 'Source reader failed'},
+        'candidates': [{'claim_id': claim['id'], 'status': 'insufficient_evidence', 'reason': 'Could not read source',
+            'span_ids': [], 'query_ids': [request['request_id']]}]}
+    saved = fact_check.submit_result(store, run['id'], failed_result, job_id=old['id'])
+    store.update_job(old['id'], 'complete', result={'execution': 'failed'})
+    store.update_job(parent['id'], 'failed', error='Fact check failed')
+    worker.resume(parent['id'])
+    retry_stage = research_plan.frozen(store, run['id'])['fact_check']
+    assert retry_stage['status'] == 'active' and retry_stage['stage_id'] != first['stage_id']
+    assert retry_stage['retry_of_stage_id'] == first['stage_id']
+    after = budget.snapshot(store, run['id'])
+    assert after['limits'] == before['limits'] and after['used'] == before['used']
+    assert fact_check.get_record(store, saved['record']['id'])['data']['execution']['status'] == 'failed'
+    with pytest.raises(ValueError):
+        fact_check.submit_result(store, run['id'], failed_result, job_id=old['id'])
+    with pytest.raises(ValueError):
+        require_fact_check_settled(store, store.one('jobs', parent['id']))
+    worker._admit_fact_check(store.one('jobs', parent['id']), brief)
+    worker._admit_fact_check(store.one('jobs', parent['id']), brief)
+    checks = store.rows("SELECT * FROM jobs WHERE kind='fact_check' ORDER BY rowid")
+    assert len(checks) == 2 and checks[-1]['status'] == 'queued'
+    assert json.loads(checks[-1]['payload'])['stage_id'] == retry_stage['stage_id']
+    assert store.one('jobs', old['id'])['status'] == 'complete'

{briefloop-0.31.6 → briefloop-0.31.7}/tests/test_fact_check_phase.py

@@ -122,3 +122,27 @@ def test_cancelled_stage_rejects_new_and_late_requests(tmp_path):
     with pytest.raises(research_plan.AdmissionError) as blocked:
         budget.reserve_search(store, run['id'], 1)
     assert blocked.value.code == 'fact_check_closed'
+
+
+def test_fact_check_search_absolute_dates_clear_relative_filter(tmp_path, monkeypatch):
+    from briefloop import tavily, websearch
+    store = Store(tmp_path)
+    store.update_settings({'search_provider': 'tavily', 'search_policy': {'primary_provider': 'tavily'}})
+    run = _run(store, period_start='2026-09-21', period_end='2026-09-27',
+               research_budget={'search_requests': 4, 'candidate_urls': 20, 'source_pages': 6})
+    _closed_round(store, run)
+    research_plan.admit_fact_check(store, run['id'], {'kind': 'task_reserve', 'limits': SHARE})
+    requests = []
+
+    def post(endpoint, payload, **kwargs):
+        requests.append(payload)
+        result = {'results': [{'title': 'Historical original', 'url': 'https://example.test/history', 'content': 'Background'}]}
+        return result, json.dumps(result).encode()
+
+    monkeypatch.setattr(tavily, '_post', post)
+    result = websearch.search('historical background', provider='tavily', time_range='year',
+                             start_date='2020-01-01', end_date='2020-12-31', store=store, run_id=run['id'])
+    assert result['status'] == 'ok'
+    assert requests[0]['start_date'] == '2020-01-01' and requests[0]['end_date'] == '2020-12-31'
+    assert 'time_range' not in requests[0]
+    assert budget.snapshot(store, run['id'])['stages']['fact_check']['search_requests'] == 1

{briefloop-0.31.6 → briefloop-0.31.7}/tests/test_native_orchestrator.py

@@ -403,3 +403,37 @@ def test_revision_responses_without_review_findings_are_told_to_send_none():
     with pytest.raises(ValueError, match='"expected": \\["review_1"\\]'):
         validate_arrays([], answer, {'review_1'})
     validate_arrays([], [], set())
+
+
+def test_guided_direct_research_without_scout_reaches_writer(tmp_path):
+    class DirectEngine(FlowEngine):
+        def next(self, sid):
+            if self.sessions[sid]['role'] == 'orchestrator' and not getattr(self, 'direct_started', False):
+                self.direct_started = True
+                self.queues[sid] = [
+                    ('save_plan', {'plan': {'summary': 'Read the supplied revenue',
+                        'reader_contract': contract(self.store, self.run['id']), 'scout_tasks': []}}),
+                    ('source_read', {'source_id': self.source['id']}),
+                    ('save_research_handoff', {'handoff': {'learnings': [], 'covered': ['Revenue read'], 'open_questions': []}}),
+                    ('workspace_action', {'request': {'action': 'finish_research_round', 'gaps': [], 'summary': 'Direct research finished'}}),
+                    ('workspace_action', {'request': {'action': 'reconciliation_save', 'reconciliation': {
+                        'status': 'not_applicable', 'examined_claim_ids': [], 'unexamined_claim_ids': [],
+                        'relations': [], 'open_questions': [], 'coverage_notes': 'No source claims in fixture'}}}),
+                    ('write_report', {'instructions': 'State the supplied revenue'}),
+                    ('finish_task', {})]
+            super().next(sid)
+
+    store, source, run, job = setup(tmp_path)
+    engine = DirectEngine(store, run, source)
+    harness = NativeHarness(store, engine)
+    worker = Worker(store)
+    worker.runtime = InteractiveRuntime(store, backends={'briefloop-native': harness})
+    try:
+        result = worker.generate(job, score=False)
+        brief = store.one('briefs', result['version_id'])
+        assert '1200' in brief['markdown']
+        assert sorted(session['role'] for session in engine.sessions.values()) == ['analyst', 'orchestrator']
+        research = json.loads((worker.folder(job) / 'research.json').read_text())
+        assert research['sources'] == []
+    finally:
+        harness.close()

{briefloop-0.31.6 → briefloop-0.31.7}/tests/test_native_scout.py

@@ -316,3 +316,55 @@ def test_source_regex_timeout_reaps_worker_and_next_search_works(tmp_path, monke
     assert len(processes) == 1 and processes[0].poll() is not None
     following = run_tool(store, config, 'source_grep', {'source_id': sid, 'pattern': '收入(USD)'})
     assert following['ok'] and f'{sid} 2 (start_char=0)' in following['content'][0]['text']
+
+
+def test_search_argument_failure_recovers_without_spending_provider_budget(tmp_path, monkeypatch):
+    from briefloop import duckduckgo, research_budget
+    store, run_id, _ = _run(tmp_path, allow_web=True, policy={'primary_provider': 'duckduckgo'})
+    config = _config(store, run_id, 'argument-recovery')
+    calls = []
+
+    class Provider:
+        validate_search = staticmethod(duckduckgo.validate_search)
+        search_note = staticmethod(lambda: 'discovery only')
+        rows = staticmethod(lambda parsed: [{'url': 'https://example.test/report', 'title': 'Report', 'content': 'snippet'}])
+
+        @staticmethod
+        def call_search(query, parameters, **kwargs):
+            calls.append(query)
+            if query == 'upstream failure':
+                raise websearch.marked('synthetic upstream failure', 'network')
+            return {}, b'{}', 'request', None
+
+    monkeypatch.setattr(websearch, 'provider_module', lambda provider: Provider)
+    arguments = {'provider': 'duckduckgo', 'query': 'orders', 'purpose': 'primary', 'reason': 'find orders'}
+    for invalid in ({'gap_id': 'week-news'}, {'max_results': 99}):
+        result = run_tool(store, config, 'web_search', {**arguments, **invalid})
+        body = json.loads(result['content'][0]['text'])
+        assert body['status'] == 'failed' and body['failure_kind'] == 'invalid_arguments'
+        assert body['recoverable'] is True and body['recovery_action'] == 'correct_arguments'
+        assert calls == []
+        assert research_budget.snapshot(store, run_id)['used']['search_requests'] == 0
+
+    recovered = json.loads(run_tool(store, config, 'web_search', arguments)['content'][0]['text'])
+    assert recovered['status'] == 'ok' and len(recovered['results']) == 1
+    assert calls == ['orders']
+    upstream_result = run_tool(store, config, 'web_search', {**arguments, 'query': 'upstream failure'})
+    upstream = json.loads(upstream_result['content'][0]['text'])
+    assert upstream['failure_kind'] == 'network'
+    assert 'recovery_action' not in upstream
+    assert research_budget.snapshot(store, run_id)['used']['search_requests'] == 2
+
+
+def test_source_read_zero_offset_preserves_requested_multiline_window(tmp_path):
+    store, run_id, sid = _run(tmp_path)
+    config = _config(store, run_id, 'zero-offset-read')
+    arguments = {'source_id': sid, 'start_line': 1, 'end_line': 3}
+    ordinary = run_tool(store, config, 'source_read', arguments)
+    zero = run_tool(store, config, 'source_read', {**arguments, 'start_char': 0})
+    assert zero == ordinary
+    assert '2: Revenue rose to USD 12 million' in zero['content'][0]['text']
+    assert '3: up from USD 10 million.' in zero['content'][0]['text']
+    assert not run_tool(store, config, 'source_read', {**arguments, 'start_char': 1})['ok']
+    single = run_tool(store, config, 'source_read', {**arguments, 'end_line': 1, 'start_char': 1})
+    assert single['ok'] and '1: eader' in single['content'][0]['text']