Funkierry · Oct 5, 2026 · Oct 5, 2026
diff --git a/.github/workflows/jobscout-deploy.yml b/.github/workflows/jobscout-deploy.yml
@@ -12,13 +12,15 @@ on:
       - 'jobscout-web/**'
       - 'deploy/jobscout/**'
       - 'docker/**'
+      - 'skills/public/**'
+      - '.dockerignore'
       - '.github/workflows/jobscout-deploy.yml'
   push:
     branches: [main]
     paths:
       - '.github/workflows/jobscout-deploy.yml'
       - 'deploy/jobscout/**'
-      - 'docker/docker-compose.jobscout.yaml'
+      - 'docker/docker-compose.jobscout*.yaml'
       - 'docker/nginx/jobscout.conf'
       - 'backend/Dockerfile'
       - 'backend/app/**'
@@ -28,6 +30,7 @@ on:
       - 'scripts/jobscout*.py'
       - 'jobscout-web/**'
       - '.dockerignore'
+      - 'skills/public/**'
 permissions:
   contents: read
 concurrency:
@@ -36,7 +39,10 @@ concurrency:
 jobs:
   container-smoke:
     runs-on: ubuntu-latest
-    timeout-minutes: 40
+    timeout-minutes: 60
+    permissions:
+      contents: read
+      packages: write
     steps:
       - uses: actions/checkout@v4
       - uses: actions/setup-python@v5
@@ -94,9 +100,73 @@ jobs:
           assert status['needs_setup'] is False
           print('Administrator restored into a fresh data volume')
           PY
+      - name: Package the tested application as commit-tagged images
+        env:
+          REPOSITORY: ${{ github.repository }}
+          COMMIT: ${{ github.sha }}
+        run: |
+          python scripts/jobscout.py down
+          prefix="ghcr.io/${REPOSITORY,,}"
+          tag="sha-${COMMIT}"
+          echo "JOBSCOUT_RELEASE_PREFIX=${prefix}" >> "$GITHUB_ENV"
+          echo "JOBSCOUT_RELEASE_TAG=${tag}" >> "$GITHUB_ENV"
+          docker build -f deploy/jobscout/gateway.Dockerfile --label "org.opencontainers.image.source=https://github.com/${REPOSITORY}" --label "org.opencontainers.image.revision=${COMMIT}" -t "${prefix}-gateway:${tag}" .
+          docker build -f deploy/jobscout/web.Dockerfile --label "org.opencontainers.image.source=https://github.com/${REPOSITORY}" --label "org.opencontainers.image.revision=${COMMIT}" -t "${prefix}-web:${tag}" .
+          docker tag jobscout-frontend:latest "${prefix}-capabilities:${tag}"
+          python - <<'PY'
+          import os
+          from pathlib import Path
+          p = Path('.jobscout/.env')
+          p.write_text(p.read_text() + '\nJOBSCOUT_DEPLOY_MODE=images\nJOBSCOUT_IMAGE_PREFIX=ghcr.io/' + os.environ['REPOSITORY'].lower() + '\nJOBSCOUT_IMAGE_TAG=sha-' + os.environ['COMMIT'] + '\nJOBSCOUT_DATA_VOLUME=jobscout-image-data\n')
+          PY
+      - name: Start packaged images and verify no source mounts
+        run: |
+          python scripts/jobscout.py up --no-pull
+          python scripts/jobscout_smoke.py
+          python - <<'PY'
+          import importlib.util, json, subprocess, urllib.request
+          from pathlib import Path
+          spec = importlib.util.spec_from_file_location('deploy', 'scripts/jobscout.py')
+          deploy = importlib.util.module_from_spec(spec)
+          spec.loader.exec_module(deploy)
+          directory = Path('.jobscout').resolve()
+          config = json.loads(deploy.compose(directory, ['config', '--format', 'json'], check=True, capture_output=True, text=True).stdout)
+          services = config['services']
+          for service in ('gateway', 'frontend', 'nginx'):
+              assert 'build' not in services[service]
+          assert {v['target'] for v in services['gateway']['volumes']} == {'/config/config.yaml', '/config/extensions_config.json', '/data'}
+          assert {v['target'] for v in services['nginx']['volumes']} == {'/usr/share/nginx/jobscout/runtime-config.js'}
+          deploy.compose(directory, ['exec', '-T', 'gateway', '/app/backend/.venv/bin/python', '-c', "from pathlib import Path; assert list(Path('/app/skills/public').glob('*/SKILL.md'))"], check=True)
+          for name in ('core', 'api-client', 'tracker-state'):
+              response = urllib.request.urlopen('http://127.0.0.1:2026/' + name + '.js')
+              assert response.status == 200 and 'javascript' in response.headers['Content-Type']
+          print('Packaged assets and public skills ready; no application source bind mounts')
+          PY
+      - name: Sign in to GitHub Container Registry
+        if: github.event_name != 'pull_request' && github.ref == 'refs/heads/main'
+        uses: docker/login-action@v3
+        with:
+          registry: ghcr.io
+          username: ${{ github.actor }}
+          password: ${{ secrets.GITHUB_TOKEN }}
+      - name: Publish tested commit images
+        if: github.event_name != 'pull_request' && github.ref == 'refs/heads/main'
+        run: |
+          for component in gateway capabilities web; do
+            docker push "${JOBSCOUT_RELEASE_PREFIX}-${component}:${JOBSCOUT_RELEASE_TAG}"
+          done
+          echo "Published image tag: \`${JOBSCOUT_RELEASE_TAG}\` (linux/amd64)" >> "$GITHUB_STEP_SUMMARY"
       - name: Service diagnostics
         if: failure()
-        run: docker compose --project-name jobscout --env-file .jobscout/.env -f docker/docker-compose.jobscout.yaml logs --tail 80
+        run: |
+          python - <<'PY'
+          import importlib.util
+          from pathlib import Path
+          spec = importlib.util.spec_from_file_location('deploy', 'scripts/jobscout.py')
+          deploy = importlib.util.module_from_spec(spec)
+          spec.loader.exec_module(deploy)
+          deploy.compose(Path('.jobscout').resolve(), ['logs', '--tail', '80'])
+          PY
       - name: Stop disposable services
         if: always()
         run: python scripts/jobscout.py down
diff --git a/README.md b/README.md
@@ -73,6 +73,8 @@ python scripts/jobscout.py up
 
 打开 **http://localhost:2026**，创建管理员账号即可进入工作台。首次构建会下载依赖并编译上游能力中心；命令会等待健康检查通过后才报告启动成功。
 
+服务器也可以直接拉取 **按完整提交号标记的预构建镜像**，省去现场安装依赖和编译。在 `.jobscout/.env` 设置 `JOBSCOUT_DEPLOY_MODE=images` 与已发布的 `JOBSCOUT_IMAGE_TAG=sha-完整提交号`，再运行同一个 `up` 命令。镜像模式要求 Compose 2.24.4+，当前发布 `linux/amd64`；具体版本获取、访问权限和升级步骤见[镜像部署](docs/jobscout/deployment.md#使用预构建镜像)。
+
 这套入口会为你准备：
 
 - **独立配置目录：** `.jobscout/`，初始化不覆盖原来根目录的本地开发配置。
@@ -193,6 +195,8 @@ Linux/macOS 将 Python 路径替换为 `backend/.venv/bin/python`，或在 backe
 
 [部署 CI](.github/workflows/jobscout-deploy.yml) 在临时 Linux 实例验证镜像构建、首次初始化、同域认证、持久化与 Chromium 启动；它不代表真实招聘网站和模型服务已经验收。前端测试还覆盖注册策略、连接重试、桌面/移动端、任务并发、取消及历史记录操作。
 
+投递工作台按 **每页 50 条**加载，支持跨页环节筛选；新增记录不会把后续页面的游标挤乱。顶部总数与状态统计仍覆盖全部投递，CSV 导出与批量检查也保持完整范围。请求处理、分页状态、报告渲染与证据展示已拆为独立模块，页面仍无需构建即可运行。实现与性能边界见[可靠性记录](docs/jobscout/reliability.md)。
+
 ## 继续了解
 
 | 想了解什么 | 文档 |

diff --git a/backend/app/application_tracker/AGENTS.md b/backend/app/application_tracker/AGENTS.md
@@ -1,5 +1,10 @@
 # Application tracker
 
+`pagination.py` returns owner-scoped cursor pages: only page rows load full evidence;
+global summaries scan compact state. Reuse MailStore resolution for mail conflicts
+and manual stages. Keep the legacy full list for exports/batches; no count claims
+based on the current page. Cursor mutation and page/full-list parity tests are offline.
+
 Refresh writes compare revision, identity and checked_at inside BEGIN IMMEDIATE; preserve notification/history atomicity. Manual, batch and scheduled checks share the per-loop refresh lock. Connections close on context exit. Gateway caches store initialization off-loop. CSV export/import share versioned text encoding and bilingual column mapping; never import status history from a CSV.
 
 Stage 7: `scheduling.py` owns owner-scoped settings, atomic per-day reservations and transactional notices. `scheduled_graph.py` accepts only the gated, trusted native scheduler context; never assemble the lead agent here. `allow_model_fallback=False` must avoid model initialization on every tracker branch. Cancellation/failure consumes reserved quota. Portal notices use the check transaction; mail notices share source resolution and the mail transaction. See `docs/jobscout/scheduled-refresh.md`.

diff --git a/backend/app/application_tracker/email/store.py b/backend/app/application_tracker/email/store.py
@@ -1,6 +1,7 @@
 """Private mail ledger and a derived view; portal history is never overwritten."""
 
 import json
+from collections import defaultdict
 from datetime import UTC, datetime
 from hashlib import sha256
 
@@ -41,13 +42,13 @@ def save(self, user_id: str, message: MailMessage, event: MailEvent) -> bool:
                 if row is None or row["company"] != event.company or row["role"] != event.role or not all(is_evidence_grounded(source, text) for text in (event.company, event.role)):
                     raise ValueError("Mail application ownership or evidence mismatch")
                 app = self.tracker._application_from_row(row)
-                previous = self._decorate([app], self._events(connection, user_id))[0]
+                previous = self._decorate([app], self._events(connection, user_id, application_ids=[app.id]))[0]
             result = connection.execute(
                 "INSERT OR IGNORE INTO jobscout_mail_events(user_id,application_id,fingerprint,event_json,received_at) VALUES(?,?,?,?,?)",
                 (user_id, event.application_id, fingerprint, event.model_dump_json(), event.received_at.astimezone(UTC).isoformat()),
             )
             if result.rowcount and previous is not None:
-                current = self._decorate([app], self._events(connection, user_id))[0]
+                current = self._decorate([app], self._events(connection, user_id, application_ids=[app.id]))[0]
                 before = (previous.source_summary or {}).get("status", previous.status.value)
                 after = current.source_summary or {}
                 if before != "未知" and after.get("source") == "email" and not after.get("conflict") and after["status"] != before:
@@ -69,26 +70,36 @@ def list_events(self, user_id: str, application_id: int | None = None) -> list[d
     def decorate(self, user_id: str, applications: list):
         # One bounded query per user listing, not per application.
         with self.tracker._connect() as connection:
-            events = self._events(connection, self.tracker._validated_user_id(user_id))
+            events = self._events(connection, self.tracker._validated_user_id(user_id), application_ids=[app.id for app in applications])
         return self._decorate(applications, events)
 
     @staticmethod
-    def _events(connection, user_id):
+    def _events(connection, user_id, *, application_ids=None, compact=False):
+        if application_ids == []:
+            return []
+        # Bind at most one page's IDs; full export keeps the original owner query.
+        bounded_ids = application_ids if application_ids is not None and len(application_ids) <= 100 else None
+        predicate = " AND application_id IN (" + ",".join("?" for _ in bounded_ids) + ")" if bounded_ids else ""
+        projection = "json_object('application_id',application_id,'status',json_extract(event_json,'$.status'),'review_reason',NULL,'received_at',json_extract(event_json,'$.received_at'))" if compact else "event_json"
         rows = connection.execute(
-            """SELECT id,event_json FROM (
-                SELECT id,event_json,DENSE_RANK() OVER (PARTITION BY application_id ORDER BY received_at DESC) AS n
-                FROM jobscout_mail_events WHERE user_id=? AND application_id IS NOT NULL
+            f"""SELECT id,event_json FROM (
+                SELECT id,{projection} AS event_json,DENSE_RANK() OVER (PARTITION BY application_id ORDER BY received_at DESC) AS n
+                FROM jobscout_mail_events WHERE user_id=? AND application_id IS NOT NULL {predicate}
                 AND json_extract(event_json,'$.review_reason') IS NULL AND json_extract(event_json,'$.status') IS NOT NULL
             ) WHERE n <= 2 ORDER BY id DESC""",
-            (user_id,),
+            (user_id, *(bounded_ids or [])),
         ).fetchall()
         return sorted([{"id": row["id"], **json.loads(row["event_json"])} for row in rows], key=lambda event: (datetime.fromisoformat(event["received_at"]), event["id"]), reverse=True)
 
     @staticmethod
     def _decorate(applications, events):
+        by_application = defaultdict(list)
+        for event in events:
+            if event["status"] and not event["review_reason"]:
+                by_application[event["application_id"]].append(event)
         results = []
         for app in applications:
-            candidates = [event for event in events if event["application_id"] == app.id and event["status"] and not event["review_reason"]]
+            candidates = by_application[app.id]
             if not candidates:
                 results.append(app)
                 continue
@@ -108,10 +119,10 @@ def _decorate(applications, events):
                 "source": "email" if selected else "portal",
                 "conflict": conflict,
                 "email_status": status,
-                "evidence": latest["quote"],
+                "evidence": latest.get("quote", ""),
                 "received_at": latest["received_at"],
-                "event_at": latest["event_at"],
-                "time_evidence": latest["time_quote"],
+                "event_at": latest.get("event_at"),
+                "time_evidence": latest.get("time_quote", ""),
                 "event_id": latest["id"],
             }
             results.append(app.model_copy(update={"source_summary": summary}))

diff --git a/backend/app/application_tracker/pagination.py b/backend/app/application_tracker/pagination.py
@@ -0,0 +1,77 @@
+"""Bounded full-row pages with compact global status projections.
+
+Counts scan small status fields, never every application's evidence/notes. Mail
+resolution stays in MailStore so pagination cannot disagree with single reads.
+"""
+
+from collections import Counter
+from datetime import datetime
+
+from pydantic import BaseModel
+
+from .email.store import TERMINAL, MailStore
+from .models import ApplicationStatus, CheckResult
+
+
+class SummaryRow(BaseModel):
+    id: int
+    status: ApplicationStatus
+    stage: str
+    stage_manual: bool
+    confidence: float
+    checked_at: datetime | None
+    check_result: CheckResult | None
+    source_summary: dict | None = None
+
+
+def presentation(row):
+    source = row.source_summary or {}
+    mail = source.get("source") == "email" and not source.get("conflict")
+    status = source["status"] if mail else row.status.value
+    stage = source["status"] if mail and not row.stage_manual else row.stage or status
+    review = bool(source.get("conflict") or status == "未知" or (not mail and ((row.checked_at and row.confidence < 0.7) or (row.check_result and row.check_result is not CheckResult.SUCCESS))))
+    if review:
+        bucket = "review"
+    elif status == "Offer":
+        bucket = "offers"
+    elif status not in TERMINAL:
+        bucket = "active"
+    else:
+        bucket = None
+    return stage, bucket
+
+
+def list_page(store, user_id, *, limit=50, before_id=None, stage=None):
+    user_id = store._validated_user_id(user_id)
+    if not 1 <= limit <= 100 or (before_id is not None and before_id < 1):
+        raise ValueError("Invalid page bounds")
+    with store._connect() as connection:
+        connection.execute("BEGIN")
+        rows = connection.execute("SELECT id,status,stage,stage_manual,confidence,checked_at,check_result FROM applications WHERE user_id=? ORDER BY id DESC", (user_id,)).fetchall()
+        projected = MailStore._decorate([SummaryRow.model_validate(dict(row)) for row in rows], MailStore._events(connection, user_id, compact=True))
+        summary = {"total": len(projected), "active": 0, "review": 0, "offers": 0}
+        stages, matching = Counter(), []
+        for row in projected:
+            label, bucket = presentation(row)
+            stages[label] += 1
+            if bucket:
+                summary[bucket] += 1
+            if stage is None or label == stage:
+                matching.append(row.id)
+        eligible = [row_id for row_id in matching if before_id is None or row_id < before_id]
+        ids = eligible[:limit]
+        items = []
+        if ids:
+            placeholders = ",".join("?" for _ in ids)
+            selected = connection.execute(f"SELECT * FROM applications WHERE user_id=? AND id IN ({placeholders}) ORDER BY id DESC LIMIT ?", (user_id, *ids, limit)).fetchall()
+            items = MailStore._decorate([store._application_from_row(row) for row in selected], MailStore._events(connection, user_id, application_ids=ids))
+    has_more = len(eligible) > limit
+    return {
+        "items": items,
+        "total": summary["total"],
+        "filtered_total": len(matching),
+        "summary": summary,
+        "stages": [{"label": label, "count": count} for label, count in stages.items()],
+        "has_more": has_more,
+        "next_before_id": ids[-1] if has_more else None,
+    }
diff --git a/backend/app/application_tracker/store.py b/backend/app/application_tracker/store.py
@@ -627,6 +627,11 @@ def import_applications(
                     updated += 1
         return ImportSummary(inserted=inserted, updated=updated, total=len(applications))
 
+    def list_applications_page(self, user_id: str, *, limit=50, before_id=None, stage=None) -> dict:
+        from .pagination import list_page
+
+        return list_page(self, user_id, limit=limit, before_id=before_id, stage=stage)
+
     def list_applications(self, user_id: str) -> list[StoredApplication]:
         user_id = self._validated_user_id(user_id)
         with self._connect() as connection:

diff --git a/backend/app/gateway/routers/jobscout.py b/backend/app/gateway/routers/jobscout.py
@@ -9,7 +9,7 @@
 from threading import Lock
 
 from anyio import CancelScope
-from fastapi import APIRouter, Depends, File, HTTPException, Request, UploadFile
+from fastapi import APIRouter, Depends, File, HTTPException, Query, Request, UploadFile
 from fastapi.encoders import jsonable_encoder
 from fastapi.responses import Response, StreamingResponse
 from pydantic import BaseModel, Field, field_validator, model_validator
@@ -515,6 +515,13 @@ async def list_tracker_applications(request: Request) -> list[StoredApplication]
     return await asyncio.to_thread((await _get_tracker_store()).list_applications, user_id)
 
 
+@router.get("/tracker/applications/page")
+@require_permission("runs", "read")
+async def list_tracker_applications_page(request: Request, limit: int = Query(50, ge=1, le=100), before_id: int | None = Query(None, ge=1), stage: str | None = Query(None, max_length=60)) -> dict:
+    del request
+    return await asyncio.to_thread((await _get_tracker_store()).list_applications_page, get_effective_user_id(), limit=limit, before_id=before_id, stage=stage)
+
+
 @router.post("/tracker/applications", response_model=StoredApplication)
 @require_permission("runs", "create")
 async def create_tracker_application(body: TrackerApplicationCreate, request: Request) -> StoredApplication: