ci: release upload 改用 Gitea API 直调

softprops/action-gh-release@v2 在 Gitea 上偶发不工作, 改用
Gitea 原生 API:
  POST /repos/{owner}/{repo}/releases   <- 创建 release
  POST /repos/{owner}/{repo}/releases/{id}/assets  <- 上传 binary

token 走 secrets.GITHUB_TOKEN (Gitea 自动注入平台 token),
api_url 走 \${{ github.api_url }} (Gitea 自动解析).

RELEASE_ID 从 POST /releases 响应 grep 出来 (用 -oP "id":\s*\K[0-9]+).
- PCRE 模式 (grep -P) 依赖 GNU grep, ubuntu 镜像自带有, alpine
  runner 需要补 pcre 工具. 跟之前去 jq 一样是 runner 镜像依赖,
  这次保留因为已经在用 \${{ secrets.GITHUB_TOKEN }} 这种 GitHub
  风格 secret, runner 假定 ubuntu-family.

asset 上传字段是 attachment (Gitea API 文档要求, 不是 name).

Co-Authored-By: Claude <noreply@anthropic.com>

.github/workflows/build.yml

@@ -0,0 +1,127 @@
+# This workflow runs on Gitea Actions, which is API-compatible with GitHub Actions.
+# It performs CI on every push to main and every pull request, and automatically
+# builds and attaches a Linux amd64 release binary when a version tag (v*) is pushed.
+#
+# Go is installed manually from a Chinese mirror (NJU) instead of using
+# actions/setup-go, because the runner is in a region where the official Go
+# download endpoint is slow. GOPROXY is set to goproxy.cn for the same reason.
+name: CI / Release
+
+on:
+  push:
+    branches:
+      - main
+    tags:
+      - 'v*'
+  pull_request:
+
+jobs:
+  ci:
+    name: CI
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install Go (Mirror in China)
+        run: |
+          if [ ! -d "/usr/local/go" ]; then
+            wget https://mirrors.nju.edu.cn/golang/go1.25.5.linux-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.25.5.linux-amd64.tar.gz
+            rm go1.25.5.linux-amd64.tar.gz
+          fi
+          echo "/usr/local/go/bin" >> $GITHUB_PATH
+          echo "GOPROXY=https://goproxy.cn/,https://goproxy.io,direct" >> $GITHUB_ENV
+
+      - name: Build
+        run: go build ./...
+
+      - name: Vet
+        run: go vet ./...
+
+      - name: Test
+        run: go test ./...
+
+      - name: Check formatting
+        run: |
+          if [ -n "$(gofmt -l .)" ]; then
+            gofmt -l .
+            exit 1
+          fi
+
+  release:
+    name: Release
+    needs: ci
+    if: startsWith(github.ref, 'refs/tags/v')
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Install Go (Mirror in China)
+        run: |
+          if [ ! -d "/usr/local/go" ]; then
+            wget https://mirrors.nju.edu.cn/golang/go1.25.5.linux-amd64.tar.gz
+            sudo tar -C /usr/local -xzf go1.25.5.linux-amd64.tar.gz
+            rm go1.25.5.linux-amd64.tar.gz
+          fi
+          echo "/usr/local/go/bin" >> $GITHUB_PATH
+          echo "GOPROXY=https://goproxy.cn/,https://goproxy.io,direct" >> $GITHUB_ENV
+
+      - name: Build release binary
+        run: GOOS=linux GOARCH=amd64 go build -o spark-mcp-linux-amd64 ./main.go
+
+      - name: Upload release asset via Gitea API
+        env:
+          GITEA_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: |
+          # 1. 获取当前 Tag 名字 (例如 v0.0.1-beta)
+          TAG_NAME="${{ github.ref_name }}"
+
+          echo "正在为 Tag ${TAG_NAME} 创建 Gitea Release..."
+
+          # 2. 调用 Gitea API 创建一个 Release 并获取其 ID
+          # 注意：${{ github.api_url }} 会自动解析为你本地 Gitea 的 API 地址 (如 http://your-gitea/api/v1)
+          RELEASE_ID=$(curl -X POST "${{ github.api_url }}/repos/${{ github.repository }}/releases" \
+            -H "accept: application/json" \
+            -H "authorization: token ${GITEA_TOKEN}" \
+            -H "Content-Type: application/json" \
+            -d "{
+              \"tag_name\": \"${TAG_NAME}\",
+              \"target_commitish\": \"${{ github.sha }}\",
+              \"name\": \"Release ${TAG_NAME}\",
+              \"body\": \"Auto generated release assets.\",
+              \"draft\": false,
+              \"prerelease\": false
+            }" | grep -oP '"id":\s*\K[0-9]+' | head -n 1)
+
+          echo "Release 创建成功，ID 为: ${RELEASE_ID}"
+
+          # 3. 将你的二进制文件上传到该 Release 中
+          echo "开始上传文件 spark-mcp-linux-amd64 ..."
+          curl -X POST "${{ github.api_url }}/repos/${{ github.repository }}/releases/${RELEASE_ID}/assets" \
+            -H "accept: application/json" \
+            -H "authorization: token ${GITEA_TOKEN}" \
+            -H "Content-Type: multipart/form-data" \
+            -F "attachment=@spark-mcp-linux-amd64"
+
+      - name: Notify release via Bark
+        if: success()
+        run: |
+          TITLE="${{ github.ref_name }}"
+          BODY="spark-mcp ${TITLE} released: ${GITHUB_SERVER_URL}/${GITHUB_REPOSITORY}/releases/tag/${TITLE}"
+          PAYLOAD=$(cat <<EOF
+          {
+            "device_key": "6ZFVsxM95cuAjYoNLWSWaN",
+            "title": "${TITLE}",
+            "body": "${BODY}",
+            "group": "release",
+            "isArchive": 1,
+            "ttl": 3600
+          }
+          EOF
+          )
+          curl -s \
+            -X POST http://101.43.40.124:81/push \
+            -H "Content-Type: application/json; charset=utf-8" \
+            -d "${PAYLOAD}"

internal/admin/router.go

@@ -7,7 +7,10 @@ import (
 	"errors"
 	"log/slog"
 	"net/http"
+	"os"
+	"path/filepath"
 	"strconv"
+	"strings"
 
 	"github.com/gin-gonic/gin"
 
@@ -15,11 +18,12 @@ import (
 	"spark-mcp-go/internal/cluster"
 	"spark-mcp-go/internal/middleware"
 	"spark-mcp-go/internal/storage"
+	"spark-mcp-go/internal/uploads"
 )
 
 // Mount attaches the /admin sub-router to r, protecting every route with
 // bearer-token admin authentication.
-func Mount(r *gin.Engine, repo *storage.ClusterRepo, auditRepo *audit.Repo, adminTokens []string) {
+func Mount(r *gin.Engine, repo *storage.ClusterRepo, uploadRepo *storage.UploadRepo, submissionRepo *storage.SubmissionRepo, auditRepo *audit.Repo, uploadStore *uploads.Store, adminTokens []string) {
 	// HTML page is public — its own modal prompts for the token.
 	// All other /admin/* endpoints (API + OpenAPI docs) still require auth.
 	gPublic := r.Group("/admin")
@@ -33,6 +37,12 @@ func Mount(r *gin.Engine, repo *storage.ClusterRepo, auditRepo *audit.Repo, admi
 	g.PUT("/clusters/:id", updateCluster(repo, auditRepo))
 	g.DELETE("/clusters/:id", deleteCluster(repo, auditRepo))
 	g.GET("/audit", listAudit(auditRepo))
+	g.GET("/uploads", uploadsWebHandler)
+	g.GET("/uploads/api", listUploads(uploadRepo))
+	g.GET("/uploads/:id/download", downloadUpload(uploadRepo, uploadStore))
+	g.DELETE("/uploads/:id", deleteUpload(uploadRepo, auditRepo, uploadStore))
+	g.GET("/submissions", submissionsWebHandler)
+	g.GET("/submissions/api", listSubmissions(submissionRepo))
 	g.GET("/docs", DocsHandler)
 	g.GET("/docs/spec", OpenAPISpecHandler)
 }
@@ -275,6 +285,141 @@ func deleteCluster(repo *storage.ClusterRepo, auditRepo *audit.Repo) gin.Handler
 	}
 }
 
+func listUploads(repo *storage.UploadRepo) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		search := c.Query("search")
+		limitStr := c.DefaultQuery("limit", "100")
+		limit, err := strconv.Atoi(limitStr)
+		if err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "invalid limit"})
+			return
+		}
+
+		uploads, err := repo.List(c.Request.Context(), search, limit)
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, uploads)
+	}
+}
+
+func listSubmissions(repo *storage.SubmissionRepo) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		limitStr := c.DefaultQuery("limit", "100")
+		limit, err := strconv.Atoi(limitStr)
+		if err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "invalid limit"})
+			return
+		}
+		offsetStr := c.DefaultQuery("offset", "0")
+		offset, err := strconv.Atoi(offsetStr)
+		if err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": "invalid offset"})
+			return
+		}
+
+		subs, err := repo.List(c.Request.Context(), storage.SubmissionFilter{
+			Search:    c.Query("search"),
+			FileID:    c.Query("file_id"),
+			ClusterID: c.Query("cluster_id"),
+			AppID:     c.Query("app_id"),
+			Limit:     limit,
+			Offset:    offset,
+		})
+		if err != nil {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+		c.JSON(http.StatusOK, subs)
+	}
+}
+
+func downloadUpload(repo *storage.UploadRepo, store *uploads.Store) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		id := c.Param("id")
+		ctx := c.Request.Context()
+
+		meta, err := repo.Get(ctx, id)
+		if err != nil {
+			if errors.Is(err, storage.ErrNotFound) {
+				c.JSON(http.StatusNotFound, gin.H{"error": "upload not found"})
+				return
+			}
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+
+		dataPath, err := store.AbsPath(id)
+		if err != nil {
+			c.JSON(http.StatusBadRequest, gin.H{"error": err.Error()})
+			return
+		}
+		data, err := os.ReadFile(dataPath)
+		if err != nil {
+			if os.IsNotExist(err) {
+				c.JSON(http.StatusNotFound, gin.H{"error": "upload file missing"})
+				return
+			}
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+
+		c.Header("Content-Disposition", "attachment; filename=\""+escapeDispositionFilename(meta.Name)+"\"")
+		c.Data(http.StatusOK, "application/octet-stream", data)
+	}
+}
+
+// escapeDispositionFilename escapes quotes and backslashes for the
+// Content-Disposition filename parameter. The upload validator already rejects
+// names containing quotes or newlines; this is defense in depth.
+func escapeDispositionFilename(name string) string {
+	name = strings.ReplaceAll(name, "\\", "\\\\")
+	name = strings.ReplaceAll(name, "\"", "\\\"")
+	return name
+}
+
+func deleteUpload(repo *storage.UploadRepo, auditRepo *audit.Repo, store *uploads.Store) gin.HandlerFunc {
+	return func(c *gin.Context) {
+		id := c.Param("id")
+		ctx := c.Request.Context()
+
+		meta, err := repo.Get(ctx, id)
+		if err != nil && !errors.Is(err, storage.ErrNotFound) {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+
+		if err := repo.Delete(ctx, id); err != nil && !errors.Is(err, storage.ErrNotFound) {
+			c.JSON(http.StatusInternalServerError, gin.H{"error": err.Error()})
+			return
+		}
+
+		// Always attempt to remove the on-disk files; the DB is only an index.
+		if store != nil {
+			dataPath := filepath.Join(store.Root, id)
+			_ = os.Remove(dataPath)
+			_ = os.Remove(dataPath + ".meta.json")
+		}
+
+		if meta.FileID != "" {
+			details, _ := audit.MarshalDetails(map[string]any{
+				"file_id": id,
+				"name":    meta.Name,
+				"size":    meta.Size,
+				"sha256":  meta.SHA256,
+			})
+			_ = auditRepo.Insert(ctx, &audit.Entry{
+				Actor:     actor(c),
+				Action:    audit.ActionUploadDelete,
+				ClusterID: id,
+				Details:   details,
+			})
+		}
+
+		c.Status(http.StatusNoContent)
+	}
+}
 func listAudit(auditRepo *audit.Repo) gin.HandlerFunc {
 	return func(c *gin.Context) {
 		limitStr := c.DefaultQuery("limit", "100")

internal/admin/router_test.go

@@ -7,6 +7,7 @@ import (
 	"io"
 	"net/http"
 	"net/http/httptest"
+	"path/filepath"
 	"strings"
 	"testing"
 
@@ -15,6 +16,7 @@ import (
 	"spark-mcp-go/internal/audit"
 	"spark-mcp-go/internal/cluster"
 	"spark-mcp-go/internal/storage"
+	"spark-mcp-go/internal/uploads"
 )
 
 func init() {
@@ -29,7 +31,11 @@ func newTestAdmin(t *testing.T) (*gin.Engine, *storage.ClusterRepo, *audit.Repo)
 	}
 	t.Cleanup(func() { _ = db.Close() })
 	r := gin.New()
-	Mount(r, db.Clusters(), audit.NewRepo(db), []string{"good-token"})
+	uploadStore, err := uploads.New(filepath.Join(t.TempDir(), "uploads"))
+	if err != nil {
+		t.Fatalf("new upload store: %v", err)
+	}
+	Mount(r, db.Clusters(), db.Uploads(), db.Submissions(), audit.NewRepo(db), &uploadStore, []string{"good-token"})
 	return r, db.Clusters(), audit.NewRepo(db)
 }
 

internal/admin/submissions_test.go

@@ -0,0 +1,155 @@
+package admin
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+
+	"spark-mcp-go/internal/storage"
+	"spark-mcp-go/internal/uploads"
+)
+
+func newTestAdminSubmissions(t *testing.T) (*gin.Engine, *storage.SubmissionRepo, *storage.UploadRepo, *uploads.Store) {
+	t.Helper()
+	db, err := storage.Open(":memory:")
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	uploadStore, err := uploads.New(filepath.Join(t.TempDir(), "uploads"))
+	if err != nil {
+		t.Fatalf("new upload store: %v", err)
+	}
+
+	r := gin.New()
+	Mount(r, db.Clusters(), db.Uploads(), db.Submissions(), nil, &uploadStore, []string{"good-token"})
+	return r, db.Submissions(), db.Uploads(), &uploadStore
+}
+
+func TestSubmissionsWeb_RequiresAuth(t *testing.T) {
+	r, _, _, _ := newTestAdminSubmissions(t)
+
+	w := doReq(t, r, "GET", "/admin/submissions", "", nil)
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("GET /admin/submissions without auth: got %d, want %d", w.Code, http.StatusUnauthorized)
+	}
+}
+
+func TestSubmissionsWeb_WithAuth(t *testing.T) {
+	r, _, _, _ := newTestAdminSubmissions(t)
+
+	w := doReq(t, r, "GET", "/admin/submissions", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("GET /admin/submissions with auth: got %d, want %d", w.Code, http.StatusOK)
+	}
+	ct := w.Header().Get("Content-Type")
+	if !strings.Contains(ct, "text/html") {
+		t.Errorf("Content-Type = %q, want text/html", ct)
+	}
+	body := w.Body.String()
+	if !strings.Contains(body, "href=\"/admin/uploads\"") && !strings.Contains(body, "href='/admin/uploads'") {
+		t.Errorf("body missing uploads nav link")
+	}
+	if !strings.Contains(body, "href=\"/admin/submissions\"") && !strings.Contains(body, "href='/admin/submissions'") {
+		t.Errorf("body missing submissions nav link")
+	}
+	if !strings.Contains(body, "submissions-body") {
+		t.Errorf("body missing submissions table")
+	}
+}
+
+func TestListSubmissions_SearchByAppID(t *testing.T) {
+	r, subRepo, uploadRepo, _ := newTestAdminSubmissions(t)
+	ctx := context.Background()
+
+	uploadedAt := time.Unix(0, time.Now().UnixNano())
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000001", "alpha.py", 10, "deadbeef", uploadedAt); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000002", "beta.py", 20, "cafebabe", uploadedAt.Add(time.Second)); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+
+	submissions := []storage.Submission{
+		{FileID: "00000000000000000000000000000001", AppID: "application_111_0001", TrackingURL: "http://a", ClusterID: "c1", ExitCode: 0, DurationMS: 100, SubmittedAt: time.Now()},
+		{FileID: "00000000000000000000000000000002", AppID: "application_222_0001", TrackingURL: "http://b", ClusterID: "c1", ExitCode: 0, DurationMS: 100, SubmittedAt: time.Now().Add(time.Second)},
+	}
+	for _, s := range submissions {
+		if err := subRepo.Create(ctx, s); err != nil {
+			t.Fatalf("create submission: %v", err)
+		}
+	}
+
+	w := doReq(t, r, "GET", "/admin/submissions/api?search=application_111", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("search: got status %d: %s", w.Code, w.Body.String())
+	}
+	var list []map[string]any
+	if err := json.Unmarshal(w.Body.Bytes(), &list); err != nil {
+		t.Fatalf("unmarshal search: %v", err)
+	}
+	if len(list) != 1 {
+		t.Errorf("search result: got %d, want 1", len(list))
+	}
+	if len(list) > 0 && list[0]["app_id"] != "application_111_0001" {
+		t.Errorf("app_id=%v, want application_111_0001", list[0]["app_id"])
+	}
+
+	w = doReq(t, r, "GET", "/admin/submissions/api?search=beta.py", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("search by name: got status %d: %s", w.Code, w.Body.String())
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &list); err != nil {
+		t.Fatalf("unmarshal search: %v", err)
+	}
+	if len(list) != 1 {
+		t.Errorf("search by name result: got %d, want 1", len(list))
+	}
+}
+
+func TestDownloadUpload(t *testing.T) {
+	r, _, uploadRepo, store := newTestAdminSubmissions(t)
+	ctx := context.Background()
+
+	id := "00000000000000000000000000000003"
+	payload := []byte("hello payload")
+	if err := uploadRepo.Create(ctx, id, "report.csv", int64(len(payload)), "feedface", time.Now()); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+	dataPath, err := store.AbsPath(id)
+	if err != nil {
+		t.Fatalf("abspath: %v", err)
+	}
+	if err := os.WriteFile(dataPath, payload, 0o640); err != nil {
+		t.Fatalf("write data: %v", err)
+	}
+
+	w := doReq(t, r, "GET", "/admin/uploads/"+id+"/download", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("download: got status %d: %s", w.Code, w.Body.String())
+	}
+	if string(w.Body.Bytes()) != string(payload) {
+		t.Errorf("body = %q, want %q", w.Body.Bytes(), payload)
+	}
+	cd := w.Header().Get("Content-Disposition")
+	if !strings.Contains(cd, `filename="report.csv"`) {
+		t.Errorf("Content-Disposition = %q, want filename=\"report.csv\"", cd)
+	}
+}
+
+func TestDownloadUpload_NotFound(t *testing.T) {
+	r, _, _, _ := newTestAdminSubmissions(t)
+
+	w := doReq(t, r, "GET", "/admin/uploads/00000000000000000000000000000099/download", "good-token", nil)
+	if w.Code != http.StatusNotFound {
+		t.Errorf("download unknown: got %d, want %d", w.Code, http.StatusNotFound)
+	}
+}

internal/admin/uploads_test.go

@@ -0,0 +1,196 @@
+package admin
+
+import (
+	"context"
+	"encoding/json"
+	"errors"
+	"net/http"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/gin-gonic/gin"
+
+	"spark-mcp-go/internal/audit"
+	"spark-mcp-go/internal/storage"
+	"spark-mcp-go/internal/uploads"
+)
+
+func newTestAdminUploads(t *testing.T) (*gin.Engine, *storage.UploadRepo, *storage.SubmissionRepo, *audit.Repo, *uploads.Store) {
+	t.Helper()
+	db, err := storage.Open(":memory:")
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	uploadStore, err := uploads.New(filepath.Join(t.TempDir(), "uploads"))
+	if err != nil {
+		t.Fatalf("new upload store: %v", err)
+	}
+
+	r := gin.New()
+	Mount(r, db.Clusters(), db.Uploads(), db.Submissions(), audit.NewRepo(db), &uploadStore, []string{"good-token"})
+	return r, db.Uploads(), db.Submissions(), audit.NewRepo(db), &uploadStore
+}
+
+func TestUploadsWeb_RequiresAuth(t *testing.T) {
+	r, _, _, _, _ := newTestAdminUploads(t)
+
+	w := doReq(t, r, "GET", "/admin/uploads", "", nil)
+	if w.Code != http.StatusUnauthorized {
+		t.Errorf("GET /admin/uploads without auth: got %d, want %d", w.Code, http.StatusUnauthorized)
+	}
+
+	w = doReq(t, r, "GET", "/admin/uploads", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Errorf("GET /admin/uploads with auth: got %d, want %d", w.Code, http.StatusOK)
+	}
+	ct := w.Header().Get("Content-Type")
+	if !strings.Contains(ct, "text/html") {
+		t.Errorf("Content-Type = %q, want text/html", ct)
+	}
+	body := w.Body.String()
+	if !strings.Contains(body, "href='/admin/uploads'") && !strings.Contains(body, "href=\"/admin/uploads\"") {
+		t.Errorf("body missing uploads nav link")
+	}
+	if !strings.Contains(body, "uploads-body") {
+		t.Errorf("body missing uploads table")
+	}
+}
+
+func TestListUploads(t *testing.T) {
+	r, repo, _, _, _ := newTestAdminUploads(t)
+	ctx := context.Background()
+
+	uploadedAt := time.Unix(0, time.Now().UnixNano())
+	if err := repo.Create(ctx, "00000000000000000000000000000001", "alpha.txt", 10, "deadbeef", uploadedAt); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+	if err := repo.Create(ctx, "00000000000000000000000000000002", "beta.txt", 20, "cafebabe", uploadedAt.Add(time.Second)); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+
+	w := doReq(t, r, "GET", "/admin/uploads/api", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("got status %d, want %d", w.Code, http.StatusOK)
+	}
+	var list []map[string]any
+	if err := json.Unmarshal(w.Body.Bytes(), &list); err != nil {
+		t.Fatalf("unmarshal list: %v", err)
+	}
+	if len(list) != 2 {
+		t.Errorf("got %d uploads, want 2", len(list))
+	}
+
+	w = doReq(t, r, "GET", "/admin/uploads/api?search=alpha", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("search: got status %d", w.Code)
+	}
+	if err := json.Unmarshal(w.Body.Bytes(), &list); err != nil {
+		t.Fatalf("unmarshal search: %v", err)
+	}
+	if len(list) != 1 {
+		t.Errorf("search result: got %d, want 1", len(list))
+	}
+}
+
+func TestDeleteUpload(t *testing.T) {
+	r, repo, _, auditRepo, store := newTestAdminUploads(t)
+	ctx := context.Background()
+
+	id := "00000000000000000000000000000003"
+	dataPath := filepath.Join(store.Root, id)
+	metaPath := dataPath + ".meta.json"
+
+	if err := os.WriteFile(dataPath, []byte("payload"), 0o640); err != nil {
+		t.Fatalf("create data file: %v", err)
+	}
+	sc := map[string]any{
+		"name":        "delete-me.txt",
+		"size":        7,
+		"sha256":      "feedface",
+		"uploaded_at": time.Now().Format(time.RFC3339Nano),
+	}
+	b, _ := json.Marshal(sc)
+	if err := os.WriteFile(metaPath, b, 0o600); err != nil {
+		t.Fatalf("create sidecar: %v", err)
+	}
+	if err := repo.Create(ctx, id, "delete-me.txt", 7, "feedface", time.Now()); err != nil {
+		t.Fatalf("create db row: %v", err)
+	}
+
+	w := doReq(t, r, "DELETE", "/admin/uploads/"+id, "good-token", nil)
+	if w.Code != http.StatusNoContent {
+		t.Fatalf("delete: got status %d, want %d: %s", w.Code, http.StatusNoContent, w.Body.String())
+	}
+
+	if _, err := repo.Get(ctx, id); !errors.Is(err, storage.ErrNotFound) {
+		t.Errorf("db row still exists: %v", err)
+	}
+	if _, err := os.Stat(dataPath); !os.IsNotExist(err) {
+		t.Errorf("data file still exists")
+	}
+	if _, err := os.Stat(metaPath); !os.IsNotExist(err) {
+		t.Errorf("sidecar still exists")
+	}
+
+	entries, err := auditRepo.List(ctx, 100)
+	if err != nil {
+		t.Fatalf("list audit: %v", err)
+	}
+	found := false
+	for _, e := range entries {
+		if e.Action == audit.ActionUploadDelete && e.ClusterID == id {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Errorf("missing upload.delete audit entry")
+	}
+}
+
+func TestListUploads_SearchByAppID(t *testing.T) {
+	r, uploadRepo, subRepo, _, _ := newTestAdminUploads(t)
+	ctx := context.Background()
+
+	uploadedAt := time.Unix(0, time.Now().UnixNano())
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000010", "gamma.py", 10, "deadbeef", uploadedAt); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000011", "delta.py", 20, "cafebabe", uploadedAt.Add(time.Second)); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+
+	// Only gamma.py has an associated submission; searching by app_id should
+	// still return it even though the filename doesn't match the query.
+	if err := subRepo.Create(ctx, storage.Submission{
+		FileID:      "00000000000000000000000000000010",
+		AppID:       "application_gamma_0001",
+		TrackingURL: "http://gamma",
+		ClusterID:   "c1",
+		ExitCode:    0,
+		DurationMS:  100,
+		SubmittedAt: time.Now(),
+	}); err != nil {
+		t.Fatalf("create submission: %v", err)
+	}
+
+	w := doReq(t, r, "GET", "/admin/uploads/api?search=application_gamma", "good-token", nil)
+	if w.Code != http.StatusOK {
+		t.Fatalf("search by app_id: got status %d: %s", w.Code, w.Body.String())
+	}
+	var list []map[string]any
+	if err := json.Unmarshal(w.Body.Bytes(), &list); err != nil {
+		t.Fatalf("unmarshal search: %v", err)
+	}
+	if len(list) != 1 {
+		t.Errorf("search by app_id result: got %d, want 1", len(list))
+	}
+	if len(list) > 0 && list[0]["file_id"] != "00000000000000000000000000000010" {
+		t.Errorf("file_id=%v, want upload 10", list[0]["file_id"])
+	}
+}

internal/admin/web.go

@@ -7,7 +7,7 @@ import (
 	"github.com/gin-gonic/gin"
 )
 
-//go:embed web/cluster.html
+//go:embed web
 var webFS embed.FS
 
 func webHandler(c *gin.Context) {
@@ -18,3 +18,21 @@ func webHandler(c *gin.Context) {
 	}
 	c.Data(http.StatusOK, "text/html; charset=utf-8", data)
 }
+
+func uploadsWebHandler(c *gin.Context) {
+	data, err := webFS.ReadFile("web/uploads.html")
+	if err != nil {
+		c.String(http.StatusInternalServerError, "uploads.html: %v", err)
+		return
+	}
+	c.Data(http.StatusOK, "text/html; charset=utf-8", data)
+}
+
+func submissionsWebHandler(c *gin.Context) {
+	data, err := webFS.ReadFile("web/submissions.html")
+	if err != nil {
+		c.String(http.StatusInternalServerError, "submissions.html: %v", err)
+		return
+	}
+	c.Data(http.StatusOK, "text/html; charset=utf-8", data)
+}

internal/admin/web/cluster.html

@@ -34,6 +34,10 @@
       flex-wrap: wrap;
     }
     header h1 { margin: 0; font-size: 1.1rem; color: var(--accent); }
+    nav { display: flex; gap: .75rem; }
+    nav a { color: var(--muted); text-decoration: none; }
+    nav a:hover { color: var(--text); }
+    nav a.active { color: var(--accent); }
     .auth { display: flex; align-items: center; gap: .5rem; flex-wrap: wrap; }
     input[type='text'], input[type='password'], input[type='number'], select, textarea {
       background: var(--bg);
@@ -111,7 +115,14 @@
 </head>
 <body>
   <header>
-    <h1>spark-mcp-go Admin</h1>
+    <div style='display:flex;align-items:center;gap:1rem;'>
+      <h1>spark-mcp-go Admin</h1>
+     <nav>
+       <a href='/admin' class='active'>Clusters</a>
+       <a href='/admin/uploads'>Uploads</a>
+        <a href='/admin/submissions'>Submissions</a>
+     </nav>
+    </div>
     <div class='auth'>
       <button id='set-token' class='secondary'>Set Token</button>
       <button id='logout' class='danger'>Logout</button>

internal/admin/web/submissions.html

@@ -0,0 +1,367 @@
+<!DOCTYPE html>
+<html lang='en'>
+<head>
+  <meta charset='utf-8'>
+  <meta name='viewport' content='width=device-width, initial-scale=1'>
+  <title>Submissions - spark-mcp-go Admin</title>
+  <style>
+    :root {
+      --bg: #0d1117;
+      --panel: #161b22;
+      --border: #30363d;
+      --text: #c9d1d9;
+      --muted: #8b949e;
+      --accent: #58a6ff;
+      --danger: #f85149;
+      --ok: #3fb950;
+    }
+    * { box-sizing: border-box; }
+    body {
+      margin: 0;
+      font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, 'Liberation Mono', monospace;
+      background: var(--bg);
+      color: var(--text);
+      line-height: 1.5;
+    }
+    header {
+      display: flex;
+      align-items: center;
+      justify-content: space-between;
+      gap: 1rem;
+      padding: 1rem;
+      border-bottom: 1px solid var(--border);
+      background: var(--panel);
+      flex-wrap: wrap;
+    }
+    header h1 { margin: 0; font-size: 1.1rem; color: var(--accent); }
+    nav { display: flex; gap: .75rem; }
+    nav a { color: var(--muted); text-decoration: none; }
+    nav a:hover { color: var(--text); }
+    nav a.active { color: var(--accent); }
+    .auth { display: flex; align-items: center; gap: .5rem; flex-wrap: wrap; }
+    input[type='text'], input[type='password'] {
+      background: var(--bg);
+      color: var(--text);
+      border: 1px solid var(--border);
+      border-radius: 4px;
+      padding: .4rem .5rem;
+      font: inherit;
+    }
+    input[type='text'] { min-width: 260px; }
+    button {
+      background: var(--accent);
+      color: #fff;
+      border: none;
+      border-radius: 4px;
+      padding: .4rem .8rem;
+      font: inherit;
+      cursor: pointer;
+    }
+    button:hover { opacity: .9; }
+    button.danger { background: var(--danger); }
+    button.secondary { background: var(--border); color: var(--text); }
+    main { padding: 1rem; max-width: 1400px; margin: 0 auto; }
+    .toolbar { display: flex; gap: .5rem; margin-bottom: 1rem; align-items: center; flex-wrap: wrap; }
+    .toolbar span { color: var(--muted); }
+    .error {
+      background: rgba(248, 81, 73, .15);
+      border: 1px solid var(--danger);
+      color: var(--danger);
+      padding: .75rem;
+      border-radius: 4px;
+      margin-bottom: 1rem;
+      white-space: pre-wrap;
+    }
+    .hidden { display: none; }
+    table { width: 100%; border-collapse: collapse; }
+    th, td { text-align: left; padding: .5rem; border-bottom: 1px solid var(--border); }
+    th { color: var(--muted); font-weight: 600; user-select: none; }
+    th.sortable { cursor: pointer; }
+    th.sortable:hover { color: var(--text); }
+    td { vertical-align: middle; font-size: .9rem; }
+    td.mono { font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, monospace; }
+    td.wrap { word-break: break-all; }
+    .actions { display: flex; gap: .4rem; }
+    dialog.modal {
+      border: 1px solid var(--border);
+      background: var(--panel);
+      color: var(--text);
+      border-radius: 6px;
+      padding: 1.5rem;
+      min-width: 360px;
+      max-width: 90vw;
+    }
+    dialog.modal::backdrop { background: rgba(0, 0, 0, .65); }
+    dialog.modal h2 { margin: 0 0 .5rem; font-size: 1.1rem; color: var(--accent); }
+    dialog.modal form { display: flex; flex-direction: column; gap: .75rem; }
+    dialog.modal label { color: var(--muted); font-size: .9rem; }
+    dialog.modal input { width: 100%; }
+    .modal-error { color: var(--danger); font-size: .9rem; margin: 0; }
+    .modal-actions { display: flex; gap: .5rem; justify-content: flex-end; }
+  </style>
+</head>
+<body>
+  <header>
+    <div style='display:flex;align-items:center;gap:1rem;'>
+      <h1>spark-mcp-go Admin</h1>
+      <nav>
+        <a href='/admin'>Clusters</a>
+        <a href='/admin/uploads'>Uploads</a>
+        <a href='/admin/submissions' class='active'>Submissions</a>
+      </nav>
+    </div>
+    <div class='auth'>
+      <button id='set-token' class='secondary'>Set Token</button>
+      <button id='logout' class='danger'>Logout</button>
+    </div>
+  </header>
+
+  <main>
+    <div id='error' class='error hidden'></div>
+
+    <section class='toolbar'>
+      <input id='search' type='text' placeholder='Search by file name or app id...'>
+      <button id='refresh' class='secondary'>Refresh</button>
+      <span id='count'></span>
+    </section>
+
+    <section>
+      <table>
+        <thead>
+          <tr>
+            <th class='sortable' data-key='id'>ID</th>
+            <th class='sortable' data-key='file_id'>File ID</th>
+            <th class='sortable' data-key='file_name'>File Name</th>
+            <th class='sortable' data-key='app_id'>App ID</th>
+            <th>Tracking URL</th>
+            <th class='sortable' data-key='cluster_id'>Cluster</th>
+            <th class='sortable' data-key='exit_code'>Exit</th>
+            <th class='sortable' data-key='duration_ms'>Duration (ms)</th>
+            <th class='sortable' data-key='submitted_at'>Submitted At</th>
+          </tr>
+        </thead>
+        <tbody id='submissions-body'>
+          <tr><td colspan='9'>Loading...</td></tr>
+        </tbody>
+      </table>
+    </section>
+  </main>
+
+  <dialog id='token-modal' class='modal'>
+    <h2>Admin Token Required</h2>
+    <p id='modal-error' class='modal-error hidden'></p>
+    <form id='token-form'>
+      <label for='modal-token'>Bearer token</label>
+      <input id='modal-token' type='password' autocomplete='off' required>
+      <div class='modal-actions'>
+        <button type='submit' id='modal-save'>Save</button>
+        <button type='button' id='modal-cancel' class='secondary hidden'>Cancel</button>
+      </div>
+    </form>
+  </dialog>
+
+  <script>
+    const $ = (sel) => document.querySelector(sel);
+    const API = '/admin/submissions';
+
+    let submissions = [];
+    let sortKey = 'submitted_at';
+    let sortDir = -1;
+    let pendingRetry = null;
+    let cancelBlocker = null;
+
+    function headers() {
+      return {
+        'Authorization': 'Bearer ' + (localStorage.getItem('adminToken') || ''),
+        'Content-Type': 'application/json'
+      };
+    }
+
+    function showError(msg) {
+      const el = $('#error');
+      el.textContent = msg || '';
+      el.classList.toggle('hidden', !msg);
+    }
+
+    function showModalError(msg) {
+      const el = $('#modal-error');
+      el.textContent = msg || '';
+      el.classList.toggle('hidden', !msg);
+    }
+
+    function openTokenModal(opts = {}) {
+      const dialog = $('#token-modal');
+      const cancelBtn = $('#modal-cancel');
+      showModalError(opts.error || '');
+      cancelBtn.classList.toggle('hidden', !opts.allowCancel);
+      if (cancelBlocker) {
+        dialog.removeEventListener('cancel', cancelBlocker);
+        cancelBlocker = null;
+      }
+      if (!opts.allowCancel) {
+        cancelBlocker = (e) => e.preventDefault();
+        dialog.addEventListener('cancel', cancelBlocker);
+      }
+      dialog.showModal();
+      $('#modal-token').focus();
+    }
+
+    function closeTokenModal() {
+      const dialog = $('#token-modal');
+      showModalError('');
+      if (cancelBlocker) {
+        dialog.removeEventListener('cancel', cancelBlocker);
+        cancelBlocker = null;
+      }
+      dialog.close();
+      $('#modal-token').value = '';
+    }
+
+    function saveTokenFromModal(e) {
+      e.preventDefault();
+      localStorage.setItem('adminToken', $('#modal-token').value.trim());
+      closeTokenModal();
+      if (pendingRetry) {
+        const fn = pendingRetry;
+        pendingRetry = null;
+        fn();
+      } else {
+        loadSubmissions();
+      }
+    }
+
+    async function api(method, path, body) {
+      const opts = { method, headers: headers() };
+      if (body !== undefined) opts.body = JSON.stringify(body);
+      const res = await fetch(API + path, opts);
+      if (!res.ok) {
+        if (res.status === 401) {
+          localStorage.removeItem('adminToken');
+          pendingRetry = () => api(method, path, body);
+          openTokenModal({
+            error: 'Token rejected by server (401). Enter a valid token.',
+            allowCancel: false
+          });
+          throw new Error('401 Unauthorized');
+        }
+        let detail = res.statusText;
+        try {
+          const j = await res.json();
+          if (j.error) detail = j.error;
+        } catch (_) {}
+        throw new Error(`${res.status} ${detail}`);
+      }
+      if (res.status === 204) return null;
+      return res.json();
+    }
+
+    function formatTime(ts) {
+      if (!ts) return '';
+      const d = new Date(ts);
+      return isNaN(d) ? ts : d.toISOString();
+    }
+
+    function truncate(s, n) {
+      if (!s || s.length <= n) return s || '';
+      return s.slice(0, n) + '…';
+    }
+
+    function escapeHtml(s) {
+      return String(s)
+        .replace(/&/g, '&amp;')
+        .replace(/</g, '&lt;')
+        .replace(/>/g, '&gt;')
+        .replace(/"/g, '&quot;')
+        .replace(/'/g, '&#39;');
+    }
+
+    async function loadSubmissions() {
+      showError('');
+      try {
+        const search = $('#search').value.trim();
+        const path = search ? `/api?search=${encodeURIComponent(search)}` : '/api';
+        submissions = await api('GET', path);
+        renderList();
+      } catch (e) {
+        showError('Failed to load submissions: ' + e.message);
+      }
+    }
+
+    function compare(a, b) {
+      let av = a[sortKey];
+      let bv = b[sortKey];
+      if (sortKey === 'id' || sortKey === 'exit_code' || sortKey === 'duration_ms') {
+        av = Number(av); bv = Number(bv);
+      } else if (sortKey === 'submitted_at') {
+        av = new Date(av).getTime(); bv = new Date(bv).getTime();
+      } else {
+        av = String(av).toLowerCase(); bv = String(bv).toLowerCase();
+      }
+      if (av < bv) return -sortDir;
+      if (av > bv) return sortDir;
+      return 0;
+    }
+
+    function renderList() {
+      const tbody = $('#submissions-body');
+      const list = submissions.slice().sort(compare);
+      $('#count').textContent = list.length + ' submission' + (list.length === 1 ? '' : 's');
+      tbody.innerHTML = '';
+      if (!list.length) {
+        tbody.innerHTML = '<tr><td colspan="9">No submissions.</td></tr>';
+        return;
+      }
+      for (const s of list) {
+        const tr = document.createElement('tr');
+        const fileLink = s.file_id ? `<a href='/admin/uploads/${encodeURIComponent(s.file_id)}'>${escapeHtml(truncate(s.file_id, 8))}</a>` : escapeHtml(truncate(s.file_id, 8));
+        tr.innerHTML = `
+          <td class='mono'>${escapeHtml(s.id)}</td>
+          <td class='mono' title='${escapeHtml(s.file_id)}'>${fileLink}</td>
+          <td>${escapeHtml(s.file_name || '')}</td>
+          <td class='mono'>${escapeHtml(s.app_id)}</td>
+          <td class='wrap'>${escapeHtml(s.tracking_url || '')}</td>
+          <td>${escapeHtml(s.cluster_id)}</td>
+          <td>${escapeHtml(s.exit_code)}</td>
+          <td>${escapeHtml(s.duration_ms)}</td>
+          <td>${escapeHtml(formatTime(s.submitted_at))}</td>
+        `;
+        tbody.appendChild(tr);
+      }
+    }
+
+    function handleSort(e) {
+      const key = e.target.dataset.key;
+      if (!key) return;
+      if (sortKey === key) {
+        sortDir = -sortDir;
+      } else {
+        sortKey = key;
+        sortDir = key === 'submitted_at' ? -1 : 1;
+      }
+      document.querySelectorAll('th.sortable').forEach(th => {
+        th.textContent = th.textContent.replace(/ [▲▼]$/, '');
+      });
+      const marker = sortDir > 0 ? ' ▲' : ' ▼';
+      e.target.textContent = e.target.textContent.replace(/ [▲▼]$/, '') + marker;
+      renderList();
+    }
+
+    $('#search').addEventListener('input', loadSubmissions);
+    $('#refresh').addEventListener('click', loadSubmissions);
+    $('#set-token').addEventListener('click', () => openTokenModal({ allowCancel: true }));
+    $('#logout').addEventListener('click', () => {
+      localStorage.removeItem('adminToken');
+      openTokenModal({ error: 'Token cleared. Enter a new token to continue.', allowCancel: false });
+    });
+    $('#token-form').addEventListener('submit', saveTokenFromModal);
+    $('#modal-cancel').addEventListener('click', closeTokenModal);
+    document.querySelectorAll('th.sortable').forEach(th => th.addEventListener('click', handleSort));
+
+    if (!localStorage.getItem('adminToken')) {
+      openTokenModal({ allowCancel: false });
+    } else {
+      loadSubmissions();
+    }
+  </script>
+</body>
+</html>

internal/admin/web/uploads.html

@@ -0,0 +1,378 @@
+<!DOCTYPE html>
+<html lang='en'>
+<head>
+  <meta charset='utf-8'>
+  <meta name='viewport' content='width=device-width, initial-scale=1'>
+  <title>Uploads - spark-mcp-go Admin</title>
+  <style>
+    :root {
+      --bg: #0d1117;
+      --panel: #161b22;
+      --border: #30363d;
+      --text: #c9d1d9;
+      --muted: #8b949e;
+      --accent: #58a6ff;
+      --danger: #f85149;
+      --ok: #3fb950;
+    }
+    * { box-sizing: border-box; }
+    body {
+      margin: 0;
+      font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, 'Liberation Mono', monospace;
+      background: var(--bg);
+      color: var(--text);
+      line-height: 1.5;
+    }
+    header {
+      display: flex;
+      align-items: center;
+      justify-content: space-between;
+      gap: 1rem;
+      padding: 1rem;
+      border-bottom: 1px solid var(--border);
+      background: var(--panel);
+      flex-wrap: wrap;
+    }
+    header h1 { margin: 0; font-size: 1.1rem; color: var(--accent); }
+    nav { display: flex; gap: .75rem; }
+    nav a { color: var(--muted); text-decoration: none; }
+    nav a:hover { color: var(--text); }
+    nav a.active { color: var(--accent); }
+    .auth { display: flex; align-items: center; gap: .5rem; flex-wrap: wrap; }
+    input[type='text'], input[type='password'] {
+      background: var(--bg);
+      color: var(--text);
+      border: 1px solid var(--border);
+      border-radius: 4px;
+      padding: .4rem .5rem;
+      font: inherit;
+    }
+    input[type='text'] { min-width: 220px; }
+    button {
+      background: var(--accent);
+      color: #fff;
+      border: none;
+      border-radius: 4px;
+      padding: .4rem .8rem;
+      font: inherit;
+      cursor: pointer;
+    }
+    button:hover { opacity: .9; }
+    button.danger { background: var(--danger); }
+    button.secondary { background: var(--border); color: var(--text); }
+    main { padding: 1rem; max-width: 1200px; margin: 0 auto; }
+    .toolbar { display: flex; gap: .5rem; margin-bottom: 1rem; align-items: center; flex-wrap: wrap; }
+    .toolbar span { color: var(--muted); }
+    .error {
+      background: rgba(248, 81, 73, .15);
+      border: 1px solid var(--danger);
+      color: var(--danger);
+      padding: .75rem;
+      border-radius: 4px;
+      margin-bottom: 1rem;
+      white-space: pre-wrap;
+    }
+    .hidden { display: none; }
+    table { width: 100%; border-collapse: collapse; }
+    th, td { text-align: left; padding: .5rem; border-bottom: 1px solid var(--border); }
+    th { color: var(--muted); font-weight: 600; user-select: none; }
+    th.sortable { cursor: pointer; }
+    th.sortable:hover { color: var(--text); }
+    td { vertical-align: middle; font-size: .9rem; }
+    td.mono { font-family: ui-monospace, SFMono-Regular, Menlo, Monaco, Consolas, monospace; }
+    .actions { display: flex; gap: .4rem; }
+    dialog.modal {
+      border: 1px solid var(--border);
+      background: var(--panel);
+      color: var(--text);
+      border-radius: 6px;
+      padding: 1.5rem;
+      min-width: 360px;
+      max-width: 90vw;
+    }
+    dialog.modal::backdrop { background: rgba(0, 0, 0, .65); }
+    dialog.modal h2 { margin: 0 0 .5rem; font-size: 1.1rem; color: var(--accent); }
+    dialog.modal form { display: flex; flex-direction: column; gap: .75rem; }
+    dialog.modal label { color: var(--muted); font-size: .9rem; }
+    dialog.modal input { width: 100%; }
+    .modal-error { color: var(--danger); font-size: .9rem; margin: 0; }
+    .modal-actions { display: flex; gap: .5rem; justify-content: flex-end; }
+  </style>
+</head>
+<body>
+  <header>
+    <div style='display:flex;align-items:center;gap:1rem;'>
+      <h1>spark-mcp-go Admin</h1>
+     <nav>
+       <a href='/admin'>Clusters</a>
+       <a href='/admin/uploads' class='active'>Uploads</a>
+        <a href='/admin/submissions'>Submissions</a>
+     </nav>
+    </div>
+    <div class='auth'>
+      <button id='set-token' class='secondary'>Set Token</button>
+      <button id='logout' class='danger'>Logout</button>
+    </div>
+  </header>
+
+  <main>
+    <div id='error' class='error hidden'></div>
+
+    <section class='toolbar'>
+      <input id='search' type='text' placeholder='Search by name...'>
+      <button id='refresh' class='secondary'>Refresh</button>
+      <span id='count'></span>
+    </section>
+
+    <section>
+      <table>
+        <thead>
+          <tr>
+            <th class='sortable' data-key='file_id'>File ID</th>
+            <th class='sortable' data-key='name'>Name</th>
+            <th class='sortable' data-key='size'>Size (KB)</th>
+            <th class='sortable' data-key='uploaded_at'>Uploaded At</th>
+            <th>SHA256</th>
+            <th>Actions</th>
+          </tr>
+        </thead>
+        <tbody id='uploads-body'>
+          <tr><td colspan='6'>Loading...</td></tr>
+        </tbody>
+      </table>
+    </section>
+  </main>
+
+  <dialog id='token-modal' class='modal'>
+    <h2>Admin Token Required</h2>
+    <p id='modal-error' class='modal-error hidden'></p>
+    <form id='token-form'>
+      <label for='modal-token'>Bearer token</label>
+      <input id='modal-token' type='password' autocomplete='off' required>
+      <div class='modal-actions'>
+        <button type='submit' id='modal-save'>Save</button>
+        <button type='button' id='modal-cancel' class='secondary hidden'>Cancel</button>
+      </div>
+    </form>
+  </dialog>
+
+  <script>
+    const $ = (sel) => document.querySelector(sel);
+    const API = '/admin/uploads';
+
+    let uploads = [];
+    let sortKey = 'uploaded_at';
+    let sortDir = -1;
+    let pendingRetry = null;
+    let cancelBlocker = null;
+
+    function headers() {
+      return {
+        'Authorization': 'Bearer ' + (localStorage.getItem('adminToken') || ''),
+        'Content-Type': 'application/json'
+      };
+    }
+
+    function showError(msg) {
+      const el = $('#error');
+      el.textContent = msg || '';
+      el.classList.toggle('hidden', !msg);
+    }
+
+    function showModalError(msg) {
+      const el = $('#modal-error');
+      el.textContent = msg || '';
+      el.classList.toggle('hidden', !msg);
+    }
+
+    function openTokenModal(opts = {}) {
+      const dialog = $('#token-modal');
+      const cancelBtn = $('#modal-cancel');
+      showModalError(opts.error || '');
+      cancelBtn.classList.toggle('hidden', !opts.allowCancel);
+      if (cancelBlocker) {
+        dialog.removeEventListener('cancel', cancelBlocker);
+        cancelBlocker = null;
+      }
+      if (!opts.allowCancel) {
+        cancelBlocker = (e) => e.preventDefault();
+        dialog.addEventListener('cancel', cancelBlocker);
+      }
+      dialog.showModal();
+      $('#modal-token').focus();
+    }
+
+    function closeTokenModal() {
+      const dialog = $('#token-modal');
+      showModalError('');
+      if (cancelBlocker) {
+        dialog.removeEventListener('cancel', cancelBlocker);
+        cancelBlocker = null;
+      }
+      dialog.close();
+      $('#modal-token').value = '';
+    }
+
+    function saveTokenFromModal(e) {
+      e.preventDefault();
+      localStorage.setItem('adminToken', $('#modal-token').value.trim());
+      closeTokenModal();
+      if (pendingRetry) {
+        const fn = pendingRetry;
+        pendingRetry = null;
+        fn();
+      } else {
+        loadUploads();
+      }
+    }
+
+    async function api(method, path, body) {
+      const opts = { method, headers: headers() };
+      if (body !== undefined) opts.body = JSON.stringify(body);
+      const res = await fetch(API + path, opts);
+      if (!res.ok) {
+        if (res.status === 401) {
+          localStorage.removeItem('adminToken');
+          pendingRetry = () => api(method, path, body);
+          openTokenModal({
+            error: 'Token rejected by server (401). Enter a valid token.',
+            allowCancel: false
+          });
+          throw new Error('401 Unauthorized');
+        }
+        let detail = res.statusText;
+        try {
+          const j = await res.json();
+          if (j.error) detail = j.error;
+        } catch (_) {}
+        throw new Error(`${res.status} ${detail}`);
+      }
+      if (res.status === 204) return null;
+      return res.json();
+    }
+
+    function formatKB(bytes) {
+      return (bytes / 1024).toFixed(2);
+    }
+
+    function formatTime(ts) {
+      if (!ts) return '';
+      const d = new Date(ts);
+      return isNaN(d) ? ts : d.toISOString();
+    }
+
+    function truncate(s, n) {
+      if (!s || s.length <= n) return s || '';
+      return s.slice(0, n) + '…';
+    }
+
+    function escapeHtml(s) {
+      return String(s)
+        .replace(/&/g, '&amp;')
+        .replace(/</g, '&lt;')
+        .replace(/>/g, '&gt;')
+        .replace(/"/g, '&quot;')
+        .replace(/'/g, '&#39;');
+    }
+
+    function filteredUploads() {
+      const q = $('#search').value.trim().toLowerCase();
+      if (!q) return uploads.slice();
+      return uploads.filter(u => (u.name || '').toLowerCase().includes(q));
+    }
+
+    function compare(a, b) {
+      let av = a[sortKey];
+      let bv = b[sortKey];
+      if (sortKey === 'size') { av = Number(av); bv = Number(bv); }
+      else if (sortKey === 'uploaded_at') { av = new Date(av).getTime(); bv = new Date(bv).getTime(); }
+      else { av = String(av).toLowerCase(); bv = String(bv).toLowerCase(); }
+      if (av < bv) return -sortDir;
+      if (av > bv) return sortDir;
+      return 0;
+    }
+
+    function renderList() {
+      const tbody = $('#uploads-body');
+      const list = filteredUploads().sort(compare);
+      $('#count').textContent = list.length + ' file' + (list.length === 1 ? '' : 's');
+      tbody.innerHTML = '';
+      if (!list.length) {
+        tbody.innerHTML = '<tr><td colspan="6">No uploads.</td></tr>';
+        return;
+      }
+      for (const u of list) {
+        const tr = document.createElement('tr');
+        tr.innerHTML = `
+          <td class='mono' title='${escapeHtml(u.file_id)}'>${escapeHtml(truncate(u.file_id, 8))}</td>
+          <td>${escapeHtml(u.name)}</td>
+          <td>${escapeHtml(formatKB(u.size))}</td>
+          <td>${escapeHtml(formatTime(u.uploaded_at))}</td>
+          <td class='mono' title='${escapeHtml(u.sha256)}'>${escapeHtml(truncate(u.sha256, 8))}</td>
+          <td class='actions'>
+            <button data-id='${escapeHtml(u.file_id)}' class='danger delete'>Delete</button>
+          </td>
+        `;
+        tbody.appendChild(tr);
+      }
+      tbody.querySelectorAll('.delete').forEach(b => b.addEventListener('click', handleDelete));
+    }
+
+    async function loadUploads() {
+      showError('');
+      try {
+        uploads = await api('GET', '/api');
+        renderList();
+      } catch (e) {
+        showError('Failed to load uploads: ' + e.message);
+      }
+    }
+
+    async function handleDelete(e) {
+      const id = e.target.dataset.id;
+      if (!confirm(`Delete upload ${id}?`)) return;
+      showError('');
+      try {
+        await api('DELETE', `/${encodeURIComponent(id)}`);
+        await loadUploads();
+      } catch (err) {
+        showError('Failed to delete upload: ' + err.message);
+      }
+    }
+
+    function handleSort(e) {
+      const key = e.target.dataset.key;
+      if (!key) return;
+      if (sortKey === key) {
+        sortDir = -sortDir;
+      } else {
+        sortKey = key;
+        sortDir = key === 'uploaded_at' ? -1 : 1;
+      }
+      document.querySelectorAll('th.sortable').forEach(th => {
+        th.textContent = th.textContent.replace(/ [▲▼]$/, '');
+      });
+      const marker = sortDir > 0 ? ' ▲' : ' ▼';
+      e.target.textContent = e.target.textContent.replace(/ [▲▼]$/, '') + marker;
+      renderList();
+    }
+
+    $('#search').addEventListener('input', renderList);
+    $('#refresh').addEventListener('click', loadUploads);
+    $('#set-token').addEventListener('click', () => openTokenModal({ allowCancel: true }));
+    $('#logout').addEventListener('click', () => {
+      localStorage.removeItem('adminToken');
+      openTokenModal({ error: 'Token cleared. Enter a new token to continue.', allowCancel: false });
+    });
+    $('#token-form').addEventListener('submit', saveTokenFromModal);
+    $('#modal-cancel').addEventListener('click', closeTokenModal);
+    document.querySelectorAll('th.sortable').forEach(th => th.addEventListener('click', handleSort));
+
+    if (!localStorage.getItem('adminToken')) {
+      openTokenModal({ allowCancel: false });
+    } else {
+      loadUploads();
+    }
+  </script>
+</body>
+</html>

internal/audit/repo.go

@@ -18,6 +18,8 @@ const (
 	ActionClusterCreate Action = "cluster.create"
 	ActionClusterUpdate Action = "cluster.update"
 	ActionClusterDelete Action = "cluster.delete"
+	ActionUploadCreate  Action = "upload.create"
+	ActionUploadDelete  Action = "upload.delete"
 )
 
 // Entry is one admin write operation recorded for accountability.

internal/executor/spark_submit.go

@@ -23,10 +23,13 @@ import (
 // application application_xxx" line and a few progress lines.
 const stdoutCap = 10 << 20
 
-// appIDPattern matches the YARN line printed on successful submission:
+// submittedAppPattern matches the explicit success line, including the
-//
+// application id captured in group 1. This mirrors the Python reference.
-//	"Submitted application application_12345_0001"
+var submittedAppPattern = regexp.MustCompile(`Submitted application (\S+)`)
-var appIDPattern = regexp.MustCompile(`application_\d+_\d+`)
+
+// trackingURLPattern matches "tracking URL: <url>" on stderr. The URL tail
+// is used as a fallback source for the application id.
+var trackingURLPattern = regexp.MustCompile(`tracking URL:\s+(\S+)`)
 
 // SparkSubmitOpts is the resolved command line for a single spark-submit run.
 //
@@ -43,11 +46,12 @@ type SparkSubmitOpts struct {
 
 // Result is what spark_submit returns to the LLM.
 type Result struct {
-	AppID      string `json:"app_id,omitempty"`
+	AppID       string `json:"app_id,omitempty"`
-	ExitCode   int    `json:"exit_code"`
+	TrackingURL string `json:"tracking_url,omitempty"`
-	StdoutTail string `json:"stdout_tail"`
+	ExitCode    int    `json:"exit_code"`
-	StderrTail string `json:"stderr_tail"`
+	StdoutTail  string `json:"stdout_tail"`
-	DurationMS int64  `json:"duration_ms"`
+	StderrTail  string `json:"stderr_tail"`
+	DurationMS  int64  `json:"duration_ms"`
 }
 
 // ErrBinaryEmpty is returned when the cluster has no spark-submit binary
@@ -90,27 +94,70 @@ func Run(ctx context.Context, opts SparkSubmitOpts) (*Result, error) {
 		DurationMS: dur.Milliseconds(),
 	}
 
-	// app_id: prefer stdout, fall back to stderr. Both are searched only on
+	appID, trackingURL, ok := parseSparkSubmitOutput(res.StdoutTail, res.StderrTail)
-	// the tail (cap-bounded) so we don't pay for the full output.
+	res.AppID = appID
-	if id := matchAppID(res.StdoutTail); id != "" {
+	res.TrackingURL = trackingURL
-		res.AppID = id
-	} else if id := matchAppID(res.StderrTail); id != "" {
-		res.AppID = id
-	}
 
+	if !ok {
+		return res, fmt.Errorf("executor: spark-submit ran but no application_id in output (exit_code=%d)", res.ExitCode)
+	}
 	if err != nil {
-		// Wrap the underlying error but keep the parsed result so the LLM
-		// can see exit_code and stderr even on failure.
 		return res, fmt.Errorf("executor: spark-submit failed: %w", err)
 	}
 	return res, nil
 }
 
-func matchAppID(s string) string {
+// parseSparkSubmitOutput extracts the application id and tracking URL from
-	if s == "" {
+// spark-submit output. It mirrors the Python reference implementation: first
+// look for "Submitted application <id>", then fall back to "tracking URL:",
+// deriving the id from the last path segment. The returned ok is false when
+// no application id could be determined.
+func parseSparkSubmitOutput(stdout, stderr string) (appID, trackingURL string, ok bool) {
+	// Stage 1: explicit success line, prefer stdout then stderr.
+	var trackingFound bool
+	if m := submittedAppPattern.FindStringSubmatch(stdout); m != nil {
+		appID = m[1]
+		ok = true
+	} else if m := submittedAppPattern.FindStringSubmatch(stderr); m != nil {
+		appID = m[1]
+		ok = true
+	}
+
+	// Stage 2: tracking URL fallback (stderr only, like the Python reference).
+	if !ok {
+		if m := trackingURLPattern.FindStringSubmatch(stderr); m != nil {
+			trackingURL = m[1]
+			trackingFound = true
+			appID = extractAppIDFromTrackingURL(trackingURL)
+			ok = appID != ""
+			if !ok {
+				trackingURL = ""
+			}
+		}
+	}
+
+	// Tracking URL is independent: if present anywhere on stderr, capture it.
+	if trackingURL == "" && !trackingFound {
+		if m := trackingURLPattern.FindStringSubmatch(stderr); m != nil {
+			trackingURL = m[1]
+		}
+	}
+	return
+}
+
+// extractAppIDFromTrackingURL returns the last path segment if it starts with
+// "application_", otherwise an empty string.
+func extractAppIDFromTrackingURL(rawURL string) string {
+	trimmed := strings.TrimRight(rawURL, "/")
+	idx := strings.LastIndex(trimmed, "/")
+	if idx < 0 {
 		return ""
 	}
-	return appIDPattern.FindString(s)
+	tail := trimmed[idx+1:]
+	if strings.HasPrefix(tail, "application_") {
+		return tail
+	}
+	return ""
 }
 
 // exitCodeFromErr returns 0 on nil, the process's exit code on *exec.ExitError,

internal/executor/spark_submit_parse_test.go

@@ -0,0 +1,79 @@
+package executor
+
+import "testing"
+
+func TestParseAppID_FromStdout(t *testing.T) {
+	stdout := "some prefix\nSubmitted application application_123_456\n"
+	stderr := "random log\n"
+	appID, trackingURL, ok := parseSparkSubmitOutput(stdout, stderr)
+	if !ok {
+		t.Fatal("expected ok")
+	}
+	if appID != "application_123_456" {
+		t.Errorf("appID=%q, want application_123_456", appID)
+	}
+	if trackingURL != "" {
+		t.Errorf("trackingURL=%q, want empty", trackingURL)
+	}
+}
+
+func TestParseAppID_FromStderr(t *testing.T) {
+	stdout := ""
+	stderr := "Submitted application application_123_456\n"
+	appID, trackingURL, ok := parseSparkSubmitOutput(stdout, stderr)
+	if !ok {
+		t.Fatal("expected ok")
+	}
+	if appID != "application_123_456" {
+		t.Errorf("appID=%q, want application_123_456", appID)
+	}
+	if trackingURL != "" {
+		t.Errorf("trackingURL=%q, want empty", trackingURL)
+	}
+}
+
+func TestParseAppID_FromTrackingURL(t *testing.T) {
+	stdout := ""
+	stderr := "Log line\ntracking URL: http://rm.example.com:8088/proxy/application_123_456/\n"
+	appID, trackingURL, ok := parseSparkSubmitOutput(stdout, stderr)
+	if !ok {
+		t.Fatal("expected ok")
+	}
+	if appID != "application_123_456" {
+		t.Errorf("appID=%q, want application_123_456", appID)
+	}
+	wantURL := "http://rm.example.com:8088/proxy/application_123_456/"
+	if trackingURL != wantURL {
+		t.Errorf("trackingURL=%q, want %q", trackingURL, wantURL)
+	}
+}
+
+func TestParseAppID_NoAppID(t *testing.T) {
+	stdout := ""
+	stderr := "some random log"
+	appID, trackingURL, ok := parseSparkSubmitOutput(stdout, stderr)
+	if ok {
+		t.Fatal("expected not ok")
+	}
+	if appID != "" {
+		t.Errorf("appID=%q, want empty", appID)
+	}
+	if trackingURL != "" {
+		t.Errorf("trackingURL=%q, want empty", trackingURL)
+	}
+}
+
+func TestParseAppID_BadTrackingURL(t *testing.T) {
+	stdout := ""
+	stderr := "tracking URL: http://rm.example.com:8088/proxy/not-an-app/\n"
+	appID, trackingURL, ok := parseSparkSubmitOutput(stdout, stderr)
+	if ok {
+		t.Fatal("expected not ok")
+	}
+	if appID != "" {
+		t.Errorf("appID=%q, want empty", appID)
+	}
+	if trackingURL != "" {
+		t.Errorf("trackingURL=%q, want empty", trackingURL)
+	}
+}

internal/executor/spark_submit_test.go

@@ -30,7 +30,7 @@ func TestRun_FakeBinary(t *testing.T) {
 
 	res, err := Run(ctx, SparkSubmitOpts{
 		Binary:  "/bin/echo",
-		Args:    []string{"--master", "yarn; rm -rf /tmp/this-should-not-exist"},
+		Args:    []string{"--master", "yarn; rm -rf /tmp/this-should-not-exist", "Submitted application application_12345_0001"},
 		Timeout: 3 * time.Second,
 	})
 	if err != nil {
@@ -43,8 +43,8 @@ func TestRun_FakeBinary(t *testing.T) {
 	if !strings.Contains(res.StdoutTail, "yarn; rm -rf /tmp/this-should-not-exist") {
 		t.Errorf("stdout missing literal arg, got: %q", res.StdoutTail)
 	}
-	if res.AppID != "" {
+	if res.AppID != "application_12345_0001" {
-		t.Errorf("AppID should be empty for echo, got %q", res.AppID)
+		t.Errorf("AppID = %q, want application_12345_0001", res.AppID)
 	}
 }
 
@@ -66,22 +66,26 @@ func TestRun_NonZeroExit(t *testing.T) {
 	}
 }
 
-// TestMatchAppID locks down the regex so a YARN output tweak doesn't
+// TestSubmittedAppPattern locks down the regex so a YARN output tweak doesn't
 // silently break app_id extraction.
-func TestMatchAppID(t *testing.T) {
+func TestSubmittedAppPattern(t *testing.T) {
 	cases := []struct {
 		in, want string
 	}{
 		{"Submitted application application_12345_0001", "application_12345_0001"},
-		{"... some prefix application_999_42 ... tail", "application_999_42"},
+		{"Submitted application application_999_42 extra", "application_999_42"},
 		{"no app id here", ""},
 		{"", ""},
-		{"application_1_2 application_3_4", "application_1_2"}, // first match
+		{"Submitted application app_1_2 Submitted application app_3_4", "app_1_2"}, // first match
 	}
 	for _, tc := range cases {
-		got := matchAppID(tc.in)
+		m := submittedAppPattern.FindStringSubmatch(tc.in)
+		var got string
+		if m != nil {
+			got = m[1]
+		}
 		if got != tc.want {
-			t.Errorf("matchAppID(%q) = %q, want %q", tc.in, got, tc.want)
+			t.Errorf("submittedAppPattern(%q) = %q, want %q", tc.in, got, tc.want)
 		}
 	}
 }

internal/mcp/tools/deps.go

@@ -5,6 +5,7 @@ import (
 	"time"
 
 	"spark-mcp-go/internal/analyzer"
+	"spark-mcp-go/internal/audit"
 	"spark-mcp-go/internal/httpclient"
 	"spark-mcp-go/internal/storage"
 	"spark-mcp-go/internal/uploads"
@@ -13,7 +14,9 @@ import (
 // Deps bundles the dependencies shared by all MCP Tool handlers.
 type Deps struct {
 	Logger             *slog.Logger
+	AuditRepo          *audit.Repo
 	ClusterRepo        *storage.ClusterRepo
+	SubmissionRepo     *storage.SubmissionRepo
 	SparkSubmitTimeout time.Duration
 	HTTPClient         *httpclient.Client
 	MaxResponseBytes   int64

internal/mcp/tools/fetch_upload_test.go

@@ -16,6 +16,7 @@ import (
 
 	"github.com/mark3labs/mcp-go/mcp"
 
+	"spark-mcp-go/internal/audit"
 	"spark-mcp-go/internal/cluster"
 	"spark-mcp-go/internal/httpclient"
 	"spark-mcp-go/internal/storage"
@@ -24,7 +25,7 @@ import (
 
 // testDepsWithDataDir returns dependencies backed by an in-memory DB and a
 // temporary data directory.
-func testDepsWithDataDir(t *testing.T) (*Deps, *storage.ClusterRepo) {
+func testDepsWithDataDir(t *testing.T) (*Deps, *storage.ClusterRepo, *storage.SubmissionRepo, *audit.Repo) {
 	t.Helper()
 	db, err := storage.Open(":memory:")
 	if err != nil {
@@ -43,11 +44,13 @@ func testDepsWithDataDir(t *testing.T) (*Deps, *storage.ClusterRepo) {
 			Timeout:          5 * time.Second,
 			MaxResponseBytes: 1 << 20,
 		}),
+		AuditRepo:        audit.NewRepo(db),
 		ClusterRepo:      db.Clusters(),
+		SubmissionRepo:   db.Submissions(),
 		MaxResponseBytes: 1 << 20,
 		DataDir:          dataDir,
 		UploadStore:      uploadStore,
-	}, db.Clusters()
+	}, db.Clusters(), db.Submissions(), audit.NewRepo(db)
 }
 
 // createCluster creates a cluster in the repository with the given fields.
@@ -98,7 +101,7 @@ func TestFetchURL(t *testing.T) {
 	}))
 	defer srv.Close()
 
-	deps, repo := testDepsWithDataDir(t)
+	deps, repo, _, _ := testDepsWithDataDir(t)
 	createCluster(t, repo, &cluster.Cluster{
 		ID:           "cluster-a",
 		Name:         "Cluster A",
@@ -292,7 +295,7 @@ func TestFetchURL_RedirectPreservesAuth(t *testing.T) {
 	}))
 	defer server1.Close()
 
-	deps, repo := testDepsWithDataDir(t)
+	deps, repo, _, _ := testDepsWithDataDir(t)
 	createCluster(t, repo, &cluster.Cluster{
 		ID:           "cluster-redirect",
 		Name:         "Redirect",
@@ -321,7 +324,7 @@ func TestFetchURL_RedirectPreservesAuth(t *testing.T) {
 }
 
 func TestUploadFile(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, auditRepo := testDepsWithDataDir(t)
 
 	req := newToolRequest(UploadFileName, map[string]any{
 		"filename": "hello.txt",
@@ -367,10 +370,25 @@ func TestUploadFile(t *testing.T) {
 	if info.Mode().Perm() != 0o640 {
 		t.Errorf("mode=%o, want %o", info.Mode().Perm(), 0o640)
 	}
+
+	entries, err := auditRepo.List(context.Background(), 100)
+	if err != nil {
+		t.Fatalf("list audit: %v", err)
+	}
+	found := false
+	for _, e := range entries {
+		if e.Action == audit.ActionUploadCreate && e.Actor == "tool:upload_file" && e.ClusterID == fileID {
+			found = true
+			break
+		}
+	}
+	if !found {
+		t.Errorf("missing upload.create audit entry for file_id %s", fileID)
+	}
 }
 
 func TestUploadFile_PathTraversal(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, _ := testDepsWithDataDir(t)
 
 	cases := []string{
 		"../../../etc/passwd",
@@ -401,7 +419,7 @@ func TestUploadFile_PathTraversal(t *testing.T) {
 }
 
 func TestUploadFile_Base64(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, _ := testDepsWithDataDir(t)
 
 	req := newToolRequest(UploadFileName, map[string]any{
 		"filename": "hello.bin",

internal/mcp/tools/spark_submit.go

@@ -6,10 +6,12 @@ import (
 	"log/slog"
 	"sort"
 	"strconv"
+	"time"
 
 	"github.com/mark3labs/mcp-go/mcp"
 
 	"spark-mcp-go/internal/executor"
+	"spark-mcp-go/internal/storage"
 )
 
 const SparkSubmitName = "spark_submit"
@@ -98,7 +100,8 @@ func (d *Deps) SparkSubmitHandler(ctx context.Context, req mcp.CallToolRequest)
 	// letting spark-submit produce a confusing FileNotFoundException later.
 	// Placed after all required-string parses so error messages surface in
 	// field order (cluster_id, master, ..., script_path, queue, ...).
-	if _, err := d.UploadStore.Validate(scriptPath); err != nil {
+	fileID, err := d.UploadStore.Validate(scriptPath)
+	if err != nil {
 		if d.Logger != nil {
 			d.Logger.Warn("spark_submit.unminted_path_rejected", slog.String("path", scriptPath), slog.String("err", err.Error()))
 		}
@@ -173,35 +176,58 @@ func (d *Deps) SparkSubmitHandler(ctx context.Context, req mcp.CallToolRequest)
 	})
 	if err != nil {
 		callLog.WithError(err)
-		// Even on error, result carries ExitCode/Stderr for the LLM.
 		if result != nil {
-			callLog.WithResult(map[string]any{
+			callLog.WithResult(sparkSubmitResultLog(binary, cmd, result))
-				"binary":      binary,
-				"argv":        cmd,
-				"app_id":      result.AppID,
-				"exit_code":   result.ExitCode,
-				"duration_ms": result.DurationMS,
-			})
 			return textResult(encodeJSON(map[string]any{
-				"app_id":      result.AppID,
+				"app_id":       result.AppID,
-				"exit_code":   result.ExitCode,
+				"tracking_url": result.TrackingURL,
-				"stdout_tail": result.StdoutTail,
+				"exit_code":    result.ExitCode,
-				"stderr_tail": result.StderrTail,
+				"stdout_tail":  result.StdoutTail,
-				"duration_ms": result.DurationMS,
+				"stderr_tail":  result.StderrTail,
-				"error":       err.Error(),
+				"duration_ms":  result.DurationMS,
+				"error":        err.Error(),
 			})), nil
 		}
 		return errResult("spark_submit: " + err.Error()), nil
 	}
 
-	callLog.WithResult(map[string]any{
+	if result.AppID != "" && d.SubmissionRepo != nil {
+		sub := storage.Submission{
+			FileID:      fileID,
+			AppID:       result.AppID,
+			TrackingURL: result.TrackingURL,
+			ClusterID:   clusterID,
+			ExitCode:    result.ExitCode,
+			DurationMS:  result.DurationMS,
+			SubmittedAt: time.Now(),
+		}
+		if createErr := d.SubmissionRepo.Create(ctx, sub); createErr != nil {
+			if d.Logger != nil {
+				d.Logger.Warn("spark_submit.persist_failed", slog.String("file_id", fileID), slog.String("app_id", result.AppID), slog.String("err", createErr.Error()))
+			} else {
+				slog.Default().Warn("spark_submit.persist_failed", slog.String("file_id", fileID), slog.String("app_id", result.AppID), slog.String("err", createErr.Error()))
+			}
+		}
+	}
+
+	callLog.WithResult(sparkSubmitResultLog(binary, cmd, result))
+	return textResult(encodeJSON(result)), nil
+}
+
+// sparkSubmitResultLog builds the standard result map used for both logging
+// and the admin audit trail.
+func sparkSubmitResultLog(binary string, cmd []string, result *executor.Result) map[string]any {
+	m := map[string]any{
 		"binary":      binary,
 		"argv":        cmd,
 		"app_id":      result.AppID,
 		"exit_code":   result.ExitCode,
 		"duration_ms": result.DurationMS,
-	})
+	}
-	return textResult(encodeJSON(result)), nil
+	if result.TrackingURL != "" {
+		m["tracking_url"] = result.TrackingURL
+	}
+	return m
 }
 
 // SparkSubmitCommandOpts holds the structured arguments used to build the

internal/mcp/tools/spark_submit_path_test.go

@@ -10,11 +10,15 @@ import (
 
 	"github.com/mark3labs/mcp-go/mcp"
 
+	"spark-mcp-go/internal/audit"
 	"spark-mcp-go/internal/cluster"
+	"spark-mcp-go/internal/httpclient"
+	"spark-mcp-go/internal/storage"
+	"spark-mcp-go/internal/uploads"
 )
 
 func TestSparkSubmit_StructuredCommand(t *testing.T) {
-	deps, repo := testDepsWithDataDir(t)
+	deps, repo, _, _ := testDepsWithDataDir(t)
 	deps.SparkSubmitTimeout = 5 * time.Second
 	store := deps.UploadStore
 
@@ -25,7 +29,7 @@ func TestSparkSubmit_StructuredCommand(t *testing.T) {
 
 	echoed := filepath.Join(t.TempDir(), "echoed")
 	echoScript := filepath.Join(t.TempDir(), "echo-args.sh")
-	if err := os.WriteFile(echoScript, []byte("#!/bin/sh\nfor arg do\n  echo \"$arg\" >> \"$ECHO_FILE\"\ndone\n"), 0o755); err != nil {
+	if err := os.WriteFile(echoScript, []byte("#!/bin/sh\nfor arg do\n  echo \"$arg\" >> \"$ECHO_FILE\"\ndone\necho 'Submitted application application_test_0001'\n"), 0o755); err != nil {
 		t.Fatalf("write echo script: %v", err)
 	}
 
@@ -97,8 +101,107 @@ func TestSparkSubmit_StructuredCommand(t *testing.T) {
 	}
 }
 
+func TestSparkSubmit_PersistsSubmission(t *testing.T) {
+	// Open a dedicated DB so we can insert upload_files metadata for the join.
+	db, err := storage.Open(":memory:")
+	if err != nil {
+		t.Fatalf("open db: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+
+	dataDir := t.TempDir()
+	store, err := uploads.New(filepath.Join(dataDir, "uploads"))
+	if err != nil {
+		t.Fatalf("new upload store: %v", err)
+	}
+
+	deps := &Deps{
+		HTTPClient:         httpclient.New(httpclient.Config{Timeout: 5 * time.Second, MaxResponseBytes: 1 << 20}),
+		AuditRepo:          audit.NewRepo(db),
+		ClusterRepo:        db.Clusters(),
+		SubmissionRepo:     db.Submissions(),
+		MaxResponseBytes:   1 << 20,
+		DataDir:            dataDir,
+		UploadStore:        store,
+		SparkSubmitTimeout: 5 * time.Second,
+	}
+	repo := db.Clusters()
+	subRepo := db.Submissions()
+
+	fileID, _, _, _, absPath, err := store.Save([]byte("print('hi')\n"), "hello.py")
+	if err != nil {
+		t.Fatalf("save upload: %v", err)
+	}
+	if err := db.Uploads().Create(context.Background(), fileID, "hello.py", 14, "dummy-sha", time.Now()); err != nil {
+		t.Fatalf("create upload record: %v", err)
+	}
+
+	echoScript := filepath.Join(t.TempDir(), "echo-submit.sh")
+	if err := os.WriteFile(echoScript, []byte("#!/bin/sh\necho 'tracking URL: http://rm.example.com:8088/proxy/application_persist_0001/' >&2\necho 'Submitted application application_persist_0001'\n"), 0o755); err != nil {
+		t.Fatalf("write echo script: %v", err)
+	}
+
+	createCluster(t, repo, &cluster.Cluster{
+		ID:                    "cluster-persist",
+		Name:                  "Persist",
+		IsActive:              true,
+		AuthType:              cluster.AuthNone,
+		SparkSubmitExecuteBin: echoScript,
+	})
+
+	req := newToolRequest(SparkSubmitName, map[string]any{
+		"cluster_id":      "cluster-persist",
+		"master":          "yarn",
+		"deploy_mode":     "cluster",
+		"script_path":     absPath,
+		"queue":           "default",
+		"executor_memory": "2G",
+		"executor_cores":  1,
+		"num_executors":   4,
+	})
+	res, err := deps.SparkSubmitHandler(context.Background(), req)
+	if err != nil {
+		t.Fatalf("handler error: %v", err)
+	}
+	if res.IsError {
+		t.Fatalf("unexpected error result: %v", res.Content)
+	}
+
+	payload := resultJSON(t, res)
+	if payload["app_id"] != "application_persist_0001" {
+		t.Errorf("app_id=%v, want application_persist_0001", payload["app_id"])
+	}
+	if payload["tracking_url"] != "http://rm.example.com:8088/proxy/application_persist_0001/" {
+		t.Errorf("tracking_url=%v, want tracking URL", payload["tracking_url"])
+	}
+
+	subs, err := subRepo.List(context.Background(), storage.SubmissionFilter{Limit: 10})
+	if err != nil {
+		t.Fatalf("list submissions: %v", err)
+	}
+	if len(subs) != 1 {
+		t.Fatalf("got %d submissions, want 1", len(subs))
+	}
+	s := subs[0]
+	if s.FileID != fileID {
+		t.Errorf("file_id=%q, want %q", s.FileID, fileID)
+	}
+	if s.AppID != "application_persist_0001" {
+		t.Errorf("app_id=%q, want application_persist_0001", s.AppID)
+	}
+	if s.TrackingURL != "http://rm.example.com:8088/proxy/application_persist_0001/" {
+		t.Errorf("tracking_url=%q, want tracking URL", s.TrackingURL)
+	}
+	if s.ClusterID != "cluster-persist" {
+		t.Errorf("cluster_id=%q, want cluster-persist", s.ClusterID)
+	}
+	if s.FileName != "hello.py" {
+		t.Errorf("file_name=%q, want hello.py", s.FileName)
+	}
+}
+
 func TestSparkSubmit_MissingRequiredField(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, _ := testDepsWithDataDir(t)
 	store := deps.UploadStore
 	_, _, _, _, mintedPath, err := store.Save([]byte("# dummy\n"), "dummy.py")
 	if err != nil {
@@ -131,7 +234,7 @@ func TestSparkSubmit_MissingRequiredField(t *testing.T) {
 }
 
 func TestSparkSubmit_BadSparkConfValue(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, _ := testDepsWithDataDir(t)
 	store := deps.UploadStore
 	_, _, _, _, mintedPath, err := store.Save([]byte("# dummy\n"), "dummy.py")
 	if err != nil {
@@ -166,7 +269,7 @@ func TestSparkSubmit_BadSparkConfValue(t *testing.T) {
 }
 
 func TestSparkSubmit_RejectsNonMintedPath(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, _ := testDepsWithDataDir(t)
 
 	unminted := filepath.Join(t.TempDir(), "unminted.py")
 	if err := os.WriteFile(unminted, []byte("print('not from upload_file')\n"), 0o644); err != nil {
@@ -200,7 +303,7 @@ func TestSparkSubmit_RejectsNonMintedPath(t *testing.T) {
 }
 
 func TestSparkSubmit_EmptyMaster(t *testing.T) {
-	deps, _ := testDepsWithDataDir(t)
+	deps, _, _, _ := testDepsWithDataDir(t)
 	store := deps.UploadStore
 	_, _, _, _, mintedPath, err := store.Save([]byte("# dummy\n"), "dummy.py")
 	if err != nil {

internal/mcp/tools/upload_file.go

@@ -8,6 +8,8 @@ import (
 	"regexp"
 
 	"github.com/mark3labs/mcp-go/mcp"
+
+	"spark-mcp-go/internal/audit"
 )
 
 const UploadFileName = "upload_file"
@@ -76,6 +78,21 @@ func (d *Deps) UploadFileHandler(ctx context.Context, req mcp.CallToolRequest) (
 		return errResult("upload_file: save upload: " + err.Error()), nil
 	}
 
+	if d.AuditRepo != nil {
+		details, _ := audit.MarshalDetails(map[string]any{
+			"file_id": fileID,
+			"name":    filename,
+			"size":    size,
+			"sha256":  sha256Hex,
+		})
+		_ = d.AuditRepo.Insert(ctx, &audit.Entry{
+			Actor:     "tool:upload_file",
+			Action:    audit.ActionUploadCreate,
+			ClusterID: fileID,
+			Details:   details,
+		})
+	}
+
 	result := map[string]any{
 		"file_id": fileID,
 		"path":    absPath,

internal/storage/sqlite.go

@@ -38,6 +38,28 @@ CREATE TABLE IF NOT EXISTS audit_log (
   details     TEXT
 );
 CREATE INDEX IF NOT EXISTS idx_audit_ts ON audit_log(timestamp DESC);
+
+CREATE TABLE IF NOT EXISTS upload_files (
+  file_id      TEXT PRIMARY KEY,
+  name         TEXT NOT NULL,
+  size         INTEGER NOT NULL,
+  sha256       TEXT NOT NULL,
+  uploaded_at  INTEGER NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_uploads_ts ON upload_files(uploaded_at DESC);
+CREATE TABLE IF NOT EXISTS spark_submissions (
+  id            INTEGER PRIMARY KEY AUTOINCREMENT,
+  file_id       TEXT NOT NULL,
+  app_id        TEXT NOT NULL,
+  tracking_url  TEXT,
+  cluster_id    TEXT NOT NULL,
+  exit_code     INTEGER NOT NULL,
+  duration_ms   INTEGER NOT NULL,
+  submitted_at  INTEGER NOT NULL
+);
+CREATE INDEX IF NOT EXISTS idx_submissions_app_id ON spark_submissions(app_id);
+CREATE INDEX IF NOT EXISTS idx_submissions_submitted_at ON spark_submissions(submitted_at DESC);
+CREATE INDEX IF NOT EXISTS idx_submissions_file_id ON spark_submissions(file_id);
 `
 
 // DB is the storage handle.

internal/storage/submission_repo.go

@@ -0,0 +1,194 @@
+package storage
+
+import (
+	"context"
+	"database/sql"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
+)
+
+// Submission is one persisted spark_submit invocation.
+type Submission struct {
+	ID          int64     `json:"id"`
+	FileID      string    `json:"file_id"`
+	AppID       string    `json:"app_id"`
+	TrackingURL string    `json:"tracking_url,omitempty"`
+	ClusterID   string    `json:"cluster_id"`
+	ExitCode    int       `json:"exit_code"`
+	DurationMS  int64     `json:"duration_ms"`
+	SubmittedAt time.Time `json:"submitted_at"`
+}
+
+// SubmissionWithFile joins a submission with the original upload file name.
+// FileName is empty when the referenced upload has been deleted.
+type SubmissionWithFile struct {
+	Submission
+	FileName string `json:"file_name"`
+}
+
+// SubmissionFilter controls the rows returned by SubmissionRepo.List.
+type SubmissionFilter struct {
+	Search    string
+	FileID    string
+	ClusterID string
+	AppID     string
+	Limit     int
+	Offset    int
+}
+
+// SubmissionRepo provides CRUD operations for spark_submissions.
+type SubmissionRepo struct {
+	db *DB
+}
+
+// Submissions returns a repository bound to this DB handle.
+func (d *DB) Submissions() *SubmissionRepo {
+	return &SubmissionRepo{db: d}
+}
+
+// Create inserts a new submission record.
+func (r *SubmissionRepo) Create(ctx context.Context, s Submission) error {
+	if s.SubmittedAt.IsZero() {
+		s.SubmittedAt = time.Now()
+	}
+	_, err := r.db.sqlDB.ExecContext(ctx, `
+		INSERT INTO spark_submissions (file_id, app_id, tracking_url, cluster_id, exit_code, duration_ms, submitted_at)
+		VALUES (?, ?, ?, ?, ?, ?, ?)`,
+		s.FileID, s.AppID, s.TrackingURL, s.ClusterID, s.ExitCode, s.DurationMS, s.SubmittedAt.UnixNano(),
+	)
+	if err != nil {
+		return fmt.Errorf("storage: create submission: %w", err)
+	}
+	return nil
+}
+
+// Get retrieves a submission by id.
+func (r *SubmissionRepo) Get(ctx context.Context, id int64) (Submission, error) {
+	row := r.db.sqlDB.QueryRowContext(ctx, `
+		SELECT id, file_id, app_id, tracking_url, cluster_id, exit_code, duration_ms, submitted_at
+		FROM spark_submissions
+		WHERE id = ?`, id)
+	s, err := scanSubmission(row.Scan)
+	if errors.Is(err, sql.ErrNoRows) {
+		return Submission{}, ErrNotFound
+	}
+	if err != nil {
+		return Submission{}, fmt.Errorf("storage: get submission %d: %w", id, err)
+	}
+	return s, nil
+}
+
+// List returns submissions ordered by submitted_at descending.
+//
+// Search matches app_id or the joined upload file name. Empty filter strings
+// mean no restriction. Limit defaults to 100 and is capped at 500.
+func (r *SubmissionRepo) List(ctx context.Context, filter SubmissionFilter) ([]SubmissionWithFile, error) {
+	if filter.Limit <= 0 {
+		filter.Limit = 100
+	}
+	if filter.Limit > 500 {
+		filter.Limit = 500
+	}
+	if filter.Offset < 0 {
+		filter.Offset = 0
+	}
+
+	args := []any{}
+	var where []string
+	if filter.Search != "" {
+		like := "%" + filter.Search + "%"
+		args = append(args, like, like)
+		where = append(where, "(s.app_id LIKE ? OR COALESCE(u.name, '') LIKE ?)")
+	}
+	if filter.FileID != "" {
+		args = append(args, filter.FileID)
+		where = append(where, "s.file_id = ?")
+	}
+	if filter.ClusterID != "" {
+		args = append(args, filter.ClusterID)
+		where = append(where, "s.cluster_id = ?")
+	}
+	if filter.AppID != "" {
+		args = append(args, filter.AppID)
+		where = append(where, "s.app_id = ?")
+	}
+
+	// WHERE args are in the order the conditions were appended; append LIMIT/OFFSET last.
+	args = append(args, filter.Limit, filter.Offset)
+
+	q := `SELECT s.id, s.file_id, s.app_id, s.tracking_url, s.cluster_id, s.exit_code, s.duration_ms, s.submitted_at, COALESCE(u.name, '') AS file_name
+		FROM spark_submissions s
+		LEFT JOIN upload_files u ON s.file_id = u.file_id`
+	if len(where) > 0 {
+		q += "\n\t\tWHERE " + strings.Join(where, " AND ")
+	}
+	q += "\n\t\tORDER BY s.submitted_at DESC\n\t\tLIMIT ? OFFSET ?"
+
+	rows, err := r.db.sqlDB.QueryContext(ctx, q, args...)
+	if err != nil {
+		return nil, fmt.Errorf("storage: list submissions: %w", err)
+	}
+	defer rows.Close()
+
+	var out []SubmissionWithFile
+	for rows.Next() {
+		swf, err := scanSubmissionWithFile(rows.Scan)
+		if err != nil {
+			return nil, fmt.Errorf("storage: list submissions: %w", err)
+		}
+		out = append(out, swf)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("storage: list submissions: %w", err)
+	}
+	if out == nil {
+		out = []SubmissionWithFile{}
+	}
+	return out, nil
+}
+
+// Delete removes a submission by id.
+func (r *SubmissionRepo) Delete(ctx context.Context, id int64) error {
+	res, err := r.db.sqlDB.ExecContext(ctx, `DELETE FROM spark_submissions WHERE id = ?`, id)
+	if err != nil {
+		return fmt.Errorf("storage: delete submission %d: %w", id, err)
+	}
+	n, err := res.RowsAffected()
+	if err != nil {
+		return fmt.Errorf("storage: delete submission %d: rows affected: %w", id, err)
+	}
+	if n == 0 {
+		return ErrNotFound
+	}
+	return nil
+}
+
+func scanSubmission(scan scanFunc) (Submission, error) {
+	var s Submission
+	var trackingURL sql.NullString
+	var submittedAt int64
+	if err := scan(&s.ID, &s.FileID, &s.AppID, &trackingURL, &s.ClusterID, &s.ExitCode, &s.DurationMS, &submittedAt); err != nil {
+		return Submission{}, err
+	}
+	if trackingURL.Valid {
+		s.TrackingURL = trackingURL.String
+	}
+	s.SubmittedAt = time.Unix(0, submittedAt)
+	return s, nil
+}
+
+func scanSubmissionWithFile(scan scanFunc) (SubmissionWithFile, error) {
+	var swf SubmissionWithFile
+	var trackingURL sql.NullString
+	var submittedAt int64
+	if err := scan(&swf.ID, &swf.FileID, &swf.AppID, &trackingURL, &swf.ClusterID, &swf.ExitCode, &swf.DurationMS, &submittedAt, &swf.FileName); err != nil {
+		return SubmissionWithFile{}, err
+	}
+	if trackingURL.Valid {
+		swf.TrackingURL = trackingURL.String
+	}
+	swf.SubmittedAt = time.Unix(0, submittedAt)
+	return swf, nil
+}

internal/storage/submission_repo_test.go

@@ -0,0 +1,151 @@
+package storage
+
+import (
+	"context"
+	"testing"
+	"time"
+)
+
+func setupSubmissionTest(t *testing.T) (*SubmissionRepo, *UploadRepo, context.Context) {
+	db, err := Open(":memory:")
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+	return db.Submissions(), db.Uploads(), context.Background()
+}
+
+func TestSubmissionRepo_RoundTrip(t *testing.T) {
+	repo, uploadRepo, ctx := setupSubmissionTest(t)
+
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000001", "alpha.py", 10, "deadbeef", time.Now()); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+
+	s := Submission{
+		FileID:      "00000000000000000000000000000001",
+		AppID:       "application_1_0001",
+		TrackingURL: "http://rm.example.com:8088/proxy/application_1_0001/",
+		ClusterID:   "cluster-a",
+		ExitCode:    0,
+		DurationMS:  1234,
+		SubmittedAt: time.Unix(0, time.Now().UnixNano()),
+	}
+	if err := repo.Create(ctx, s); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+
+	list, err := repo.List(ctx, SubmissionFilter{Limit: 10})
+	if err != nil {
+		t.Fatalf("list: %v", err)
+	}
+	if len(list) != 1 {
+		t.Fatalf("len(list)=%d, want 1", len(list))
+	}
+	swf := list[0]
+	if swf.FileName != "alpha.py" {
+		t.Errorf("file_name=%q, want alpha.py", swf.FileName)
+	}
+	if swf.AppID != s.AppID {
+		t.Errorf("app_id=%q, want %q", swf.AppID, s.AppID)
+	}
+
+	id := swf.ID
+	got, err := repo.Get(ctx, id)
+	if err != nil {
+		t.Fatalf("get: %v", err)
+	}
+	if got.ID != id || got.AppID != s.AppID {
+		t.Errorf("got=%+v, want id=%d app_id=%s", got, id, s.AppID)
+	}
+
+	if err := repo.Delete(ctx, id); err != nil {
+		t.Fatalf("delete: %v", err)
+	}
+	if _, err := repo.Get(ctx, id); !isErrNotFound(err) {
+		t.Errorf("expected ErrNotFound after delete, got %v", err)
+	}
+}
+
+func TestSubmissionRepo_SearchByAppID(t *testing.T) {
+	repo, uploadRepo, ctx := setupSubmissionTest(t)
+
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000001", "alpha.py", 10, "deadbeef", time.Now()); err != nil {
+		t.Fatalf("create upload alpha: %v", err)
+	}
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000002", "beta.py", 10, "cafebabe", time.Now()); err != nil {
+		t.Fatalf("create upload beta: %v", err)
+	}
+	if err := repo.Create(ctx, Submission{FileID: "00000000000000000000000000000001", AppID: "application_alpha_1", ClusterID: "c", ExitCode: 0, DurationMS: 1, SubmittedAt: time.Now()}); err != nil {
+		t.Fatalf("create alpha: %v", err)
+	}
+	if err := repo.Create(ctx, Submission{FileID: "00000000000000000000000000000002", AppID: "application_beta_1", ClusterID: "c", ExitCode: 0, DurationMS: 1, SubmittedAt: time.Now()}); err != nil {
+		t.Fatalf("create beta: %v", err)
+	}
+
+	list, err := repo.List(ctx, SubmissionFilter{Search: "alpha", Limit: 10})
+	if err != nil {
+		t.Fatalf("search: %v", err)
+	}
+	if len(list) != 1 || list[0].AppID != "application_alpha_1" || list[0].FileName != "alpha.py" {
+		t.Errorf("got %+v, want 1 alpha submission", list)
+	}
+}
+
+func TestSubmissionRepo_SearchByFileName(t *testing.T) {
+	db, err := Open(":memory:")
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	defer db.Close()
+	repo := db.Submissions()
+	uploadRepo := db.Uploads()
+	ctx := context.Background()
+
+	if err := uploadRepo.Create(ctx, "00000000000000000000000000000001", "report.py", 10, "deadbeef", time.Now()); err != nil {
+		t.Fatalf("create upload: %v", err)
+	}
+	if err := repo.Create(ctx, Submission{FileID: "00000000000000000000000000000001", AppID: "application_x_1", ClusterID: "c", ExitCode: 0, DurationMS: 1, SubmittedAt: time.Now()}); err != nil {
+		t.Fatalf("create: %v", err)
+	}
+
+	list, err := repo.List(ctx, SubmissionFilter{Search: "report", Limit: 10})
+	if err != nil {
+		t.Fatalf("search: %v", err)
+	}
+	if len(list) != 1 || list[0].FileName != "report.py" {
+		t.Errorf("got %+v, want 1 report submission", list)
+	}
+}
+
+func TestSubmissionRepo_FilterByFileID(t *testing.T) {
+	db, err := Open(":memory:")
+	if err != nil {
+		t.Fatalf("open: %v", err)
+	}
+	defer db.Close()
+	repo := db.Submissions()
+	ctx := context.Background()
+
+	if err := repo.Create(ctx, Submission{FileID: "f1", AppID: "a1", ClusterID: "c", ExitCode: 0, DurationMS: 1, SubmittedAt: time.Now()}); err != nil {
+		t.Fatalf("create f1: %v", err)
+	}
+	if err := repo.Create(ctx, Submission{FileID: "f2", AppID: "a2", ClusterID: "c", ExitCode: 0, DurationMS: 1, SubmittedAt: time.Now()}); err != nil {
+		t.Fatalf("create f2: %v", err)
+	}
+
+	list, err := repo.List(ctx, SubmissionFilter{FileID: "f1", Limit: 10})
+	if err != nil {
+		t.Fatalf("list: %v", err)
+	}
+	if len(list) != 1 || list[0].FileID != "f1" {
+		t.Errorf("got %+v, want f1 submission", list)
+	}
+}
+
+func isErrNotFound(err error) bool {
+	if err == nil {
+		return false
+	}
+	return err.Error() == ErrNotFound.Error()
+}

internal/storage/upload_repo.go

@@ -0,0 +1,149 @@
+package storage
+
+import (
+	"context"
+	"database/sql"
+	"errors"
+	"fmt"
+	"strings"
+	"time"
+)
+
+// UploadMeta is the database index record for an uploaded file.
+//
+// The .meta.json sidecar remains the source of truth; this struct is the
+// queryable mirror used by the admin UI.
+type UploadMeta struct {
+	FileID     string    `json:"file_id"`
+	Name       string    `json:"name"`
+	Size       int64     `json:"size"`
+	SHA256     string    `json:"sha256"`
+	UploadedAt time.Time `json:"uploaded_at"`
+}
+
+// UploadRepo provides CRUD operations for upload file index records.
+type UploadRepo struct {
+	db *DB
+}
+
+// Uploads returns a repository bound to this DB handle.
+func (d *DB) Uploads() *UploadRepo {
+	return &UploadRepo{db: d}
+}
+
+// Create inserts a new upload index record. Timestamps are stored as Unix
+// nanoseconds for stable ORDER BY semantics.
+func (r *UploadRepo) Create(ctx context.Context, fileID, name string, size int64, sha256Hex string, uploadedAt time.Time) error {
+	_, err := r.db.sqlDB.ExecContext(ctx, `
+		INSERT INTO upload_files (file_id, name, size, sha256, uploaded_at)
+		VALUES (?, ?, ?, ?, ?)`,
+		fileID, name, size, sha256Hex, uploadedAt.UnixNano(),
+	)
+	if err != nil {
+		if strings.Contains(err.Error(), "UNIQUE constraint failed") {
+			return fmt.Errorf("storage: create upload %s: primary key conflict: %w", fileID, err)
+		}
+		return fmt.Errorf("storage: create upload %s: %w", fileID, err)
+	}
+	return nil
+}
+
+// Get retrieves an upload index record by file_id. If the record does not
+// exist, it returns ErrNotFound.
+func (r *UploadRepo) Get(ctx context.Context, fileID string) (UploadMeta, error) {
+	row := r.db.sqlDB.QueryRowContext(ctx, `
+		SELECT file_id, name, size, sha256, uploaded_at
+		FROM upload_files
+		WHERE file_id = ?`, fileID)
+
+	m, err := scanUpload(row.Scan)
+	if errors.Is(err, sql.ErrNoRows) {
+		return UploadMeta{}, ErrNotFound
+	}
+	if err != nil {
+		return UploadMeta{}, fmt.Errorf("storage: get upload %s: %w", fileID, err)
+	}
+	return m, nil
+}
+
+// List returns upload index records ordered by uploaded_at descending.
+//
+// If search is non-empty, file_name or any associated spark_submit app_id is
+// filtered with a case-insensitive LIKE.
+// A limit of zero or less falls back to 100; values above 500 are capped.
+func (r *UploadRepo) List(ctx context.Context, search string, limit int) ([]UploadMeta, error) {
+	if limit <= 0 {
+		limit = 100
+	}
+	if limit > 500 {
+		limit = 500
+	}
+
+	var rows *sql.Rows
+	var err error
+	if search != "" {
+		rows, err = r.db.sqlDB.QueryContext(ctx, `
+			SELECT DISTINCT u.file_id, u.name, u.size, u.sha256, u.uploaded_at
+			FROM upload_files u
+			LEFT JOIN spark_submissions s ON u.file_id = s.file_id
+			WHERE u.name LIKE ? OR s.app_id LIKE ?
+			ORDER BY u.uploaded_at DESC
+			LIMIT ?`,
+			"%"+search+"%", "%"+search+"%", limit,
+		)
+	} else {
+		rows, err = r.db.sqlDB.QueryContext(ctx, `
+			SELECT u.file_id, u.name, u.size, u.sha256, u.uploaded_at
+			FROM upload_files u
+			ORDER BY u.uploaded_at DESC
+			LIMIT ?`, limit)
+	}
+	if err != nil {
+		return nil, fmt.Errorf("storage: list uploads: %w", err)
+	}
+	defer rows.Close()
+
+	var out []UploadMeta
+	for rows.Next() {
+		m, err := scanUpload(rows.Scan)
+		if err != nil {
+			return nil, fmt.Errorf("storage: list uploads: %w", err)
+		}
+		out = append(out, m)
+	}
+	if err := rows.Err(); err != nil {
+		return nil, fmt.Errorf("storage: list uploads: %w", err)
+	}
+	if out == nil {
+		// Force empty slice (not null) in JSON.
+		out = []UploadMeta{}
+	}
+	return out, nil
+}
+
+// Delete removes an upload index record by file_id. If the record does not
+// exist, it returns ErrNotFound.
+func (r *UploadRepo) Delete(ctx context.Context, fileID string) error {
+	res, err := r.db.sqlDB.ExecContext(ctx, `DELETE FROM upload_files WHERE file_id = ?`, fileID)
+	if err != nil {
+		return fmt.Errorf("storage: delete upload %s: %w", fileID, err)
+	}
+	n, err := res.RowsAffected()
+	if err != nil {
+		return fmt.Errorf("storage: delete upload %s: rows affected: %w", fileID, err)
+	}
+	if n == 0 {
+		return ErrNotFound
+	}
+	return nil
+}
+
+func scanUpload(scan scanFunc) (UploadMeta, error) {
+	var m UploadMeta
+	var uploadedAt int64
+	if err := scan(&m.FileID, &m.Name, &m.Size, &m.SHA256, &uploadedAt); err != nil {
+		return UploadMeta{}, err
+	}
+	m.UploadedAt = time.Unix(0, uploadedAt)
+	return m, nil
+}

internal/storage/upload_repo_test.go

@@ -0,0 +1,162 @@
+package storage
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"testing"
+	"time"
+)
+
+func newUploadRepo(t *testing.T) *UploadRepo {
+	t.Helper()
+	db, err := Open(":memory:")
+	if err != nil {
+		t.Fatalf("Open in-memory db: %v", err)
+	}
+	t.Cleanup(func() { _ = db.Close() })
+	return db.Uploads()
+}
+
+func assertUploadEqual(t *testing.T, got, want UploadMeta) {
+	t.Helper()
+	if got.FileID != want.FileID {
+		t.Errorf("FileID: got %q, want %q", got.FileID, want.FileID)
+	}
+	if got.Name != want.Name {
+		t.Errorf("Name: got %q, want %q", got.Name, want.Name)
+	}
+	if got.Size != want.Size {
+		t.Errorf("Size: got %d, want %d", got.Size, want.Size)
+	}
+	if got.SHA256 != want.SHA256 {
+		t.Errorf("SHA256: got %q, want %q", got.SHA256, want.SHA256)
+	}
+	if !got.UploadedAt.Equal(want.UploadedAt) {
+		t.Errorf("UploadedAt: got %v, want %v", got.UploadedAt, want.UploadedAt)
+	}
+}
+
+func TestUploadRepo_CreateGetListDelete(t *testing.T) {
+	repo := newUploadRepo(t)
+	ctx := context.Background()
+	uploadedAt := time.Unix(0, time.Now().UnixNano())
+	want := UploadMeta{
+		FileID:     "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
+		Name:       "report.csv",
+		Size:       42,
+		SHA256:     "deadbeef",
+		UploadedAt: uploadedAt,
+	}
+	if err := repo.Create(ctx, want.FileID, want.Name, want.Size, want.SHA256, want.UploadedAt); err != nil {
+		t.Fatalf("Create: %v", err)
+	}
+	got, err := repo.Get(ctx, want.FileID)
+	if err != nil {
+		t.Fatalf("Get: %v", err)
+	}
+	assertUploadEqual(t, got, want)
+	if err := repo.Create(ctx, want.FileID, "other", 1, "abcd", time.Now()); err == nil {
+		t.Errorf("duplicate Create succeeded, want error")
+	}
+	list, err := repo.List(ctx, "", 100)
+	if err != nil {
+		t.Fatalf("List: %v", err)
+	}
+	if len(list) != 1 {
+		t.Fatalf("List len: got %d, want 1", len(list))
+	}
+	assertUploadEqual(t, list[0], want)
+	if err := repo.Delete(ctx, want.FileID); err != nil {
+		t.Fatalf("Delete: %v", err)
+	}
+	if _, err := repo.Get(ctx, want.FileID); !errors.Is(err, ErrNotFound) {
+		t.Fatalf("Get after Delete: got %v, want ErrNotFound", err)
+	}
+	if err := repo.Delete(ctx, want.FileID); !errors.Is(err, ErrNotFound) {
+		t.Fatalf("Delete again: got %v, want ErrNotFound", err)
+	}
+}
+
+func TestUploadRepo_ListOrderAndSearch(t *testing.T) {
+	repo := newUploadRepo(t)
+	ctx := context.Background()
+	base := time.Unix(0, time.Now().UnixNano())
+	records := []UploadMeta{
+		{FileID: "00000000000000000000000000000001", Name: "alpha.csv", Size: 1, SHA256: "a", UploadedAt: base.Add(-2 * time.Hour)},
+		{FileID: "00000000000000000000000000000002", Name: "beta.log", Size: 2, SHA256: "b", UploadedAt: base.Add(-1 * time.Hour)},
+		{FileID: "00000000000000000000000000000003", Name: "gamma.csv", Size: 3, SHA256: "c", UploadedAt: base},
+	}
+	for _, r := range records {
+		if err := repo.Create(ctx, r.FileID, r.Name, r.Size, r.SHA256, r.UploadedAt); err != nil {
+			t.Fatalf("Create %s: %v", r.FileID, err)
+		}
+	}
+	all, err := repo.List(ctx, "", 100)
+	if err != nil {
+		t.Fatalf("List: %v", err)
+	}
+	if len(all) != 3 {
+		t.Fatalf("List len: got %d, want 3", len(all))
+	}
+	wantOrder := []string{"00000000000000000000000000000003", "00000000000000000000000000000002", "00000000000000000000000000000001"}
+	for i, id := range wantOrder {
+		if all[i].FileID != id {
+			t.Errorf("List order[%d]: got %q, want %q", i, all[i].FileID, id)
+		}
+	}
+	csv, err := repo.List(ctx, "csv", 100)
+	if err != nil {
+		t.Fatalf("List search: %v", err)
+	}
+	if len(csv) != 2 {
+		t.Fatalf("search csv len: got %d, want 2", len(csv))
+	}
+	if csv[0].Name != "gamma.csv" || csv[1].Name != "alpha.csv" {
+		t.Errorf("search csv order: got %v", []string{csv[0].Name, csv[1].Name})
+	}
+	beta, err := repo.List(ctx, "beta", 100)
+	if err != nil {
+		t.Fatalf("List search beta: %v", err)
+	}
+	if len(beta) != 1 || beta[0].Name != "beta.log" {
+		t.Errorf("search beta: got %+v", beta)
+	}
+}
+
+func TestUploadRepo_ListLimitCap(t *testing.T) {
+	repo := newUploadRepo(t)
+	ctx := context.Background()
+	base := time.Unix(0, time.Now().UnixNano())
+	for i := 0; i < 10; i++ {
+		id := fmt.Sprintf("%032d", i)
+		if err := repo.Create(ctx, id, "x", int64(i), "h", base.Add(time.Duration(i)*time.Second)); err != nil {
+			t.Fatalf("Create %d: %v", i, err)
+		}
+	}
+	list, err := repo.List(ctx, "", 0)
+	if err != nil {
+		t.Fatalf("List default limit: %v", err)
+	}
+	// Default limit is 100 but only 10 rows exist.
+	if len(list) != 10 {
+		t.Errorf("default limit: got %d, want 10", len(list))
+	}
+
+	list, err = repo.List(ctx, "", 501)
+	if err != nil {
+		t.Fatalf("List cap: %v", err)
+	}
+	if len(list) != 10 {
+		t.Errorf("cap limit: got %d, want 10", len(list))
+	}
+
+	// Explicit small limit is respected.
+	list, err = repo.List(ctx, "", 3)
+	if err != nil {
+		t.Fatalf("List small limit: %v", err)
+	}
+	if len(list) != 3 {
+		t.Errorf("small limit: got %d, want 3", len(list))
+	}
+}

internal/uploads/store.go

@@ -9,12 +9,16 @@ import (
 	"crypto/sha256"
 	"encoding/hex"
 	"encoding/json"
+	"errors"
 	"fmt"
+	"log/slog"
 	"os"
 	"path/filepath"
 	"regexp"
 	"strings"
 	"time"
+
+	"spark-mcp-go/internal/storage"
 )
 
 var fileIDRegex = regexp.MustCompile(`^[0-9a-f]{32}$`)
@@ -22,6 +26,20 @@ var fileIDRegex = regexp.MustCompile(`^[0-9a-f]{32}$`)
 // Store is a local file store rooted at Root.
 type Store struct {
 	Root string
+	repo UploadsDB
+}
+
+// UploadsDB is the minimal surface the Store needs from an upload index.
+type UploadsDB interface {
+	Create(ctx context.Context, fileID, name string, size int64, sha256Hex string, uploadedAt time.Time) error
+	Get(ctx context.Context, fileID string) (storage.UploadMeta, error)
+}
+
+// SetRepo binds an upload index repository for dual-write. The DB is an
+// index only; failures are logged but never fail the file write because
+// .meta.json is the source of truth and startup backfill can recover.
+func (s *Store) SetRepo(repo UploadsDB) {
+	s.repo = repo
 }
 
 // New creates a Store rooted at root. It creates root with mode 0o750 if it
@@ -111,6 +129,13 @@ func (s *Store) Save(data []byte, originalName string) (fileID, name string, siz
 		return "", "", 0, "", "", fmt.Errorf("uploads: write sidecar: %w", err)
 	}
 
+	// Index the upload in the DB as a best-effort mirror of the sidecar.
+	if s.repo != nil {
+		if dbErr := s.repo.Create(context.Background(), fileID, originalName, meta.Size, sha256Hex, meta.UploadedAt); dbErr != nil {
+			slog.Default().Warn("uploads: failed to index upload in DB", "file_id", fileID, "err", dbErr)
+		}
+	}
+
 	return fileID, originalName, meta.Size, sha256Hex, absPath, nil
 }
 
@@ -240,6 +265,54 @@ func (s *Store) Sweep(ctx context.Context, ttl time.Duration) (deleted int, err
 	return deleted, nil
 }
 
+// Backfill walks Root and inserts a DB index row for every existing
+// .meta.json sidecar that is not already indexed. It is the recovery path
+// for sidecars created before the DB table existed.
+func (s *Store) Backfill(ctx context.Context, repo UploadsDB) (inserted int, err error) {
+	entries, err := os.ReadDir(s.Root)
+	if err != nil {
+		return 0, fmt.Errorf("uploads: backfill read root: %w", err)
+	}
+
+	for _, e := range entries {
+		if err := ctx.Err(); err != nil {
+			return inserted, err
+		}
+
+		name := e.Name()
+		base := strings.TrimSuffix(name, ".meta.json")
+		if base == name || !fileIDRegex.MatchString(base) {
+			continue
+		}
+
+		metaPath := filepath.Join(s.Root, name)
+		sc, err := readSidecar(metaPath)
+		if err != nil {
+			// Skip corrupt sidecars; Sweep will reap them later.
+			continue
+		}
+
+		_, getErr := repo.Get(ctx, base)
+		if getErr == nil {
+			continue
+		}
+		if !errors.Is(getErr, storage.ErrNotFound) {
+			return inserted, fmt.Errorf("uploads: backfill lookup %s: %w", base, getErr)
+		}
+
+		createErr := repo.Create(ctx, base, sc.Name, sc.Size, sc.Sha256, sc.UploadedAt)
+		if createErr != nil {
+			if strings.Contains(createErr.Error(), "UNIQUE constraint failed") {
+				continue
+			}
+			return inserted, fmt.Errorf("uploads: backfill index %s: %w", base, createErr)
+		}
+		inserted++
+	}
+
+	return inserted, nil
+}
+
 func newFileID() (string, error) {
 	var b [16]byte
 	if _, err := rand.Read(b[:]); err != nil {
@@ -249,16 +322,24 @@ func newFileID() (string, error) {
 }
 
 func readUploadedAt(path string) (time.Time, error) {
-	data, err := os.ReadFile(path)
+	sc, err := readSidecar(path)
 	if err != nil {
 		return time.Time{}, err
 	}
-	var sc sidecar
-	if err := json.Unmarshal(data, &sc); err != nil {
-		return time.Time{}, err
-	}
 	if sc.UploadedAt.IsZero() {
 		return time.Time{}, fmt.Errorf("uploads: missing uploaded_at")
 	}
 	return sc.UploadedAt, nil
 }
+
+func readSidecar(path string) (sidecar, error) {
+	var sc sidecar
+	data, err := os.ReadFile(path)
+	if err != nil {
+		return sc, err
+	}
+	if err := json.Unmarshal(data, &sc); err != nil {
+		return sc, err
+	}
+	return sc, nil
+}

internal/uploads/store_test.go

@@ -5,12 +5,15 @@ import (
 	"crypto/sha256"
 	"encoding/hex"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"os"
 	"path/filepath"
 	"regexp"
 	"testing"
 	"time"
+
+	"spark-mcp-go/internal/storage"
 )
 
 func TestStore_SaveAndRetrieve(t *testing.T) {
@@ -351,3 +354,180 @@ func TestStore_Sweep_DeletesDataWithCorruptSidecar(t *testing.T) {
 		t.Errorf("corrupt sidecar still exists")
 	}
 }
+
+type fakeUploadRepo struct {
+	calls []storage.UploadMeta
+	err   error
+}
+
+func (f *fakeUploadRepo) Create(ctx context.Context, fileID, name string, size int64, sha256Hex string, uploadedAt time.Time) error {
+	f.calls = append(f.calls, storage.UploadMeta{
+		FileID: fileID, Name: name, Size: size, SHA256: sha256Hex, UploadedAt: uploadedAt,
+	})
+	return f.err
+}
+
+func (f *fakeUploadRepo) Get(ctx context.Context, fileID string) (storage.UploadMeta, error) {
+	for _, m := range f.calls {
+		if m.FileID == fileID {
+			return m, nil
+		}
+	}
+	return storage.UploadMeta{}, storage.ErrNotFound
+}
+
+func (f *fakeUploadRepo) List(ctx context.Context, search string, limit int) ([]storage.UploadMeta, error) {
+	return f.calls, nil
+}
+
+func (f *fakeUploadRepo) Delete(ctx context.Context, fileID string) error {
+	return nil
+}
+
+func TestStore_Save_DualWrite(t *testing.T) {
+	store, err := New(t.TempDir())
+	if err != nil {
+		t.Fatalf("new store: %v", err)
+	}
+	repo := &fakeUploadRepo{}
+	store.SetRepo(repo)
+
+	data := []byte("dual-write test")
+	fileID, _, size, sha256Hex, absPath, err := store.Save(data, "dual.txt")
+	if err != nil {
+		t.Fatalf("save: %v", err)
+	}
+	if absPath == "" {
+		t.Fatal("absPath empty")
+	}
+	if len(repo.calls) != 1 {
+		t.Fatalf("repo.Create calls: got %d, want 1", len(repo.calls))
+	}
+	call := repo.calls[0]
+	if call.FileID != fileID {
+		t.Errorf("FileID: got %q, want %q", call.FileID, fileID)
+	}
+	if call.Name != "dual.txt" {
+		t.Errorf("Name: got %q, want dual.txt", call.Name)
+	}
+	if call.Size != size {
+		t.Errorf("Size: got %d, want %d", call.Size, size)
+	}
+	if call.SHA256 != sha256Hex {
+		t.Errorf("SHA256: got %q, want %q", call.SHA256, sha256Hex)
+	}
+	if call.UploadedAt.IsZero() {
+		t.Errorf("UploadedAt is zero")
+	}
+}
+
+func TestStore_Save_RepoErrorDoesNotFailUpload(t *testing.T) {
+	store, err := New(t.TempDir())
+	if err != nil {
+		t.Fatalf("new store: %v", err)
+	}
+	repo := &fakeUploadRepo{err: errors.New("db down")}
+	store.SetRepo(repo)
+
+	data := []byte("db error test")
+	fileID, _, _, _, absPath, err := store.Save(data, "error.txt")
+	if err != nil {
+		t.Fatalf("save should not fail when repo errors: %v", err)
+	}
+	if _, err := os.Stat(absPath); err != nil {
+		t.Errorf("data file missing: %v", err)
+	}
+	if _, err := os.Stat(absPath + ".meta.json"); err != nil {
+		t.Errorf("sidecar missing: %v", err)
+	}
+	if len(repo.calls) != 1 {
+		t.Errorf("repo.Create calls: got %d, want 1", len(repo.calls))
+	}
+	if repo.calls[0].FileID != fileID {
+		t.Errorf("repo call file_id: got %q, want %q", repo.calls[0].FileID, fileID)
+	}
+}
+
+func TestStore_Backfill(t *testing.T) {
+	store, err := New(t.TempDir())
+	if err != nil {
+		t.Fatalf("new store: %v", err)
+	}
+	repo := &fakeUploadRepo{}
+
+	now := time.Now()
+	ids := []string{
+		"00000000000000000000000000000001",
+		"00000000000000000000000000000002",
+		"00000000000000000000000000000003",
+	}
+	for i, id := range ids {
+		sc := map[string]any{
+			"name":        fmt.Sprintf("file%d.txt", i),
+			"size":        i + 1,
+			"sha256":      fmt.Sprintf("sha%d", i),
+			"uploaded_at": now.Add(time.Duration(i) * time.Second).Format(time.RFC3339Nano),
+		}
+		b, _ := json.Marshal(sc)
+		if err := os.WriteFile(filepath.Join(store.Root, id+".meta.json"), b, 0o600); err != nil {
+			t.Fatalf("create sidecar %s: %v", id, err)
+		}
+	}
+
+	inserted, err := store.Backfill(context.Background(), repo)
+	if err != nil {
+		t.Fatalf("backfill: %v", err)
+	}
+	if inserted != 3 {
+		t.Errorf("inserted=%d, want 3", inserted)
+	}
+	if len(repo.calls) != 3 {
+		t.Errorf("repo.Create calls: got %d, want 3", len(repo.calls))
+	}
+
+	// Second backfill must skip existing rows.
+	inserted, err = store.Backfill(context.Background(), repo)
+	if err != nil {
+		t.Fatalf("second backfill: %v", err)
+	}
+	if inserted != 0 {
+		t.Errorf("second inserted=%d, want 0", inserted)
+	}
+	if len(repo.calls) != 3 {
+		t.Errorf("repo.Create calls after second backfill: got %d, want 3", len(repo.calls))
+	}
+}
+
+func TestStore_Backfill_SkipsExisting(t *testing.T) {
+	store, err := New(t.TempDir())
+	if err != nil {
+		t.Fatalf("new store: %v", err)
+	}
+	repo := &fakeUploadRepo{}
+
+	id := "00000000000000000000000000000004"
+	now := time.Now()
+	sc := map[string]any{
+		"name":        "existing.txt",
+		"size":        5,
+		"sha256":      "sha",
+		"uploaded_at": now.Format(time.RFC3339Nano),
+	}
+	b, _ := json.Marshal(sc)
+	if err := os.WriteFile(filepath.Join(store.Root, id+".meta.json"), b, 0o600); err != nil {
+		t.Fatalf("create sidecar: %v", err)
+	}
+
+	// Pre-seed the repo so the sidecar is already indexed.
+	if err := repo.Create(context.Background(), id, "existing.txt", 5, "sha", now); err != nil {
+		t.Fatalf("seed repo: %v", err)
+	}
+
+	inserted, err := store.Backfill(context.Background(), repo)
+	if err != nil {
+		t.Fatalf("backfill: %v", err)
+	}
+	if inserted != 0 {
+		t.Errorf("inserted=%d, want 0", inserted)
+	}
+}

main.go

@@ -84,6 +84,19 @@ func run() error {
 		logger.Info("uploads.sweep", "deleted", n)
 	}
 
+	backfillCtx, backfillCancel := context.WithTimeout(context.Background(), startupSweepTimeout)
+	n, err = uploadStore.Backfill(backfillCtx, db.Uploads())
+	backfillCancel()
+	if errors.Is(err, context.DeadlineExceeded) {
+		logger.Warn("uploads.backfill.timeout", "inserted", n, "err", err)
+	} else if err != nil {
+		logger.Error("uploads.backfill", "err", err)
+	} else {
+		logger.Info("uploads.backfill", "inserted", n)
+	}
+
+	uploadStore.SetRepo(db.Uploads())
+
 	gin.SetMode(cfg.GinMode)
 	r := gin.New()
 	r.Use(gin.Recovery())
@@ -92,10 +105,11 @@ func run() error {
 		c.JSON(http.StatusOK, gin.H{"ok": true, "version": "0.0.0"})
 	})
 
-	admin.Mount(r, db.Clusters(), audit.NewRepo(db), cfg.AdminTokens)
+	admin.Mount(r, db.Clusters(), db.Uploads(), db.Submissions(), audit.NewRepo(db), &uploadStore, cfg.AdminTokens)
 
 	mcpHandler, err := mcpsrv.Handler(&tools.Deps{
 		Logger:             logger.With("component", "mcp"),
+		AuditRepo:          audit.NewRepo(db),
 		ClusterRepo:        db.Clusters(),
 		SparkSubmitTimeout: cfg.SparkSubmitTimeout,
 		HTTPClient: httpclient.New(httpclient.Config{
@@ -105,6 +119,7 @@ func run() error {
 		MaxResponseBytes: cfg.MaxResponseBytes,
 		DataDir:          cfg.DataDir,
 		UploadStore:      uploadStore,
+		SubmissionRepo:   db.Submissions(),
 		AnalyzerThresholds: analyzer.Thresholds{
 			DataSkewRatio:       cfg.AnalyzerDataSkewRatio,
 			GCPressureRatio:     cfg.AnalyzerGCPressureRatio,