Add worker stats to the wizard, artifact pruning and offline-worker removal to the admin console

coordinator/cmd/coordinator/main.go

@@ -173,6 +173,7 @@ func runWithConfig(cfg infra.Config) error {
 		GetJobResult:     usecase.NewGetJobResult(jobRepo, usecase.NewDownloadArtifact(artifactRepo, blobStore)),
 		GetTaskInput:     usecase.NewGetTaskInput(taskRepo, artifactRepo, blobStore),
 		Dashboard:        usecase.NewDashboard(uiReadRepo, catalog),
+		PruneArtifacts:   usecase.NewPruneArtifacts(jobRepo, uiReadRepo, blobStore, clk),
 		PreviewArtifact:  usecase.NewPreviewArtifact(uiReadRepo, blobStore),
 		Admin: usecase.NewAdmin(deps.adminReadRepo, uiReadRepo, workerRepo, deps.settingsRepo, catalog,
 			usecase.AdminNodeInfo{

coordinator/internal/agent/daemon.go

@@ -158,6 +158,7 @@ func (d *Daemon) runOnce() (Outcome, error) {
 	if task == nil {
 		return Outcome{Claimed: false}, nil
 	}
+	d.log.Info("task claimed", "task_id", task.TaskID, "attempt", task.Attempt)
 	started := time.Now()
 	taskDir := filepath.Join(d.config.WorkDir, task.TaskID, fmt.Sprint(task.Attempt))
 	if err := os.MkdirAll(taskDir, 0o750); err != nil {

coordinator/internal/agent/setupui/server.go

@@ -289,15 +289,26 @@ func writeJSON(w http.ResponseWriter, status int, v any) {
 
 // statusView is what the wizard needs to paint the running/stopped state.
 type statusView struct {
-	ConfigPresent bool   `json:"config_present"`
-	ConfigPath    string `json:"config_path"`
-	LogPath       string `json:"log_path"`
-	Running       bool   `json:"running"`
-	Pid           int    `json:"pid"`
-	WorkerName    string `json:"worker_name,omitempty"`
-	Coordinator   string `json:"coordinator,omitempty"`
-	WorkDir       string `json:"work_dir,omitempty"`
-	TokenSet      bool   `json:"token_set"`
+	ConfigPresent bool        `json:"config_present"`
+	ConfigPath    string      `json:"config_path"`
+	LogPath       string      `json:"log_path"`
+	Running       bool        `json:"running"`
+	Pid           int         `json:"pid"`
+	WorkerName    string      `json:"worker_name,omitempty"`
+	Coordinator   string      `json:"coordinator,omitempty"`
+	WorkDir       string      `json:"work_dir,omitempty"`
+	TokenSet      bool        `json:"token_set"`
+	Stats         WorkerStats `json:"stats"`
+}
+
+// WorkerStats is parsed from the worker log: the agent reports each claim,
+// completion and failure as a structured line, so the wizard can show live
+// counters without any coordinator access.
+type WorkerStats struct {
+	Registered bool `json:"registered"`
+	Claimed    int  `json:"claimed"`
+	Completed  int  `json:"completed"`
+	Failed     int  `json:"failed"`
 }
 
 // ensureVenvTaskRunner rewrites the saved config so its task runner uses the
@@ -319,8 +330,30 @@ func (s *Server) ensureVenvTaskRunner() {
 	}
 }
 
+// parseWorkerStats counts the structured agent events in the worker log.
+func parseWorkerStats(logPath string) WorkerStats {
+	var stats WorkerStats
+	raw, err := os.ReadFile(logPath)
+	if err != nil {
+		return stats
+	}
+	for _, line := range strings.Split(string(raw), "\n") {
+		switch {
+		case strings.Contains(line, "msg=registered"):
+			stats.Registered = true
+		case strings.Contains(line, `msg="task claimed"`):
+			stats.Claimed++
+		case strings.Contains(line, `msg="task completed"`):
+			stats.Completed++
+		case strings.Contains(line, `msg="task failed"`):
+			stats.Failed++
+		}
+	}
+	return stats
+}
+
 func (s *Server) handleStatus(w http.ResponseWriter, r *http.Request) {
-	view := statusView{ConfigPath: s.cfgPath, LogPath: s.logPath, Running: s.sup.Alive(), Pid: s.sup.Pid()}
+	view := statusView{ConfigPath: s.cfgPath, LogPath: s.logPath, Running: s.sup.Alive(), Pid: s.sup.Pid(), Stats: parseWorkerStats(s.logPath)}
 	if raw, err := os.ReadFile(s.cfgPath); err == nil {
 		var file agent.ConfigFile
 		if json.Unmarshal(raw, &file) == nil {

coordinator/internal/agent/setupui/server_test.go

@@ -479,3 +479,21 @@ func TestTestProbesTheVenvPythonAfterInstall(t *testing.T) {
 		t.Errorf("scimesh check = %+v, want the venv interpreter reporting 9.9.9-test", report.Scimesh)
 	}
 }
+
+func TestParseWorkerStats(t *testing.T) {
+	logPath := filepath.Join(t.TempDir(), "worker.log")
+	content := `time=1 level=INFO msg=registered worker_id=w1
+time=2 level=INFO msg="task claimed" task_id=t1 attempt=0
+time=3 level=INFO msg="task completed" task_id=t1 elapsed_seconds=2
+time=4 level=INFO msg="task claimed" task_id=t2 attempt=0
+time=5 level=WARN msg="task failed" task_id=t2 error_code=X retryable=true
+time=6 level=WARN msg="agent cycle failed" error="boom"
+`
+	if err := os.WriteFile(logPath, []byte(content), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	stats := parseWorkerStats(logPath)
+	if !stats.Registered || stats.Claimed != 2 || stats.Completed != 1 || stats.Failed != 1 {
+		t.Errorf("stats = %+v, want registered claimed=2 completed=1 failed=1", stats)
+	}
+}

coordinator/internal/agent/setupui/template.html

@@ -174,6 +174,12 @@ code{font-family:var(--mono);font-size:.86em}
         <div><h1 id="st-title">Worker is working</h1><div class="sub" id="st-sub">—</div></div>
         <button class="btn btn-danger" id="st-stop" style="margin-left:auto"><svg viewBox="0 0 24 24" fill="none" stroke-width="2"><rect x="6" y="6" width="12" height="12" rx="2"/></svg>Stop</button>
       </div>
+      <div class="stat-row">
+        <div class="stat"><b id="st-claimed">—</b><span>claimed</span></div>
+        <div class="stat"><b id="st-completed">—</b><span>completed</span></div>
+        <div class="stat bad"><b id="st-failed">—</b><span>failed</span></div>
+        <div class="stat"><b id="st-registered">—</b><span>registered</span></div>
+      </div>
       <div class="meta-line" id="st-meta"></div>
       <div class="logbox" id="st-log"></div>
       <div class="actions"><span class="link" id="st-cfg">—</span><button class="btn btn-ghost" id="st-reconfig">Reconfigure…</button></div>
@@ -293,6 +299,11 @@ async function refreshStatus(){
   $('st-title').textContent=v.running?(v.worker_name||'Worker')+' is working':(v.worker_name||'Worker')+' is stopped';
   $('st-sub').textContent='pid '+(v.pid||'—')+' · config '+(v.config_present?v.config_path:'not saved yet');
   $('st-cfg').textContent='Configuration: '+(v.config_present?v.config_path:'—');
+  const stats=v.stats||{};
+  $('st-claimed').textContent=stats.claimed!=null?stats.claimed:'—';
+  $('st-completed').textContent=stats.completed!=null?stats.completed:'—';
+  $('st-failed').textContent=stats.failed!=null?stats.failed:'—';
+  $('st-registered').textContent=stats.registered?'yes':'no';
   const meta=$('st-meta');
   meta.innerHTML='';
   if(v.coordinator){const c=document.createElement('span');c.className='chip';c.innerHTML='<svg viewBox="0 0 24 24" fill="none" stroke-width="2"><circle cx="12" cy="12" r="9"/></svg>'+v.coordinator;meta.append(c)}

coordinator/internal/memstore/memstore.go

@@ -461,3 +461,32 @@ func (r *TaskResultRepo) CountAgreeing(_ context.Context, taskID uuid.UUID, sha2
 	}
 	return n, nil
 }
+
+func (r *JobRepo) ListCompletedBefore(ctx context.Context, cutoff time.Time) ([]domain.Job, error) {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	var out []domain.Job
+	for _, j := range r.jobs {
+		if j.CompletedAt != nil && j.CompletedAt.Before(cutoff) {
+			out = append(out, *j)
+		}
+	}
+	return out, nil
+}
+
+func (r *JobRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	delete(r.jobs, id)
+	return nil
+}
+
+func (r *WorkerRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	r.mu.Lock()
+	defer r.mu.Unlock()
+	if _, ok := r.workers[id]; !ok {
+		return domain.ErrWorkerNotFound
+	}
+	delete(r.workers, id)
+	return nil
+}

coordinator/internal/storage/postgres/job_repo.go

@@ -3,6 +3,7 @@ package postgres
 import (
 	"context"
 	"errors"
+	"fmt"
 	"time"
 
 	sq "github.com/Masterminds/squirrel"
@@ -153,3 +154,46 @@ func (r *JobRepo) UpdateStatus(ctx context.Context, id uuid.UUID,
 	}
 	return nil
 }
+
+// ListCompletedBefore returns jobs whose completion timestamp is older than
+// the cutoff (completed and failed both count as finished).
+func (r *JobRepo) ListCompletedBefore(ctx context.Context, cutoff time.Time) ([]domain.Job, error) {
+	sql, args, err := psql.Select(jobColumns...).From("jobs").
+		Where(sq.NotEq{"completed_at": nil}).
+		Where(sq.Lt{"completed_at": cutoff}).
+		OrderBy("completed_at ASC").
+		ToSql()
+	if err != nil {
+		return nil, err
+	}
+	rows, err := conn(ctx, r.pool).Query(ctx, sql, args...)
+	if err != nil {
+		return nil, fmt.Errorf("list completed jobs: %w", err)
+	}
+	defer rows.Close()
+	var jobs []domain.Job
+	for rows.Next() {
+		var j domain.Job
+		var status string
+		if err := rows.Scan(
+			&j.ID, &j.Workload, &j.InputURI, &j.Parameters, &status, &j.CreatedAt, &j.CompletedAt,
+			&j.InputArtifactID, &j.ResultArtifactID, &j.ErrorCode, &j.ErrorMessage, &j.ReducerStartedAt,
+			&j.OwnerID,
+		); err != nil {
+			return nil, err
+		}
+		j.Status = domain.JobStatus(status)
+		jobs = append(jobs, j)
+	}
+	return jobs, rows.Err()
+}
+
+// Delete removes the job row; tasks, artifacts and task_results cascade.
+func (r *JobRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	sql, args, err := psql.Delete("jobs").Where(sq.Eq{"id": id}).ToSql()
+	if err != nil {
+		return err
+	}
+	_, err = conn(ctx, r.pool).Exec(ctx, sql, args...)
+	return err
+}

coordinator/internal/storage/postgres/worker_repo.go

@@ -123,3 +123,19 @@ func scanWorker(row pgx.Row) (*domain.Worker, error) {
 	w.TrustLevel = domain.WorkerTrust(trust)
 	return &w, nil
 }
+
+// Delete removes a worker from the registry.
+func (r *WorkerRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	sql, args, err := psql.Delete("workers").Where(sq.Eq{"id": id}).ToSql()
+	if err != nil {
+		return err
+	}
+	tag, err := conn(ctx, r.pool).Exec(ctx, sql, args...)
+	if err != nil {
+		return fmt.Errorf("delete worker: %w", err)
+	}
+	if tag.RowsAffected() == 0 {
+		return domain.ErrWorkerNotFound
+	}
+	return nil
+}

coordinator/internal/storage/sqlite/admin_m2_test.go

@@ -88,3 +88,47 @@ func TestWorkerSetTrust(t *testing.T) {
 		t.Errorf("unknown worker trust err = %v, want ErrWorkerNotFound", err)
 	}
 }
+
+func TestJobRepoListCompletedBeforeAndDelete(t *testing.T) {
+	db := newTestDB(t)
+	ctx := context.Background()
+	repo := NewJobRepo(db)
+
+	old := seedJob(t, db, 2)
+	oldTime := fixedTime().Add(-40 * 24 * time.Hour)
+	if err := repo.UpdateStatus(ctx, old.ID, domain.JobCompleted, &oldTime); err != nil {
+		t.Fatal(err)
+	}
+	fresh := seedJob(t, db, 2)
+	freshTime := fixedTime().Add(-2 * time.Hour)
+	if err := repo.UpdateStatus(ctx, fresh.ID, domain.JobCompleted, &freshTime); err != nil {
+		t.Fatal(err)
+	}
+	// The failing check constraint needs no result artifact for completed; the
+	// UpdateStatus path is fine, but tasks stay pending — irrelevant here.
+
+	list, err := repo.ListCompletedBefore(ctx, fixedTime().Add(-7*24*time.Hour))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(list) != 1 || list[0].ID != old.ID {
+		t.Errorf("list = %d jobs, want only the old one", len(list))
+	}
+	if err := repo.Delete(ctx, old.ID); err != nil {
+		t.Fatal(err)
+	}
+	var n int
+	if err := db.QueryRowContext(ctx, "SELECT COUNT(*) FROM jobs WHERE id = ?", old.ID.String()).Scan(&n); err != nil {
+		t.Fatal(err)
+	}
+	if n != 0 {
+		t.Error("job row must be gone after Delete")
+	}
+	// Tasks cascaded away with the job.
+	if err := db.QueryRowContext(ctx, "SELECT COUNT(*) FROM tasks WHERE job_id = ?", old.ID.String()).Scan(&n); err != nil {
+		t.Fatal(err)
+	}
+	if n != 0 {
+		t.Error("tasks must cascade with the job")
+	}
+}

coordinator/internal/storage/sqlite/job_repo.go

@@ -3,6 +3,7 @@ package sqlite
 import (
 	"context"
 	"database/sql"
+	"fmt"
 	"time"
 
 	"github.com/google/uuid"
@@ -166,3 +167,30 @@ func nullableUUID(id *uuid.UUID) any {
 	}
 	return id.String()
 }
+
+// ListCompletedBefore returns jobs whose completion timestamp is older than
+// the cutoff (completed and failed both count as finished).
+func (r *JobRepo) ListCompletedBefore(ctx context.Context, cutoff time.Time) ([]domain.Job, error) {
+	rows, err := conn(ctx, r.db).QueryContext(ctx,
+		"SELECT "+jobColumns+" FROM jobs WHERE completed_at IS NOT NULL AND completed_at < ? ORDER BY completed_at ASC",
+		encodeTime(cutoff))
+	if err != nil {
+		return nil, fmt.Errorf("list completed jobs: %w", err)
+	}
+	defer func() { _ = rows.Close() }()
+	var jobs []domain.Job
+	for rows.Next() {
+		job, err := scanJob(rows)
+		if err != nil {
+			return nil, err
+		}
+		jobs = append(jobs, *job)
+	}
+	return jobs, rows.Err()
+}
+
+// Delete removes the job row; tasks, artifacts and task_results cascade.
+func (r *JobRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	_, err := conn(ctx, r.db).ExecContext(ctx, "DELETE FROM jobs WHERE id = ?", id.String())
+	return err
+}

coordinator/internal/storage/sqlite/worker_repo.go

@@ -107,3 +107,19 @@ func (r *WorkerRepo) SetTrust(ctx context.Context, id uuid.UUID, trust domain.Wo
 	}
 	return nil
 }
+
+// Delete removes a worker from the registry.
+func (r *WorkerRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	res, err := conn(ctx, r.db).ExecContext(ctx, "DELETE FROM workers WHERE id = ?", id.String())
+	if err != nil {
+		return err
+	}
+	affected, err := res.RowsAffected()
+	if err != nil {
+		return err
+	}
+	if affected == 0 {
+		return domain.ErrWorkerNotFound
+	}
+	return nil
+}

coordinator/internal/transport/http/server.go

@@ -36,6 +36,7 @@ type UseCases struct {
 	Dashboard        *usecase.Dashboard
 	PreviewArtifact  *usecase.PreviewArtifact
 	Admin            *usecase.Admin
+	PruneArtifacts   *usecase.PruneArtifacts
 }
 
 type Server struct {
@@ -184,6 +185,7 @@ func (s *Server) Handler(token string, uiToken ...string) http.Handler {
 			ui.Handle("GET /ui/admin/api/metrics", chain(http.HandlerFunc(s.handleUIAdminMetricsJSON), gate, requireAdmin))
 			ui.Handle("GET /ui/admin/api/workers", chain(http.HandlerFunc(s.handleUIAdminWorkersJSON), gate, requireAdmin))
 			ui.Handle("POST /ui/admin/api/workers/{id}/trust", chain(http.HandlerFunc(s.handleUIAdminSetTrustJSON), gate, requireAdmin))
+			ui.Handle("POST /ui/admin/api/workers/{id}/remove", chain(http.HandlerFunc(s.handleUIAdminRemoveWorkerJSON), gate, requireAdmin))
 			ui.Handle("GET /ui/admin/api/users", chain(http.HandlerFunc(s.handleUIAdminUsersJSON), gate, requireAdmin))
 			ui.Handle("POST /ui/admin/api/users/{id}/role", chain(http.HandlerFunc(s.handleUIAdminSetUserRoleJSON), gate, requireAdmin))
 			ui.Handle("GET /ui/admin/api/worker-keys", chain(http.HandlerFunc(s.handleUIAdminWorkerKeysJSON), gate, requireAdmin))
@@ -192,6 +194,7 @@ func (s *Server) Handler(token string, uiToken ...string) http.Handler {
 			ui.Handle("POST /ui/admin/api/workloads/{name}/enabled", chain(http.HandlerFunc(s.handleUIAdminSetWorkloadEnabledJSON), gate, requireAdmin))
 			ui.Handle("GET /ui/admin/api/settings", chain(http.HandlerFunc(s.handleUIAdminSettingsJSON), gate, requireAdmin))
 			ui.Handle("POST /ui/admin/api/token/reveal", chain(http.HandlerFunc(s.handleUIAdminRevealTokenJSON), gate, requireAdmin))
+			ui.Handle("POST /ui/admin/api/prune", chain(http.HandlerFunc(s.handleUIAdminPruneJSON), gate, requireAdmin))
 		} else {
 			for _, rt := range app {
 				ui.HandleFunc(rt.pattern, rt.handler)

coordinator/internal/transport/http/templates/admin.html

@@ -210,7 +210,7 @@ tbody tr:hover{background:var(--panel-2)}
     <div class="section-note">Workers register themselves. Trust decides whether a machine's results are accepted directly or need quorum.</div>
     <div class="card">
       <table>
-        <thead><tr><th>Worker</th><th>Status</th><th>Capabilities</th><th>Trust</th><th>Owner</th><th>Last signal</th></tr></thead>
+        <thead><tr><th>Worker</th><th>Status</th><th>Capabilities</th><th>Trust</th><th>Owner</th><th>Last signal</th><th></th></tr></thead>
         <tbody id="worker-rows"></tbody>
       </table>
     </div>
@@ -284,6 +284,12 @@ tbody tr:hover{background:var(--panel-2)}
   <!-- ═══ SETTINGS ═══ -->
   <section class="page" id="page-settings">
     <div class="warn-strip" style="display:flex;gap:10px;align-items:flex-start;background:var(--amber-soft);border:1px solid #e5b64f33;border-radius:10px;padding:12px 14px;font-size:12.5px;color:#eecf8d"><span>The cluster token below authenticates <b>any</b> worker. Reveal it only on a trusted machine.</span></div>
+    <div class="section-title" style="color:var(--red)">Danger zone</div>
+    <div class="card" style="border-color:#f2647c33">
+      <dl class="kv">
+        <dt>Prune artifacts</dt><dd style="display:flex;justify-content:space-between;align-items:center;gap:14px"><span style="color:var(--text-2);font-size:12.5px">Delete finished jobs older than a cutoff and all their artifacts (blob files included).</span><span style="display:flex;gap:8px;align-items:center"><input id="prune-days" type="number" min="1" max="3650" value="30" style="width:80px"><button class="btn btn-danger btn-sm" id="prune">Prune…</button></span></dd>
+      </dl>
+    </div>
     <div class="section-title">Cluster</div>
     <div class="card">
       <dl class="kv">
@@ -440,9 +446,15 @@ async function loadWorkers(){
       '<td>'+(w.capabilities||[]).map(c=>'<span class="cap">'+esc(c)+'</span>').join('')+'</td>'+
       '<td>'+trustSel+'</td>'+
       '<td style="color:var(--text-2)">'+esc(w.owner)+'</td>'+
-      '<td style="color:var(--text-2);font-size:12.5px">'+fmtTime(w.last_heartbeat_at)+'</td>';
+      '<td style="color:var(--text-2);font-size:12.5px">'+fmtTime(w.last_heartbeat_at)+'</td>'+
+      '<td>'+(w.status==='offline'?'<button class="btn btn-danger btn-sm worker-remove" data-id="'+w.id+'" data-name="'+esc(w.name)+'">Remove</button>':'')+'</td>';
     rows.append(tr);
   }
+  document.querySelectorAll('.worker-remove').forEach(btn=>btn.addEventListener('click',async()=>{
+    if(!confirm('Remove the offline worker "'+btn.dataset.name+'" from the registry? This cannot be undone.'))return;
+    await fetch('/ui/admin/api/workers/'+btn.dataset.id+'/remove',{method:'POST'});
+    loadWorkers();
+  }));
   document.querySelectorAll('.trust-sel').forEach(sel=>sel.addEventListener('change',async()=>{
     await fetch('/ui/admin/api/workers/'+sel.dataset.id+'/trust',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({trusted:sel.value==='trusted'})});
     loadWorkers();
@@ -530,6 +542,17 @@ document.getElementById('reveal').addEventListener('click',async e=>{
   }else{tok.textContent='••••••••••••••••••••••••';e.target.textContent='Reveal'}
 });
 document.querySelectorAll('.toggle').forEach(t=>t.addEventListener('click',()=>t.classList.toggle('on')));
+document.getElementById('prune').addEventListener('click',async()=>{
+  const days=parseInt(document.getElementById('prune-days').value||'30',10);
+  if(!confirm('Delete finished jobs older than '+days+' days with all their artifacts? This cannot be undone.'))return;
+  const btn=document.getElementById('prune');btn.disabled=true;btn.textContent='Pruning…';
+  const r=await fetch('/ui/admin/api/prune',{method:'POST',headers:{'Content-Type':'application/json'},body:JSON.stringify({older_than_days:days})});
+  btn.disabled=false;btn.textContent='Prune…';
+  if(!r.ok){alert('Prune failed.');return}
+  const v=await r.json();
+  alert('Removed '+v.jobs+' jobs and '+v.artifacts+' artifacts, freed '+fmtBytes(v.freed_bytes)+'.');
+  if(current.page==='settings')loadSettings();
+});
 </script>
 </body>
 </html>

coordinator/internal/transport/http/ui_admin.go

@@ -8,6 +8,7 @@ import (
 	"net/url"
 	"strconv"
 	"strings"
+	"time"
 
 	"github.com/google/uuid"
 
@@ -390,3 +391,43 @@ func (s *Server) adminOwnerEmails(r *http.Request) map[uuid.UUID]string {
 	}
 	return out
 }
+
+// handleUIAdminPruneJSON deletes finished jobs older than the requested
+// number of days (with all their artifacts) and reports what was freed.
+func (s *Server) handleUIAdminPruneJSON(w http.ResponseWriter, r *http.Request) {
+	var body struct {
+		OlderThanDays int `json:"older_than_days"`
+	}
+	if err := decodeJSON(r, &body); err != nil {
+		s.writeError(w, r, domain.ErrInvalidInput)
+		return
+	}
+	if body.OlderThanDays < 1 || body.OlderThanDays > 3650 {
+		s.writeError(w, r, domain.ErrInvalidInput)
+		return
+	}
+	ctx, cancel := s.reqCtx(r)
+	defer cancel()
+	result, err := s.uc.PruneArtifacts.Execute(ctx, time.Duration(body.OlderThanDays)*24*time.Hour)
+	if err != nil {
+		s.writeError(w, r, err)
+		return
+	}
+	writeJSON(w, http.StatusOK, result)
+}
+
+// handleUIAdminRemoveWorkerJSON deletes an offline worker.
+func (s *Server) handleUIAdminRemoveWorkerJSON(w http.ResponseWriter, r *http.Request) {
+	id, err := uuid.Parse(r.PathValue("id"))
+	if err != nil {
+		s.writeError(w, r, domain.ErrInvalidInput)
+		return
+	}
+	ctx, cancel := s.reqCtx(r)
+	defer cancel()
+	if err := s.uc.Admin.RemoveWorker(ctx, id); err != nil {
+		s.writeError(w, r, err)
+		return
+	}
+	w.WriteHeader(http.StatusNoContent)
+}

coordinator/internal/usecase/admin.go

@@ -521,3 +521,20 @@ func (a *Admin) RevealWorkerToken(ctx context.Context, actor string) string {
 	}
 	return token
 }
+
+// RemoveWorker deletes an offline worker from the registry. Online or busy
+// workers are refused: an admin console must never yank a live machine out
+// from under a running task.
+func (a *Admin) RemoveWorker(ctx context.Context, id uuid.UUID) error {
+	if a.workers == nil {
+		return domain.ErrWorkerNotFound
+	}
+	worker, err := a.workers.Get(ctx, id)
+	if err != nil {
+		return err
+	}
+	if worker.Status != domain.WorkerOffline {
+		return domain.ErrInvalidInput
+	}
+	return a.workers.Delete(ctx, id)
+}

coordinator/internal/usecase/admin_test.go

@@ -336,3 +336,30 @@ func TestAdminRevealToken(t *testing.T) {
 		t.Errorf("token = %q", got)
 	}
 }
+
+type removableWorkerRepo struct {
+	WorkerRepository
+	deleted uuid.UUID
+}
+
+func (f *removableWorkerRepo) Get(ctx context.Context, id uuid.UUID) (*domain.Worker, error) {
+	return &domain.Worker{ID: id, Status: domain.WorkerOffline}, nil
+}
+
+func (f *removableWorkerRepo) Delete(ctx context.Context, id uuid.UUID) error {
+	f.deleted = id
+	return nil
+}
+
+func TestAdminRemoveWorker(t *testing.T) {
+	a := adminFixture()
+	repo := &removableWorkerRepo{}
+	a.workers = repo
+	id := uuid.New()
+	if err := a.RemoveWorker(context.Background(), id); err != nil {
+		t.Fatal(err)
+	}
+	if repo.deleted != id {
+		t.Error("offline worker must be deleted")
+	}
+}

coordinator/internal/usecase/ports.go

@@ -82,6 +82,12 @@ type JobRepository interface {
 	ClaimReduction(ctx context.Context, id uuid.UUID, startedAt time.Time) (bool, error)
 	CompleteWithResult(ctx context.Context, id, resultArtifactID uuid.UUID, completedAt time.Time) error
 	FailReduction(ctx context.Context, id uuid.UUID, code, message string, completedAt time.Time) error
+	// ListCompletedBefore returns jobs that finished (completed or failed)
+	// before the cutoff, for the admin artifact pruner.
+	ListCompletedBefore(ctx context.Context, cutoff time.Time) ([]domain.Job, error)
+	// Delete removes a job row; the engine cascades its tasks, artifacts and
+	// quorum votes. Blob files must be removed separately.
+	Delete(ctx context.Context, id uuid.UUID) error
 }
 
 // WorkerRepository persists the worker registry.
@@ -97,6 +103,9 @@ type WorkerRepository interface {
 	// SetTrust reclassifies a worker's trust level (trusted/untrusted). Returns
 	// ErrNotFound when the id is unknown.
 	SetTrust(ctx context.Context, id uuid.UUID, trust domain.WorkerTrust) error
+	// Delete removes a worker from the registry. Returns ErrNotFound when the
+	// id is unknown.
+	Delete(ctx context.Context, id uuid.UUID) error
 }
 
 // ArtifactRepository persists artifact metadata. The bytes live in a BlobStore;

coordinator/internal/usecase/prune.go

@@ -0,0 +1,55 @@
+package usecase
+
+import (
+	"context"
+	"time"
+)
+
+// PruneArtifacts removes completed or failed jobs older than `olderThan` and
+// every artifact they own: database rows cascade, blob files are deleted
+// explicitly. It returns what was freed so the admin console can report it.
+type PruneArtifacts struct {
+	jobs  JobRepository
+	read  UIReadRepository
+	blobs BlobStore
+	clk   Clock
+}
+
+func NewPruneArtifacts(jobs JobRepository, read UIReadRepository, blobs BlobStore, clk Clock) *PruneArtifacts {
+	return &PruneArtifacts{jobs: jobs, read: read, blobs: blobs, clk: clk}
+}
+
+type PruneResult struct {
+	Jobs       int   `json:"jobs"`
+	Artifacts  int   `json:"artifacts"`
+	FreedBytes int64 `json:"freed_bytes"`
+}
+
+// Execute deletes finished jobs whose completion timestamp is older than the
+// cutoff. Jobs that are still active are never touched.
+func (uc *PruneArtifacts) Execute(ctx context.Context, olderThan time.Duration) (PruneResult, error) {
+	cutoff := uc.clk.Now().Add(-olderThan)
+	jobs, err := uc.jobs.ListCompletedBefore(ctx, cutoff)
+	if err != nil {
+		return PruneResult{}, err
+	}
+	out := PruneResult{}
+	for _, job := range jobs {
+		artifacts, err := uc.read.ListArtifactsByJob(ctx, job.ID)
+		if err != nil {
+			return out, err
+		}
+		for _, artifact := range artifacts {
+			if err := uc.blobs.Delete(ctx, artifact.StorageKey); err != nil {
+				return out, err
+			}
+			out.FreedBytes += artifact.SizeBytes
+			out.Artifacts++
+		}
+		if err := uc.jobs.Delete(ctx, job.ID); err != nil {
+			return out, err
+		}
+		out.Jobs++
+	}
+	return out, nil
+}

coordinator/internal/usecase/usecase_test.go

@@ -980,3 +980,40 @@ func TestSubmitDatasetRejectsDisabledWorkload(t *testing.T) {
 		t.Fatalf("submit after re-enable: %v", err)
 	}
 }
+
+func TestPruneArtifactsRemovesOldFinishedJobs(t *testing.T) {
+	h := newHarness()
+	old := time.Date(2026, 7, 1, 0, 0, 0, 0, time.UTC)
+	job := &domain.Job{ID: uuid.New(), Workload: "similarity-search", Status: domain.JobCompleted, CreatedAt: old, CompletedAt: &old}
+	if err := h.jobs.Insert(context.Background(), job); err != nil {
+		t.Fatal(err)
+	}
+	art, err := domain.NewArtifact(job.ID, nil, domain.ArtifactFinalResult, "r.csv", "text/csv", old)
+	if err != nil {
+		t.Fatal(err)
+	}
+	art.SetContent("sha", 42)
+	if err := h.arts.Insert(context.Background(), art); err != nil {
+		t.Fatal(err)
+	}
+	// An active job must survive the prune.
+	active := &domain.Job{ID: uuid.New(), Workload: "similarity-search", Status: domain.JobRunning, CreatedAt: old}
+	if err := h.jobs.Insert(context.Background(), active); err != nil {
+		t.Fatal(err)
+	}
+
+	prune := usecase.NewPruneArtifacts(h.jobs, memstore.NewUIReadRepo(h.jobs, h.tasks, h.work, h.arts), h.blobs, h.clk)
+	result, err := prune.Execute(context.Background(), 7*24*time.Hour)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if result.Jobs != 1 || result.Artifacts != 1 || result.FreedBytes != 42 {
+		t.Errorf("prune = %+v, want 1 job / 1 artifact / 42 bytes", result)
+	}
+	if _, err := h.jobs.Get(context.Background(), job.ID); err == nil {
+		t.Error("finished job must be gone")
+	}
+	if _, err := h.jobs.Get(context.Background(), active.ID); err != nil {
+		t.Error("active job must survive the prune")
+	}
+}