diff --git a/agent/tools.go b/agent/tools.go
index cf78dcd..f30cdd5 100644
--- a/agent/tools.go
+++ b/agent/tools.go
@@ -58,7 +58,7 @@ type ToolSet struct {
 	ownerNode int64
 	GoalMet   bool
 	Reason    string
-	wrote     int
+	writes    WriteCounts
 	// killWork, if set, terminates a running work by intent id (engine callback,
 	// wired by the planner). nil = the kill_work tool reports unavailable.
 	killWork func(intentID int64) error
@@ -86,9 +86,29 @@ type EnrichTrigger interface {
 	ProbeSite(id int64, url string)
 }
 
-// Wrote reports how many facts this run wrote back (so the engine can tell
-// "explored but persisted nothing" apart from a completed intent).
-func (t *ToolSet) Wrote() int { return t.wrote }
+// WriteCounts breaks down what a worker persisted this run, by node kind, so the
+// engine can log an accurate "wrote back" summary instead of lumping assets and
+// findings under "facts" (record_fact → Facts, insert_assets → Assets,
+// report_finding → Findings; each element of a batch counts once).
+type WriteCounts struct {
+	Facts    int
+	Assets   int
+	Findings int
+}
+
+// Total is every node persisted this run, regardless of kind — the
+// "explored but persisted nothing" signal (Total == 0).
+func (w WriteCounts) Total() int { return w.Facts + w.Assets + w.Findings }
+
+// String renders the per-kind breakdown for logs, e.g. "事实1 资产25 漏洞0".
+func (w WriteCounts) String() string {
+	return fmt.Sprintf("事实%d 资产%d 漏洞%d", w.Facts, w.Assets, w.Findings)
+}
+
+// Writes reports what this run wrote back, split by node kind (so the engine can
+// tell "explored but persisted nothing" apart from a completed intent, and log an
+// honest breakdown instead of calling assets/findings "facts").
+func (t *ToolSet) Writes() WriteCounts { return t.writes }
 
 func NewToolSet(ts *db.ExplorationStore, worker string) *ToolSet {
 	return &ToolSet{ts: ts, worker: worker}
@@ -594,7 +614,7 @@ func (t *ToolSet) addFinding() actool.CoreTool {
 				// conversation context: no exploration store available, cannot record finding
 				return actool.Errorf("report_finding 需要任务上下文（exploration store 未初始化）"), nil
 			}
-			t.wrote++
+			t.writes.Findings++
 			return actool.Text(fmt.Sprintf("finding recorded: %d", id)), nil
 		})
 }
@@ -642,7 +662,7 @@ func (t *ToolSet) recordOneFact(it factItem, defaultIntent int64) (int64, error)
 	if intent > 0 {
 		_ = t.ts.Link(intent, db.RelYields, id) // chain: intent -> fact
 	}
-	t.wrote++
+	t.writes.Facts++
 	return id, nil
 }
 
diff --git a/agent/tools_insert.go b/agent/tools_insert.go
index ba7c7c4..f1df3b0 100644
--- a/agent/tools_insert.go
+++ b/agent/tools_insert.go
@@ -272,7 +272,7 @@ func (t *ToolSet) insertAssets() actool.CoreTool {
 					continue
 				}
 				results = append(results, result{Index: i, ID: id, Type: typ})
-				t.wrote++
+				t.writes.Assets++
 				t.anchorOwner(id)
 			}
 
diff --git a/agent/worker.go b/agent/worker.go
index c476889..2370ce7 100644
--- a/agent/worker.go
+++ b/agent/worker.go
@@ -193,9 +193,10 @@ func renderWorkerGraphOverview(data map[string]any) string {
 // Execute runs one intent. hooks (the per-task Guard) gates every tool call; may
 // be nil. emit, if non-nil, receives one ActivityRecord per execution step.
 // Returns the terminal reason (so the engine can distinguish completed vs
-// max_turns) and how many facts were written back (so an intent that explored but
-// persisted nothing isn't mistaken for done).
-func (w *Worker) Execute(ctx context.Context, name string, taskID int64, as *db.AssetStore, ts *db.ExplorationStore, intent *db.Node, hooks harness.HookRunner, emit func(db.Activity), enr EnrichTrigger) (harness.TerminalReason, int, error) {
+// max_turns) and a per-kind breakdown of what was written back (so an intent that
+// explored but persisted nothing isn't mistaken for done, and the engine can log
+// facts/assets/findings separately instead of lumping them under "facts").
+func (w *Worker) Execute(ctx context.Context, name string, taskID int64, as *db.AssetStore, ts *db.ExplorationStore, intent *db.Node, hooks harness.HookRunner, emit func(db.Activity), enr EnrichTrigger) (harness.TerminalReason, WriteCounts, error) {
 	tsx := NewToolSet(ts, name)
 	tsx.SetTaskID(taskID)
 	if as != nil {
@@ -298,5 +299,5 @@ func (w *Worker) Execute(ctx context.Context, name string, taskID int64, as *db.
 		defer runCancel()
 	}
 	_, reason, err := captureRunSession(runCtx, s, input, emitWrap)
-	return reason, tsx.Wrote(), err
+	return reason, tsx.Writes(), err
 }
diff --git a/server/engine.go b/server/engine.go
index 690b2bd..9694911 100644
--- a/server/engine.go
+++ b/server/engine.go
@@ -532,7 +532,7 @@ func (e *Engine) workerLoop(ctx context.Context, t *Task, name string) {
 		// 不要误标 blocked。此时 worker 通常已在 settlement 阶段把结果写回。
 		if ectx.Err() != nil && e.isSettling(t.ID) {
 			_ = t.Store.SetIntentState(intent.ID, "exhausted")
-			log.Printf("[worker %s] task %s 意图 #%d 因任务超时收尾结束(exhausted)，写回 %d 条", name, t.ID, intent.ID, wrote)
+			log.Printf("[worker %s] task %s 意图 #%d 因任务超时收尾结束(exhausted)，写回 %s", name, t.ID, intent.ID, wrote)
 			e.touch(t.ID)
 			continue
 		}
@@ -563,13 +563,13 @@ func (e *Engine) workerLoop(ctx context.Context, t *Task, name string) {
 			state = "blocked"
 		case reason == harness.ReasonMaxTurns:
 			state = "exhausted"
-			log.Printf("[worker %s] intent %d 撞步数上限(exhausted)，本次写回事实 %d 条", name, intent.ID, wrote)
+			log.Printf("[worker %s] intent %d 撞步数上限(exhausted)，本次写回 %s", name, intent.ID, wrote)
 		case reason == harness.ReasonTimeout:
 			state = "exhausted"
-			log.Printf("[worker %s] intent %d 运行超时(exhausted)，收尾后写回事实 %d 条", name, intent.ID, wrote)
+			log.Printf("[worker %s] intent %d 运行超时(exhausted)，收尾后写回 %s", name, intent.ID, wrote)
 		}
 		_ = t.Store.SetIntentState(intent.ID, state)
-		log.Printf("[worker %s] task %s 意图 #%d 结束: %s (写回 %d 条事实)", name, t.ID, intent.ID, state, wrote)
+		log.Printf("[worker %s] task %s 意图 #%d 结束: %s (写回 %s)", name, t.ID, intent.ID, state, wrote)
 		e.touch(t.ID)
 		t.NotifyDone(intent.ID) // results changed the graph -> wake the planner (with the just-finished intent id)
 	}