diff --git a/mateclaw-server/src/main/java/vip/mate/agent/graph/plan/node/StepExecutionNode.java b/mateclaw-server/src/main/java/vip/mate/agent/graph/plan/node/StepExecutionNode.java index 59c4ca8a..4c33ff59 100644 --- a/mateclaw-server/src/main/java/vip/mate/agent/graph/plan/node/StepExecutionNode.java +++ b/mateclaw-server/src/main/java/vip/mate/agent/graph/plan/node/StepExecutionNode.java @@ -30,6 +30,7 @@ import vip.mate.planning.service.PlanningService; import vip.mate.agent.context.ChatOrigin; import vip.mate.skill.runtime.SkillCatalogRenderer; import vip.mate.tool.builtin.DelegateAgentTool; +import vip.mate.tool.builtin.DelegateAgentTool.ChildResult; import vip.mate.tool.builtin.DelegationContext; import vip.mate.tool.builtin.ToolExecutionContext; @@ -678,7 +679,7 @@ public class StepExecutionNode implements NodeAction { // Seed the delegation context with the plan's REAL conversation id (from // graph state) so the delegated child conversation is parented to it and // stays hidden from the user's conversation list. The ChatOrigin in the - // plan-execute path carries no conversationId, so delegateByAgentId can't + // plan-execute path carries no conversationId, so the delegation can't // derive the parent on its own — we provide it here. boolean seeded = false; if (conversationId != null && !conversationId.isBlank() @@ -687,20 +688,30 @@ public class StepExecutionNode implements NodeAction { DelegationContext.enter(conversationId, Set.of(), conversationId, null, 0); seeded = true; } - String result; + ChildResult childResult = null; + String delegateError = null; try { - result = delegateAgentTool.delegateByAgentId(assignedAgentId, step, chatOrigin); + childResult = delegateAgentTool.delegateByAgentIdStructured(assignedAgentId, step, chatOrigin); } catch (Exception e) { log.error("[StepExecution] Delegated step {} threw: {}", stepIndex, e.getMessage(), e); - result = "[错误] 委派执行异常：" + e.getMessage(); + delegateError = e.getMessage(); } finally { if (seeded) { DelegationContext.exit(); } } - String finalResult = result != null ? result : ""; - boolean failed = finalResult.isEmpty() || finalResult.startsWith("[错误]"); + // Branch on the structured outcome instead of pattern-matching an error + // prefix out of the reply text: a successful child with non-empty content + // is the only "ok" case; blank / error / missing all count as failure. + boolean ok = childResult != null && childResult.success() && !childResult.isBlank(); + String finalResult = ok + ? (childResult.result() != null ? childResult.result() : "") + : "[错误] 委派执行失败：" + (delegateError != null ? delegateError + : childResult != null && childResult.error() != null ? childResult.error() + : childResult != null && childResult.isBlank() ? "子 Agent 返回内容为空" + : "未知错误"); + boolean failed = !ok; if (failed) { planningService.updateSubPlanFailure(planId, stepIndex, finalResult); } else { diff --git a/mateclaw-server/src/main/java/vip/mate/tool/builtin/DelegateAgentTool.java b/mateclaw-server/src/main/java/vip/mate/tool/builtin/DelegateAgentTool.java index 476d461d..275cd1c3 100644 --- a/mateclaw-server/src/main/java/vip/mate/tool/builtin/DelegateAgentTool.java +++ b/mateclaw-server/src/main/java/vip/mate/tool/builtin/DelegateAgentTool.java @@ -13,6 +13,7 @@ import org.springframework.lang.Nullable; import org.springframework.stereotype.Component; import org.springframework.beans.factory.annotation.Value; import vip.mate.agent.AgentService; +import vip.mate.agent.AgentService.ChatResult; import vip.mate.agent.context.ChatOrigin; import vip.mate.agent.delegation.SubagentRegistry; import vip.mate.agent.model.AgentEntity; @@ -242,6 +243,38 @@ public class DelegateAgentTool { return "[错误] 未找到名为「" + agentName + "」的已启用 Agent。" + availableAgentsHint(); } + // Execute via the shared single-task core (registry, relay, broadcast, + // child run). Returns the structured ChildResult plus the child's + // session handle so the parent can follow up on this exact sub-agent. + SingleDelegation sd = executeSingleDelegation(target, task, inheritParentContext, ctx); + ChildResult result = sd.result(); + + String response = result.toToolResponse(target.getName()); + // Surface the child's session handle so the parent can follow up on this + // exact sub-agent (its conversation persists past this call) via + // send_to_subagent, instead of re-spawning a fresh, context-less child. + if (result.success() && sd.childConversationId() != null) { + response += "\n\n[session_id: " + sd.childConversationId() + + " — to follow up with this sub-agent, call send_to_subagent(session_id, message)]"; + } + return response; + } + + /** Carrier for a single-task delegation: the structured child result plus + * the child conversation handle (null when spawning was short-circuited). */ + private record SingleDelegation(ChildResult result, String childConversationId) {} + + /** + * Shared execution core for single-task delegation, used by both the + * LLM-facing {@link #delegateToAgent} tool and the id-based + * {@link #delegateByAgentIdStructured} (per-step plan delegation). Handles + * spawn-pause, child conversation creation, optional parent-context + * inheritance, sub-agent registry, event relay/broadcast, and the child + * run — returning the structured {@link ChildResult} so callers decide how + * to format it (tool string vs. plan step bookkeeping). + */ + private SingleDelegation executeSingleDelegation(AgentEntity target, String task, + Boolean inheritParentContext, ToolContext ctx) { String parentConversationId = resolveParentConversationId(); // Root (human-facing) conversation at the top of the delegation tree. // At depth 0 the immediate parent IS the root; deeper layers carry it @@ -249,14 +282,15 @@ public class DelegateAgentTool { String rootConversationId = DelegationContext.rootConversationId(); if (rootConversationId == null) rootConversationId = parentConversationId; String parentSubagentId = DelegationContext.currentSubagentId(); - int childDepth = depth + 1; + int childDepth = DelegationContext.currentDepth() + 1; // Spawn-pause: short-circuit before creating child state when either the // immediate parent or the root tree is paused, so no conversation rows / // relays / registry entries leak. if ((parentConversationId != null && subagentRegistry.isSpawnPaused(parentConversationId)) || (rootConversationId != null && subagentRegistry.isSpawnPaused(rootConversationId))) { - return "[错误] Spawning paused for this conversation; resume via /api/v1/subagents/spawn-pause"; + return new SingleDelegation(ChildResult.ofError(0, target.getName(), + "Spawning paused for this conversation; resume via /api/v1/subagents/spawn-pause"), null); } String childConversationId = createChildConv(target, parentConversationId); @@ -275,12 +309,12 @@ public class DelegateAgentTool { } log.info("Agent delegation: depth={}, target={}({}), childConv={}, parentConv={}", - depth + 1, target.getName(), target.getId(), childConversationId, parentConversationId); + childDepth, target.getName(), target.getId(), childConversationId, parentConversationId); // Register the live sub-agent first so its stable id rides on every // event. Disposable is null in the synchronous single-task path because - // the executor blocks on AgentService#chat directly — there is no Flux - // subscription to dispose. Interrupts here are best-effort (status flip). + // the executor blocks on AgentService#chatWithUsage directly — there is + // no Flux subscription to dispose. Interrupts here are best-effort (status flip). String subagentId = parentConversationId != null ? subagentRegistry.register(parentConversationId, childConversationId, target.getId(), task, null, parentSubagentId, childDepth, rootConversationId) @@ -326,16 +360,7 @@ public class DelegateAgentTool { broadcastEnd(rootConversationId, childConversationId, target.getName(), result, subagentId, parentSubagentId, childDepth); } - - String response = result.toToolResponse(target.getName()); - // Surface the child's session handle so the parent can follow up on this - // exact sub-agent (its conversation persists past this call) via - // send_to_subagent, instead of re-spawning a fresh, context-less child. - if (result.success() && childConversationId != null) { - response += "\n\n[session_id: " + childConversationId - + " — to follow up with this sub-agent, call send_to_subagent(session_id, message)]"; - } - return response; + return new SingleDelegation(result, childConversationId); } /** @@ -347,12 +372,30 @@ public class DelegateAgentTool { * child's reply text, or an error string when the agent is missing/disabled. */ public String delegateByAgentId(Long agentId, String task, ChatOrigin parentOrigin) { + ChildResult result = delegateByAgentIdStructured(agentId, task, parentOrigin); + return result.toToolResponse(result.agentName()); + } + + /** + * Structured variant of {@link #delegateByAgentId}: runs the same isolated + * child execution but returns the full {@link ChildResult} instead of a + * formatted string. Used by per-step plan delegation so the plan graph can + * branch on {@code success()} / {@code isBlank()} / {@code outcome()} and + * read token usage, rather than pattern-matching an error prefix out of a + * string. Never throws — agent-resolution and depth-guard failures come + * back as {@code outcome="error"} results. + */ + public ChildResult delegateByAgentIdStructured(Long agentId, String task, ChatOrigin parentOrigin) { if (agentId == null) { - return "[错误] 未指定委派 Agent。"; + return ChildResult.ofError(0, "?", "未指定委派 Agent。"); } AgentEntity target = agentMapper.selectById(agentId); if (target == null || !Boolean.TRUE.equals(target.getEnabled())) { - return "[错误] 未找到 id=" + agentId + " 的已启用 Agent。"; + return ChildResult.ofError(0, "id=" + agentId, "未找到 id=" + agentId + " 的已启用 Agent。"); + } + if (DelegationContext.currentDepth() >= MAX_DELEGATION_DEPTH) { + return ChildResult.ofError(0, target.getName(), + "委派层级已达上限（" + MAX_DELEGATION_DEPTH + " 层）"); } ChatOrigin origin = parentOrigin != null ? parentOrigin : ChatOrigin.EMPTY; ToolContext ctx = origin.toToolContext(); @@ -372,7 +415,7 @@ public class DelegateAgentTool { DelegationContext.enter(parentConvId, Set.of(), parentConvId, null, 0); } try { - return delegateToAgent(target.getName(), task, false, ctx); + return executeSingleDelegation(target, task, false, ctx).result(); } finally { if (seedContext) { DelegationContext.exit(); @@ -543,6 +586,8 @@ public class DelegateAgentTool { payload.put("trimmedLength", r.trimmedLength); payload.put("blank", r.isBlank()); payload.put("durationMs", r.durationMs); + payload.put("promptTokens", r.promptTokens); + payload.put("completionTokens", r.completionTokens); payload.put("resultPreview", r.success ? truncate(r.result, 400) : (r.error != null ? r.error : "error")); @@ -637,6 +682,8 @@ public class DelegateAgentTool { m.put("trimmedLength", r.trimmedLength); m.put("blank", r.isBlank()); m.put("durationMs", r.durationMs); + m.put("promptTokens", r.promptTokens); + m.put("completionTokens", r.completionTokens); // childConversationId + subagentId for stable frontend tree lookup prepared.stream() .filter(p -> p.index == r.taskIndex) @@ -666,6 +713,8 @@ public class DelegateAgentTool { long timeoutCount = results.stream().filter(r -> "timeout".equals(r.outcome)).count(); long cancelledCount = results.stream().filter(r -> "cancelled".equals(r.outcome)).count(); long errorCount = results.stream().filter(r -> "error".equals(r.outcome)).count(); + long tokensInTotal = results.stream().mapToLong(r -> r.promptTokens).sum(); + long tokensOutTotal = results.stream().mapToLong(r -> r.completionTokens).sum(); StringBuilder sb = new StringBuilder(); @@ -680,6 +729,8 @@ public class DelegateAgentTool { .append(" cancelled=").append(cancelledCount) .append(" error=").append(errorCount) .append(" durationMs=").append(totalDurationMs) + .append(" tokensIn=").append(tokensInTotal) + .append(" tokensOut=").append(tokensOutTotal) .append("\n\n"); // Important: this result is from the current execution. Any timeout entries in the @@ -700,6 +751,8 @@ public class DelegateAgentTool { .append(" | contentLength=").append(r.trimmedLength).append("chars") .append(" | rawLength=").append(r.rawLength).append("chars") .append(" | duration=").append(r.durationMs / 1000).append("s") + .append(" | tokensIn=").append(r.promptTokens) + .append(" | tokensOut=").append(r.completionTokens) .append("\n\n"); switch (r.outcome) { @@ -1048,11 +1101,17 @@ public class DelegateAgentTool { ChatOrigin childOrigin = (parentOrigin != null ? parentOrigin : ChatOrigin.EMPTY) .withAgent(target.getId()) .withConversationId(childConversationId); - String rawResult = agentService.chat(target.getId(), task, childConversationId, childOrigin); + // chatWithUsage runs the same StateGraph as chat() (so the child + // message is persisted identically) but also surfaces the child's + // token usage from the graph's _usage_final event, so the parent can + // see what each sub-agent cost. + ChatResult chatResult = agentService.chatWithUsage( + target.getId(), task, childConversationId, childOrigin); long durationMs = System.currentTimeMillis() - startTime; + String rawResult = chatResult.content(); // Measure lengths before truncation so ChildResult carries accurate metadata. return ChildResult.ofSuccess(taskIndex, target.getName(), rawResult, durationMs, - MAX_RESULT_LENGTH); + MAX_RESULT_LENGTH, chatResult.promptTokens(), chatResult.completionTokens()); } catch (Exception e) { log.error("Child agent failed: taskIndex={}, agent={}, error={}", taskIndex, target.getName(), e.getMessage()); @@ -1079,21 +1138,25 @@ public class DelegateAgentTool { *

{@code rawLength} and {@code trimmedLength} are measured before truncation and reflect the * true content length. */ - private record ChildResult( + public record ChildResult( int taskIndex, String agentName, boolean success, String result, String error, long durationMs, - /** "success" | "blank_success" | "timeout" | "error" */ + /** "success" | "blank_success" | "timeout" | "cancelled" | "error" */ String outcome, - int rawLength, int trimmedLength) { + int rawLength, int trimmedLength, + /** Child token usage captured from the graph's _usage_final event; + * 0 for non-success outcomes (timeout / error / cancelled). */ + int promptTokens, int completionTokens) { /** Whether the child returned no usable content (blank_success). */ - boolean isBlank() { return "blank_success".equals(outcome); } + public boolean isBlank() { return "blank_success".equals(outcome); } /** * Factory for a successful child execution. * Measures lengths from the raw result before applying the truncation limit. */ - static ChildResult ofSuccess(int idx, String name, String rawResult, long ms, int maxLen) { + static ChildResult ofSuccess(int idx, String name, String rawResult, long ms, int maxLen, + int promptTokens, int completionTokens) { String safe = rawResult != null ? rawResult : ""; String trimmed = safe.trim(); boolean blank = trimmed.isEmpty(); @@ -1102,7 +1165,8 @@ public class DelegateAgentTool { truncate(safe, maxLen), null, ms, blank ? "blank_success" : "success", - safe.length(), trimmed.length()); + safe.length(), trimmed.length(), + Math.max(0, promptTokens), Math.max(0, completionTokens)); } /** @@ -1113,14 +1177,14 @@ public class DelegateAgentTool { String msg = err != null ? err : "Unknown error"; boolean isTimeout = msg.contains("超时") || msg.toLowerCase().contains("timeout"); return new ChildResult(idx, name, false, null, msg, 0, - isTimeout ? "timeout" : "error", 0, 0); + isTimeout ? "timeout" : "error", 0, 0, 0, 0); } /** Factory for an explicit timeout (parallel window exceeded). */ static ChildResult ofTimeout(int idx, String name, int timeoutSec) { String msg = "超时 (" + timeoutSec + "s)"; return new ChildResult(idx, name, false, null, msg, (long) timeoutSec * 1000L, - "timeout", 0, 0); + "timeout", 0, 0, 0, 0); } /** @@ -1130,7 +1194,7 @@ public class DelegateAgentTool { */ static ChildResult ofCancelled(int idx, String name) { return new ChildResult(idx, name, false, null, - "已取消（必需子任务失败，触发提前收束）", 0, "cancelled", 0, 0); + "已取消（必需子任务失败，触发提前收束）", 0, "cancelled", 0, 0, 0, 0); } // Legacy shims — kept for callers that pre-date the factory methods @@ -1140,15 +1204,19 @@ public class DelegateAgentTool { String trimmed = safe.trim(); boolean blank = trimmed.isEmpty(); return new ChildResult(idx, name, true, safe, null, ms, - blank ? "blank_success" : "success", safe.length(), trimmed.length()); + blank ? "blank_success" : "success", safe.length(), trimmed.length(), 0, 0); } static ChildResult error(int idx, String name, String err) { return ofError(idx, name, err); } String toToolResponse(String agentName) { - if (success) return "[Agent「" + agentName + "」的回复]\n\n" + (result != null ? result : ""); - return "[错误] Agent「" + agentName + "」执行失败: " + error; + if (!success) return "[错误] Agent「" + agentName + "」执行失败: " + error; + String body = "[Agent「" + agentName + "」的回复]\n\n" + (result != null ? result : ""); + if (promptTokens > 0 || completionTokens > 0) { + body += "\n\n[usage: tokensIn=" + promptTokens + " tokensOut=" + completionTokens + "]"; + } + return body; } private static String truncate(String text, int maxLength) { @@ -1372,6 +1440,8 @@ public class DelegateAgentTool { Map ev = delegationPayload(subagentId, parentSubagentId, depth, childConvId, agentName); ev.put("success", result.success); ev.put("durationMs", result.durationMs); + ev.put("promptTokens", result.promptTokens); + ev.put("completionTokens", result.completionTokens); ev.put("resultPreview", result.success ? truncate(result.result, 200) : (result.error != null ? result.error : "")); streamTracker.broadcastObject(rootConvId, "delegation_end", ev); diff --git a/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateAgentToolTest.java b/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateAgentToolTest.java index 33be4673..3c379571 100644 --- a/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateAgentToolTest.java +++ b/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateAgentToolTest.java @@ -15,6 +15,7 @@ import org.mockito.Mock; import org.mockito.Spy; import org.mockito.junit.jupiter.MockitoExtension; import vip.mate.agent.AgentService; +import vip.mate.agent.AgentService.ChatResult; import vip.mate.agent.delegation.SubagentRegistry; import vip.mate.agent.model.AgentEntity; import vip.mate.agent.repository.AgentMapper; @@ -180,9 +181,9 @@ class DelegateAgentToolTest { // the test thread returns immediately. The orphan keeps sleeping on a // virtual thread until JVM teardown — that's the same behavior as // production (cancel is best-effort). - when(agentService.chat(anyLong(), anyString(), anyString(), any())).thenAnswer(invocation -> { + when(agentService.chatWithUsage(anyLong(), anyString(), anyString(), any())).thenAnswer(invocation -> { Thread.sleep(10_000); - return "should not reach here"; + return ChatResult.contentOnly("should not reach here"); }); // Set a conversationId so resolveParentConversationId works @@ -239,13 +240,13 @@ class DelegateAgentToolTest { when(streamTracker.isRunning(any())).thenReturn(false); // FastAgent completes immediately - when(agentService.chat(eq(10L), anyString(), anyString(), any())) - .thenReturn("Fast result completed successfully"); + when(agentService.chatWithUsage(eq(10L), anyString(), anyString(), any())) + .thenReturn(ChatResult.contentOnly("Fast result completed successfully")); // SlowAgent blocks longer than the (test-overridden) 3 s budget. - when(agentService.chat(eq(11L), anyString(), anyString(), any())).thenAnswer(invocation -> { + when(agentService.chatWithUsage(eq(11L), anyString(), anyString(), any())).thenAnswer(invocation -> { Thread.sleep(10_000); - return "should not reach here"; + return ChatResult.contentOnly("should not reach here"); }); ToolExecutionContext.set("parent-mixed", "admin"); @@ -286,12 +287,12 @@ class DelegateAgentToolTest { .thenReturn(slowAgent); when(streamTracker.isRunning(any())).thenReturn(false); // Required FailAgent errors immediately → arms fail-fast. - when(agentService.chat(eq(20L), anyString(), anyString(), any())) + when(agentService.chatWithUsage(eq(20L), anyString(), anyString(), any())) .thenThrow(new RuntimeException("boom")); // SlowAgent would block well past the 3 s test budget; fail-fast cancels it. - when(agentService.chat(eq(21L), anyString(), anyString(), any())).thenAnswer(inv -> { + when(agentService.chatWithUsage(eq(21L), anyString(), anyString(), any())).thenAnswer(inv -> { Thread.sleep(10_000); - return "unreachable"; + return ChatResult.contentOnly("unreachable"); }); ToolExecutionContext.set("parent-ff", "admin"); @@ -325,10 +326,10 @@ class DelegateAgentToolTest { .thenReturn(optAgent) .thenReturn(okAgent); when(streamTracker.isRunning(any())).thenReturn(false); - when(agentService.chat(eq(30L), anyString(), anyString(), any())) + when(agentService.chatWithUsage(eq(30L), anyString(), anyString(), any())) .thenThrow(new RuntimeException("opt boom")); - when(agentService.chat(eq(31L), anyString(), anyString(), any())) - .thenReturn("ok result done"); + when(agentService.chatWithUsage(eq(31L), anyString(), anyString(), any())) + .thenReturn(ChatResult.contentOnly("ok result done")); ToolExecutionContext.set("parent-opt", "admin"); String json = "[{\"agentName\":\"OptAgent\",\"task\":\"a\",\"optional\":true}," @@ -354,9 +355,9 @@ class DelegateAgentToolTest { when(agentMapper.selectOne(any(LambdaQueryWrapper.class))).thenReturn(longAgent); when(streamTracker.isRunning(any())).thenReturn(false); // Sleeps 4 s — beyond the 3 s test budget, but within the 6 s override. - when(agentService.chat(eq(40L), anyString(), anyString(), any())).thenAnswer(inv -> { + when(agentService.chatWithUsage(eq(40L), anyString(), anyString(), any())).thenAnswer(inv -> { Thread.sleep(4_000); - return "long task finished ok"; + return ChatResult.contentOnly("long task finished ok"); }); ToolExecutionContext.set("parent-to", "admin"); @@ -368,4 +369,60 @@ class DelegateAgentToolTest { "long task should complete within the widened budget: " + result); assertFalse(result.contains("timeout=1"), "should not time out with the override: " + result); } + + // ===== token usage surfacing ===== + + @Test + @DisplayName("delegateToAgent surfaces the child's token usage in the reply") + void delegateToAgentSurfacesTokenUsage() { + AgentEntity agent = new AgentEntity(); + agent.setId(50L); + agent.setName("Worker"); + agent.setEnabled(true); + agent.setWorkspaceId(1L); + when(agentMapper.selectOne(any(LambdaQueryWrapper.class))).thenReturn(agent); + when(streamTracker.isRunning(any())).thenReturn(false); + when(agentService.chatWithUsage(eq(50L), anyString(), anyString(), any())) + .thenReturn(new ChatResult("done with work", 120, 45, null, null)); + + ToolExecutionContext.set("parent-usage", "admin"); + String result = delegateAgentTool.delegateToAgent("Worker", "do the thing", null, null); + + assertTrue(result.contains("tokensIn=120"), "reply should surface prompt tokens: " + result); + assertTrue(result.contains("tokensOut=45"), "reply should surface completion tokens: " + result); + } + + @Test + @DisplayName("delegateParallel aggregates child token usage in the header and per-row lines") + void delegateParallelAggregatesTokenUsage() { + AgentEntity a = new AgentEntity(); + a.setId(60L); + a.setName("AgentA"); + a.setEnabled(true); + a.setWorkspaceId(1L); + AgentEntity b = new AgentEntity(); + b.setId(61L); + b.setName("AgentB"); + b.setEnabled(true); + b.setWorkspaceId(1L); + when(agentMapper.selectOne(any(LambdaQueryWrapper.class))) + .thenReturn(a) + .thenReturn(b); + when(streamTracker.isRunning(any())).thenReturn(false); + when(agentService.chatWithUsage(eq(60L), anyString(), anyString(), any())) + .thenReturn(new ChatResult("result A", 100, 30, null, null)); + when(agentService.chatWithUsage(eq(61L), anyString(), anyString(), any())) + .thenReturn(new ChatResult("result B", 80, 20, null, null)); + + ToolExecutionContext.set("parent-usage-parallel", "admin"); + String json = "[{\"agentName\":\"AgentA\",\"task\":\"a\"},{\"agentName\":\"AgentB\",\"task\":\"b\"}]"; + String result = delegateAgentTool.delegateParallel(json, null); + + // Machine header carries the batch totals (180 in, 50 out). + assertTrue(result.contains("tokensIn=180"), "header should aggregate prompt tokens: " + result); + assertTrue(result.contains("tokensOut=50"), "header should aggregate completion tokens: " + result); + // Per-row lines carry each child's own usage. + assertTrue(result.contains("tokensIn=100"), "row A should carry its prompt tokens: " + result); + assertTrue(result.contains("tokensIn=80"), "row B should carry its prompt tokens: " + result); + } } diff --git a/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateEventSequenceTest.java b/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateEventSequenceTest.java index f36cabe2..4c734217 100644 --- a/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateEventSequenceTest.java +++ b/mateclaw-server/src/test/java/vip/mate/tool/builtin/DelegateEventSequenceTest.java @@ -16,6 +16,7 @@ import org.mockito.Mock; import org.mockito.junit.jupiter.MockitoExtension; import org.mockito.Spy; import vip.mate.agent.AgentService; +import vip.mate.agent.AgentService.ChatResult; import vip.mate.agent.delegation.SubagentRegistry; import vip.mate.agent.model.AgentEntity; import vip.mate.agent.repository.AgentMapper; @@ -108,14 +109,14 @@ class DelegateEventSequenceTest { }); // During chat(), simulate the child broadcasting a tool_call_started event - when(agentService.chat(eq(100L), eq("summarize the report"), anyString(), any())) + when(agentService.chatWithUsage(eq(100L), eq("summarize the report"), anyString(), any())) .thenAnswer(invocation -> { // The relay listener should have been registered by now — fire it BiConsumer relay = relayRef.get(); assertNotNull(relay, "Relay should be registered before child chat starts"); relay.accept("tool_call_started", "{\"name\":\"searchWeb\"}"); relay.accept("tool_call_completed", "{\"name\":\"searchWeb\",\"success\":true}"); - return "The report shows growth of 15% YoY."; + return ChatResult.contentOnly("The report shows growth of 15% YoY."); }); // Act @@ -170,8 +171,10 @@ class DelegateEventSequenceTest { when(streamTracker.addBatchedEventRelay(anyString(), anyString(), anyInt(), anyLong(), any())) .thenReturn(() -> {}); - when(agentService.chat(eq(101L), anyString(), anyString(), any())).thenReturn("Result A"); - when(agentService.chat(eq(102L), anyString(), anyString(), any())).thenReturn("Result B"); + when(agentService.chatWithUsage(eq(101L), anyString(), anyString(), any())) + .thenReturn(ChatResult.contentOnly("Result A")); + when(agentService.chatWithUsage(eq(102L), anyString(), anyString(), any())) + .thenReturn(ChatResult.contentOnly("Result B")); String json = "[{\"agentName\":\"AgentA\",\"task\":\"task A\"},{\"agentName\":\"AgentB\",\"task\":\"task B\"}]"; @@ -203,7 +206,8 @@ class DelegateEventSequenceTest { ToolExecutionContext.set("inactive-parent", "admin"); when(streamTracker.isRunning("inactive-parent")).thenReturn(false); - when(agentService.chat(eq(200L), anyString(), anyString(), any())).thenReturn("done"); + when(agentService.chatWithUsage(eq(200L), anyString(), anyString(), any())) + .thenReturn(ChatResult.contentOnly("done")); // Act delegateAgentTool.delegateToAgent("QuietAgent", "quiet task", null, null); @@ -235,7 +239,7 @@ class DelegateEventSequenceTest { return (Runnable) () -> {}; }); - when(agentService.chat(eq(300L), anyString(), anyString(), any())) + when(agentService.chatWithUsage(eq(300L), anyString(), anyString(), any())) .thenAnswer(invocation -> { BiConsumer relay = relayRef.get(); // These should produce delegation_progress: @@ -244,7 +248,7 @@ class DelegateEventSequenceTest { // These should be ignored by the relay filter: relay.accept("heartbeat", "{}"); relay.accept("token", "{\"text\":\"hello\"}"); - return "filtered result"; + return ChatResult.contentOnly("filtered result"); }); delegateAgentTool.delegateToAgent("FilterAgent", "filter task", null, null); @@ -294,18 +298,19 @@ class DelegateEventSequenceTest { // ToolExecutionContext to the Child's own conversation. Reproduce that so // the grandchild's immediate parent resolves to childConv, while its // events must still target rootConv (carried via DelegationContext). - when(agentService.chat(eq(100L), anyString(), anyString(), any())) + when(agentService.chatWithUsage(eq(100L), anyString(), anyString(), any())) .thenAnswer(inv -> { String childConv = inv.getArgument(2); ToolExecutionContext.set(childConv, "admin"); try { - return delegateAgentTool.delegateToAgent("Grandchild", "gtask", null, null); + return ChatResult.contentOnly( + delegateAgentTool.delegateToAgent("Grandchild", "gtask", null, null)); } finally { ToolExecutionContext.set(rootConv, "admin"); } }); - when(agentService.chat(eq(200L), anyString(), anyString(), any())) - .thenReturn("grandchild done"); + when(agentService.chatWithUsage(eq(200L), anyString(), anyString(), any())) + .thenReturn(ChatResult.contentOnly("grandchild done")); delegateAgentTool.delegateToAgent("Child", "ctask", null, null); diff --git a/mateclaw-ui/src/components/chat/DelegationNodeView.vue b/mateclaw-ui/src/components/chat/DelegationNodeView.vue index cab4906e..885e0f87 100644 --- a/mateclaw-ui/src/components/chat/DelegationNodeView.vue +++ b/mateclaw-ui/src/components/chat/DelegationNodeView.vue @@ -42,6 +42,13 @@ const progress = computed(() => { return n ? `${n} ${n === 1 ? 'tool' : 'tools'}` : '' }) +// Compact token cost, shown once the subagent finishes (e.g. "3.2k tok"). +const tokenLabel = computed(() => { + const t = (props.node.promptTokens || 0) + (props.node.completionTokens || 0) + if (!t) return '' + return `${t >= 1000 ? (t / 1000).toFixed(1) + 'k' : t} tok` +}) + function stepStatus(i: number): 'pending' | 'running' | 'completed' { const p = plan.value if (!p) return 'pending' @@ -63,6 +70,7 @@ function stepStatus(i: number): 'pending' | 'running' | 'completed' { {{ node.agentName }} {{ progress }} + {{ tokenLabel }} 0) parts.push(`${tok >= 1000 ? (tok / 1000).toFixed(1) + 'k' : tok} tok`) + return parts.length ? ` (${parts.join(' · ')})` : '' + } + function markDelegComplete(segs: MessageSegment[], subagentId: string | undefined, childConvId: string | undefined, - success: boolean, resultPreview?: string, durationMs?: number): boolean { + success: boolean, resultPreview?: string, durationMs?: number, + promptTokens?: number, completionTokens?: number): boolean { const seg = findDelegSegment(segs, subagentId, childConvId) if (seg) { seg.status = success ? 'completed' : 'error' seg.toolSuccess = success if (resultPreview) seg.toolResult = resultPreview - if (durationMs) seg.toolArgs = (seg.toolArgs || '').trimEnd() + ` (${Math.round(durationMs / 1000)}s)` + const suffix = delegMetaSuffix(durationMs, promptTokens, completionTokens) + if (suffix) seg.toolArgs = (seg.toolArgs || '').trimEnd() + suffix return true } if (subagentId) { @@ -1089,6 +1100,8 @@ export function useChat(options: UseChatOptions): UseChatReturn { node.status = success ? 'completed' : 'error' if (resultPreview) node.result = resultPreview if (durationMs) node.durationMs = durationMs + if (promptTokens) node.promptTokens = promptTokens + if (completionTokens) node.completionTokens = completionTokens return true } } @@ -1232,7 +1245,7 @@ export function useChat(options: UseChatOptions): UseChatReturn { if (isStaleEvent(data)) return if (!currentAssistantId.value) return markDelegComplete(currentSegments.value, data.subagentId, data.childConversationId, - !!data.success, data.resultPreview, data.durationMs) + !!data.success, data.resultPreview, data.durationMs, data.promptTokens, data.completionTokens) flushSegmentsToMessage() }) @@ -1250,7 +1263,7 @@ export function useChat(options: UseChatOptions): UseChatReturn { : !!(cr.subagentId && findNode(segs.flatMap(s => s.childTimeline?.children || []), cr.subagentId)?.status === 'running') if (stillRunning) { markDelegComplete(segs, cr.subagentId, cr.childConversationId, !!cr.success, - cr.error || undefined, cr.durationMs) + cr.error || undefined, cr.durationMs, cr.promptTokens, cr.completionTokens) } } } else { @@ -1261,7 +1274,7 @@ export function useChat(options: UseChatOptions): UseChatReturn { } } else { markDelegComplete(segs, data.subagentId, data.childConversationId, - !!data.success, data.resultPreview, data.durationMs) + !!data.success, data.resultPreview, data.durationMs, data.promptTokens, data.completionTokens) } flushSegmentsToMessage() }) diff --git a/mateclaw-ui/src/types/index.ts b/mateclaw-ui/src/types/index.ts index 442228cb..6bd3de32 100644 --- a/mateclaw-ui/src/types/index.ts +++ b/mateclaw-ui/src/types/index.ts @@ -171,6 +171,9 @@ export interface DelegationNode { tools?: DelegationToolEntry[] result?: string durationMs?: number + /** Child token usage relayed from delegation_end / delegation_child_complete. */ + promptTokens?: number + completionTokens?: number /** Heartbeat watchdog flagged this subagent as making no observable progress. */ stale?: boolean /** Spawned via fire-and-forget delegation: runs detached, result via task_output. */