Merge pull request #885 from pikasTech/fix/v02-agentrun-result-contract-cleanup

fix(v02): report AgentRun result contract
2026-06-05 02:27:53 +08:00
parent 8a48ed2dfe 1162c0e5bd
commit 869746374b
2 changed files with 96 additions and 61 deletions
@@ -2,17 +2,6 @@ import { randomUUID } from "node:crypto";

 import { defaultCodeAgentTraceStore } from "./code-agent-trace-store.ts";
 import { decorateCodeAgentSession } from "./code-agent-session-lifecycle.ts";
-import {
-  CODEX_APP_SERVER_PROTOCOL,
-  CODEX_APP_SERVER_WIRE_API,
-  CODEX_STDIO_BACKEND,
-  CODEX_STDIO_CAPABILITY_LEVEL,
-  CODEX_STDIO_IMPLEMENTATION_TYPE,
-  CODEX_STDIO_PROVIDER,
-  CODEX_STDIO_RUNNER_KIND,
-  CODEX_STDIO_SESSION_MODE,
-  longLivedSessionGate
-} from "./codex-stdio-session.ts";
 import {
  parsePositiveInteger,
  safeConversationId,
@@ -31,9 +20,13 @@ const DEFAULT_PROVIDER_ID = "G14";
 const DEFAULT_REPO_URL = "http://git-mirror-http.devops-infra.svc.cluster.local/pikasTech/HWLAB.git";
 const DEFAULT_RUNNER_NAMESPACE = "agentrun-v01";
 const DEFAULT_TIMEOUT_MS = 1_200_000;
+const AGENTRUN_BACKEND_PREFIX = ADAPTER_ID;
 const AGENTRUN_RUNNER_KIND = "agentrun-v01-shared-runner";
 const AGENTRUN_SESSION_MODE = "agentrun-v01-durable-session";
 const AGENTRUN_IMPLEMENTATION_TYPE = "agentrun-v01-shared-execution-infra";
+const AGENTRUN_CAPABILITY_LEVEL = "agentrun-v01-shared-code-agent-session";
+const AGENTRUN_PROVIDER_TRACE_PROTOCOL = "agentrun-v01-jsonrpc";
+const AGENTRUN_PROVIDER_TRACE_WIRE_API = "agentrun-v01-command-result";
 const AGENTRUN_BACKENDS = Object.freeze(["codex", "deepseek", "minimax-m3"]);
 const THREAD_CONTINUITY_POLICY = "hwlab-agentrun-v01-reuse-runner-thread";
 const SESSION_POLICY_RUN_LOCAL = "hwlab-agentrun-v01-session-runner-reuse";
@@ -82,7 +75,7 @@ export function describeAgentRunAdapterAvailability(env = process.env, options =
  const managerHost = hostForUrl(managerUrl);
  const repoHost = hostForUrl(repoUrl);
  const status = ready ? "agentrun-ready" : "blocked";
-  const capabilityLevel = ready ? CODEX_STDIO_CAPABILITY_LEVEL : "blocked";
+  const capabilityLevel = ready ? AGENTRUN_CAPABILITY_LEVEL : "blocked";

  return {
    endpoint: "POST /v1/agent/chat",
@@ -224,13 +217,13 @@ export function initialAgentRunChatResult({ params = {}, options = {}, traceId }
    messageId: `msg_${safeTraceId(traceId)?.slice(4) || randomUUID()}`,
    createdAt: timestamp,
    updatedAt: timestamp,
-    provider: CODEX_STDIO_PROVIDER,
+    provider: providerForBackendProfile(backendProfile),
    model: modelForBackendProfile(backendProfile, env),
-    backend: CODEX_STDIO_BACKEND,
+    backend: backendForBackendProfile(backendProfile),
    infrastructureBackend: `agentrun-v01/${backendProfile}`,
-    capabilityLevel: CODEX_STDIO_CAPABILITY_LEVEL,
-    sessionMode: CODEX_STDIO_SESSION_MODE,
-    implementationType: CODEX_STDIO_IMPLEMENTATION_TYPE,
+    capabilityLevel: AGENTRUN_CAPABILITY_LEVEL,
+    sessionMode: AGENTRUN_SESSION_MODE,
+    implementationType: AGENTRUN_IMPLEMENTATION_TYPE,
    threadContinuityPolicy: THREAD_CONTINUITY_POLICY,
    agentRun: {
      adapter: ADAPTER_ID,
@@ -623,13 +616,13 @@ export async function loadPersistedAgentRunResult(traceId, options = {}) {
    createdAt: session.startedAt ?? session.updatedAt ?? nowIso(options.now),
    updatedAt: session.updatedAt ?? nowIso(options.now),
    ownerUserId: session.ownerUserId,
-    provider: CODEX_STDIO_PROVIDER,
+    provider: providerForBackendProfile(agentRun.backendProfile ?? "deepseek"),
    model: modelForBackendProfile(agentRun.backendProfile, options.env ?? process.env),
-    backend: CODEX_STDIO_BACKEND,
+    backend: backendForBackendProfile(agentRun.backendProfile ?? "deepseek"),
    infrastructureBackend: `agentrun-v01/${agentRun.backendProfile ?? "deepseek"}`,
-    capabilityLevel: CODEX_STDIO_CAPABILITY_LEVEL,
-    sessionMode: CODEX_STDIO_SESSION_MODE,
-    implementationType: CODEX_STDIO_IMPLEMENTATION_TYPE,
+    capabilityLevel: AGENTRUN_CAPABILITY_LEVEL,
+    sessionMode: AGENTRUN_SESSION_MODE,
+    implementationType: AGENTRUN_IMPLEMENTATION_TYPE,
    finalResponse: session.session?.finalResponse ?? null,
    traceSummary: session.session?.traceSummary ?? null,
    agentRun: { ...agentRun, adapter: ADAPTER_ID, valuesPrinted: false },
@@ -1025,12 +1018,12 @@ function providerCredentialSecretRef(profile, env) {

 function agentRunProviderTrace({ base = {}, result = {}, terminalStatus = null } = {}) {
  return {
-    transport: "stdio",
-    protocol: CODEX_APP_SERVER_PROTOCOL,
-    wireApi: CODEX_APP_SERVER_WIRE_API,
-    runnerKind: CODEX_STDIO_RUNNER_KIND,
-    implementationType: CODEX_STDIO_IMPLEMENTATION_TYPE,
-    command: "codex app-server --listen stdio://",
+    transport: "agentrun-v01",
+    protocol: AGENTRUN_PROVIDER_TRACE_PROTOCOL,
+    wireApi: AGENTRUN_PROVIDER_TRACE_WIRE_API,
+    runnerKind: AGENTRUN_RUNNER_KIND,
+    implementationType: AGENTRUN_IMPLEMENTATION_TYPE,
+    command: "agentrun.v01.command.turn",
    toolName: "agentrun.v01.command.turn",
    source: ADAPTER_ID,
    backendProfile: base.agentRun?.backendProfile ?? null,
@@ -1062,15 +1055,15 @@ function agentRunFailureAttribution({ code, message, canceled = false } = {}) {
    return {
      category: "provider_invalid_tool_call",
      retryable: true,
-      userMessage: "Codex app-server/provider 返回了无效 tool-call arguments JSON；这不是 HWLAB device-pod 或 Cloud API 端点失败，请查看 providerTrace.failureKind 和 runnerTrace 定位上游 tool_call_id。",
-      summary: message || "Codex app-server/provider returned invalid tool-call arguments JSON."
+      userMessage: "AgentRun/provider 返回了无效 tool-call arguments JSON；这不是 HWLAB device-pod 或 Cloud API 端点失败，请查看 providerTrace.failureKind 和 runnerTrace 定位上游 tool_call_id。",
+      summary: message || "AgentRun/provider returned invalid tool-call arguments JSON."
    };
  }
  if (code === "thread-resume-failed") {
    return {
      category: "thread_resume_failed",
      retryable: true,
-      userMessage: "AgentRun 复用的 Codex thread 已失效；当前 turn 会按失败终止，Workbench 会清理 stale continuation 指针，下一轮自动从新 session/thread 启动。",
+      userMessage: "AgentRun 复用的 thread 已失效；当前 turn 会按失败终止，Workbench 会清理 stale continuation 指针，下一轮自动从新 session/thread 启动。",
      summary: message || "AgentRun thread/resume failed for an existing thread."
    };
  }
@@ -1083,24 +1076,29 @@ function agentRunFailureAttribution({ code, message, canceled = false } = {}) {
 }

 function agentRunLongLivedSessionGate(base = {}) {
-  return longLivedSessionGate({
-    provider: CODEX_STDIO_PROVIDER,
-    runnerKind: CODEX_STDIO_RUNNER_KIND,
-    session: agentRunSessionSummary(base, "idle"),
-    sessionMode: CODEX_STDIO_SESSION_MODE,
-    implementationType: CODEX_STDIO_IMPLEMENTATION_TYPE,
-    codexStdioFeasibility: {
-      ready: true,
-      canStartLongLivedCodexStdio: true,
-      source: ADAPTER_ID,
-      valuesPrinted: false
-    }
-  });
+  const session = agentRunSessionSummary(base, "idle");
+  return {
+    status: "pass",
+    pass: true,
+    requiredCapability: AGENTRUN_CAPABILITY_LEVEL,
+    currentCapability: AGENTRUN_IMPLEMENTATION_TYPE,
+    sessionId: session.sessionId ?? null,
+    sessionStatus: session.status ?? null,
+    runnerKind: AGENTRUN_RUNNER_KIND,
+    provider: providerForBackendProfile(base.agentRun?.backendProfile ?? "deepseek"),
+    sessionMode: AGENTRUN_SESSION_MODE,
+    implementationType: AGENTRUN_IMPLEMENTATION_TYPE,
+    adapter: ADAPTER_ID,
+    delegatedToAgentRun: true,
+    codexStdio: false,
+    blockers: [],
+    valuesPrinted: false
+  };
 }

 function agentRunToolCalls(result = {}, status = "completed") {
  return [{
-    name: "codex-app-server.turn",
+    name: "agentrun.v01.command.turn",
    status,
    runId: result?.runId ?? null,
    commandId: result?.commandId ?? null,
@@ -1584,11 +1582,12 @@ function agentRunSessionSummary(base, status) {
    conversationId: base.conversationId ?? base.agentRun?.conversationId ?? null,
    threadId: base.threadId ?? base.agentRun?.threadId ?? null,
    status,
-    sessionMode: CODEX_STDIO_SESSION_MODE,
-    runnerKind: CODEX_STDIO_RUNNER_KIND,
+    sessionMode: AGENTRUN_SESSION_MODE,
+    runnerKind: AGENTRUN_RUNNER_KIND,
    lastTraceId: base.traceId ?? base.agentRun?.traceId ?? null,
    longLivedSession: true,
-    codexStdio: true,
+    codexStdio: false,
+    delegatedToAgentRun: true,
    writeCapable: true,
    durable: true,
    durableSession: true,
@@ -1624,7 +1623,7 @@ function agentRunSessionReuseSummary(base, reused, options = {}) {

 function agentRunRunnerSummary(mapping = {}) {
  return {
-    kind: CODEX_STDIO_RUNNER_KIND,
+    kind: AGENTRUN_RUNNER_KIND,
    adapter: ADAPTER_ID,
    runId: mapping.runId ?? null,
    commandId: mapping.commandId ?? null,
@@ -1633,11 +1632,12 @@ function agentRunRunnerSummary(mapping = {}) {
    jobName: mapping.jobName ?? null,
    namespace: mapping.namespace ?? null,
    backendProfile: mapping.backendProfile ?? null,
-    provider: CODEX_STDIO_PROVIDER,
-    sessionMode: CODEX_STDIO_SESSION_MODE,
-    implementationType: CODEX_STDIO_IMPLEMENTATION_TYPE,
-    capabilityLevel: CODEX_STDIO_CAPABILITY_LEVEL,
-    codexStdio: true,
+    provider: providerForBackendProfile(mapping.backendProfile ?? "deepseek"),
+    sessionMode: AGENTRUN_SESSION_MODE,
+    implementationType: AGENTRUN_IMPLEMENTATION_TYPE,
+    capabilityLevel: AGENTRUN_CAPABILITY_LEVEL,
+    codexStdio: false,
+    delegatedToAgentRun: true,
    writeCapable: true,
    durableSession: true,
    longLivedSession: true,
@@ -1648,15 +1648,19 @@ function agentRunRunnerSummary(mapping = {}) {

 function agentRunTraceMeta(env = process.env) {
  return {
-    runnerKind: CODEX_STDIO_RUNNER_KIND,
+    runnerKind: AGENTRUN_RUNNER_KIND,
    workspace: resolveAgentRunRepoUrl(env),
    sandbox: firstNonEmpty(env.HWLAB_CODE_AGENT_AGENTRUN_SANDBOX, env.HWLAB_CODE_AGENT_CODEX_SANDBOX, "danger-full-access"),
-    protocol: CODEX_APP_SERVER_PROTOCOL,
-    sessionMode: CODEX_STDIO_SESSION_MODE,
-    implementationType: CODEX_STDIO_IMPLEMENTATION_TYPE
+    protocol: AGENTRUN_PROVIDER_TRACE_PROTOCOL,
+    sessionMode: AGENTRUN_SESSION_MODE,
+    implementationType: AGENTRUN_IMPLEMENTATION_TYPE
  };
 }

+function backendForBackendProfile(profile) {
+  return `${AGENTRUN_BACKEND_PREFIX}/${resolveAgentRunBackendProfile({}, { providerProfile: profile })}`;
+}
+
 function resolveAgentRunRepoUrl(env = process.env) {
  const raw = firstNonEmpty(env.HWLAB_CODE_AGENT_AGENTRUN_REPO_URL, env.HWLAB_BOOT_READ_URL, DEFAULT_REPO_URL);
  const url = new URL(raw);
@@ -437,15 +437,32 @@ test("cloud api /v1/agent/chat delegates v0.2 turns to AgentRun v0.1 over adapte
    const payload = await pollAgentResult(port, traceId);
    validateCodeAgentChatSchema(payload);
    assert.equal(payload.status, "completed");
-    assert.equal(payload.provider, "codex-stdio");
+    assert.equal(payload.provider, "deepseek");
+    assert.equal(payload.backend, "agentrun-v01/deepseek");
    assert.equal(payload.infrastructureBackend, "agentrun-v01/deepseek");
-    assert.equal(payload.providerTrace.protocol, "codex-app-server-jsonrpc-stdio");
-    assert.equal(payload.providerTrace.command, "codex app-server --listen stdio://");
+    assert.equal(payload.capabilityLevel, "agentrun-v01-shared-code-agent-session");
+    assert.equal(payload.sessionMode, "agentrun-v01-durable-session");
+    assert.equal(payload.implementationType, "agentrun-v01-shared-execution-infra");
+    assert.equal(payload.runner.kind, "agentrun-v01-shared-runner");
+    assert.equal(payload.runner.provider, "deepseek");
+    assert.equal(payload.runner.codexStdio, false);
+    assert.equal(payload.runner.delegatedToAgentRun, true);
+    assert.equal(payload.providerTrace.protocol, "agentrun-v01-jsonrpc");
+    assert.equal(payload.providerTrace.command, "agentrun.v01.command.turn");
+    assert.equal(payload.providerTrace.runnerKind, "agentrun-v01-shared-runner");
    assert.equal(payload.providerTrace.terminalStatus, "completed");
    assert.equal(payload.longLivedSessionGate.status, "pass");
+    assert.equal(payload.longLivedSessionGate.provider, "deepseek");
+    assert.equal(payload.longLivedSessionGate.codexStdio, false);
+    assert.equal(payload.longLivedSessionGate.delegatedToAgentRun, true);
    assert.equal(payload.agentRun.runId, "run_hwlab_adapter");
    assert.equal(payload.agentRun.commandId, "cmd_hwlab_adapter");
    assert.equal(payload.agentRun.jobName, "agentrun-v01-runner-hwlab-adapter");
+    const serializedPayload = JSON.stringify(payload);
+    assert.equal(serializedPayload.includes("repo-owned-codex"), false);
+    assert.equal(serializedPayload.includes("codex-app-server-stdio"), false);
+    assert.equal(serializedPayload.includes("\"provider\":\"codex-stdio\""), false);
+    assert.equal(serializedPayload.includes("\"codexStdio\":true"), false);
    assert.match(payload.reply.content, /AgentRun adapter/u);
    assert.ok(payload.runnerTrace.events.some((event) => event.label === "agentrun:backend:runner-job-created"));
    const bundleEvent = payload.runnerTrace.events.find((event) => event.label === "agentrun:backend:resource-bundle-materialized");
@@ -560,6 +577,10 @@ test("cloud api /v1/agent/chat delegates v0.2 turns to AgentRun v0.1 over adapte
    const secondPayload = await pollAgentResult(port, secondTraceId);
    validateCodeAgentChatSchema(secondPayload);
    assert.equal(secondPayload.status, "completed");
+    assert.equal(secondPayload.provider, "deepseek");
+    assert.equal(secondPayload.backend, "agentrun-v01/deepseek");
+    assert.equal(secondPayload.runner.kind, "agentrun-v01-shared-runner");
+    assert.equal(secondPayload.runner.codexStdio, false);
    assert.equal(secondPayload.agentRun.runId, "run_hwlab_adapter");
    assert.equal(secondPayload.agentRun.commandId, "cmd_hwlab_adapter_second");
    assert.equal(secondPayload.providerTrace.commandId, "cmd_hwlab_adapter_second");
@@ -889,6 +910,7 @@ test("cloud api AgentRun adapter exposes invalid tool-call attribution in result
    assert.equal(payload.error.code, "provider-invalid-tool-call");
    assert.equal(payload.error.category, "provider_invalid_tool_call");
    assert.match(payload.error.userMessage, /无效 tool-call arguments JSON/u);
+    assert.match(payload.error.userMessage, /AgentRun\/provider/u);
    assert.match(payload.error.userMessage, /不是 HWLAB device-pod/u);
    assert.equal(payload.blocker.category, "provider_invalid_tool_call");
    assert.match(payload.blocker.summary, /invalid function arguments json string/u);
@@ -897,6 +919,10 @@ test("cloud api AgentRun adapter exposes invalid tool-call attribution in result
    assert.equal(payload.sessionReuse.threadId, "thread_invalid_tool");
    assert.equal(payload.agentRun.reuseEligible, false);
    assert.equal(payload.reuseEligible, false);
+    const serializedPayload = JSON.stringify(payload);
+    assert.equal(serializedPayload.includes("repo-owned-codex"), false);
+    assert.equal(serializedPayload.includes("codex-app-server-stdio"), false);
+    assert.equal(serializedPayload.includes("\"provider\":\"codex-stdio\""), false);
  } finally {
    await new Promise((resolve, reject) => server.close((error) => (error ? reject(error) : resolve())));
    await new Promise((resolve, reject) => agentRunServer.close((error) => (error ? reject(error) : resolve())));
@@ -1031,7 +1057,12 @@ test("cloud api AgentRun adapter maps minimax-m3 provider profile to AgentRun ba
    const payload = await pollAgentResult(port, traceId);
    validateCodeAgentChatSchema(payload);
    assert.equal(payload.status, "completed");
+    assert.equal(payload.provider, "minimax-m3");
+    assert.equal(payload.backend, "agentrun-v01/minimax-m3");
    assert.equal(payload.infrastructureBackend, "agentrun-v01/minimax-m3");
+    assert.equal(payload.runner.kind, "agentrun-v01-shared-runner");
+    assert.equal(payload.runner.provider, "minimax-m3");
+    assert.equal(payload.runner.codexStdio, false);
    assert.equal(payload.agentRun.backendProfile, "minimax-m3");
    assert.equal(payload.sessionId, "ses_server-test-minimax-m3");
    assert.equal(payload.agentRun.sessionId, "ses_agentrun_minimax_m3_server_test_minimax_m3");