Files
pikasTech-HWLAB/internal/cloud/code-agent-session-registry.test.mjs
T
Lyon 8c8942d232 fix: route M3 IO before Codex stdio
M3 IO 中文请求先走受控 Skill CLI/HWLAB API 路径,避免 Codex stdio 150s 超时;补充 true/false/read 与结构化 blocker 测试。
2026-05-24 09:24:03 +08:00

1791 lines
70 KiB
JavaScript

import assert from "node:assert/strict";
import { existsSync } from "node:fs";
import { chmod, mkdir, mkdtemp, rm, writeFile } from "node:fs/promises";
import os from "node:os";
import path from "node:path";
import test from "node:test";
import { createCodeAgentSessionRegistry } from "./code-agent-session-registry.mjs";
import { handleCodeAgentChat, validateCodeAgentChatSchema } from "./code-agent-chat.mjs";
import { createCodexStdioSessionManager } from "./codex-stdio-session.mjs";
import {
classifyCodexRunnerCapability,
classifyCodeAgentChatReadiness
} from "../../scripts/src/code-agent-response-contract.mjs";
import {
HWLAB_M3_IO_API_BASE_URL_ENV,
HWLAB_M3_IO_CAPABILITY_LEVELS,
HWLAB_M3_IO_API_ROUTE,
HWLAB_M3_STATUS_API_ROUTE
} from "../../skills/hwlab-agent-runtime/scripts/src/m3-io-skill-client.mjs";
test("code agent session registry creates, reuses, and expires sessions without storing secrets", () => {
const registry = createCodeAgentSessionRegistry({
idleTimeoutMs: 1000,
idFactory: () => "ses_test-registry"
});
const first = registry.acquire({
conversationId: "cnv_registry",
workspace: "/workspace/hwlab",
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_registry_1",
now: "2026-05-23T00:00:00.000Z"
});
assert.equal(first.ok, true);
assert.equal(first.reused, false);
assert.equal(first.session.sessionId, "ses_test-registry");
assert.equal(first.session.status, "busy");
assert.equal(first.session.workspace, "/workspace/hwlab");
assert.equal(first.session.sandbox, "read-only");
assert.equal(first.session.runnerKind, "hwlab-readonly-runner");
assert.equal(first.session.sessionMode, "controlled-readonly-session-registry");
assert.equal(first.session.capabilityLevel, "read-only-session-tools");
assert.equal(first.session.longLivedSession, true);
assert.equal(first.session.durableSession, false);
assert.equal(first.session.codexStdio, false);
assert.equal(first.session.writeCapable, false);
assert.equal(first.session.readOnly, true);
assert.equal(first.session.idleTimeoutMs, 1000);
assert.equal(first.session.lastTraceId, "trc_registry_1");
assert.equal(first.session.secretMaterialStored, false);
assert.equal(first.session.valuesRedacted, true);
const describedCreating = registry.describe();
assert.equal(describedCreating.sessionMode, "controlled-readonly-session-registry");
assert.equal(describedCreating.longLivedSession, true);
assert.equal(describedCreating.longLivedCodexStdio, false);
assert.equal(describedCreating.recentSessions[0].sessionId, "ses_test-registry");
assert.equal(describedCreating.recentSessions[0].sessionMode, "controlled-readonly-session-registry");
assert.equal(describedCreating.recentSessions[0].workspace, "/workspace/hwlab");
assert.equal(describedCreating.recentSessions[0].secretMaterialStored, false);
registry.release(first.session.sessionId, {
conversationId: "cnv_registry",
traceId: "trc_registry_1",
reused: first.session.reused,
now: "2026-05-23T00:00:00.010Z"
});
const second = registry.acquire({
conversationId: "cnv_registry",
workspace: "/workspace/hwlab",
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_registry_2",
now: "2026-05-23T00:00:00.500Z"
});
assert.equal(second.ok, true);
assert.equal(second.reused, true);
assert.equal(second.session.sessionId, "ses_test-registry");
assert.equal(second.session.turn, 2);
assert.equal(second.session.lastTraceId, "trc_registry_2");
registry.release(second.session.sessionId, {
conversationId: "cnv_registry",
traceId: "trc_registry_2",
reused: second.session.reused,
now: "2026-05-23T00:00:00.510Z"
});
const expired = registry.acquire({
conversationId: "cnv_registry",
workspace: "/workspace/hwlab",
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_registry_expired",
now: "2026-05-23T00:00:02.000Z"
});
assert.equal(expired.ok, false);
assert.equal(expired.code, "session_expired");
assert.equal(expired.session.status, "expired");
assert.equal(expired.blocker.sourceIssue, "pikasTech/HWLAB#317");
});
test("code agent session registry stores bounded non-sensitive conversation facts", () => {
const registry = createCodeAgentSessionRegistry({
maxConversationFacts: 2,
maxFactSkills: 2,
idFactory: () => "ses_fact_registry"
});
const acquired = registry.acquire({
conversationId: "cnv_fact_registry",
workspace: "/workspace/hwlab",
traceId: "trc_fact_registry_1",
now: "2026-05-23T00:00:00.000Z"
});
registry.release(acquired.session.sessionId, {
conversationId: "cnv_fact_registry",
traceId: "trc_fact_registry_1",
now: "2026-05-23T00:00:00.010Z"
});
registry.recordFact("cnv_fact_registry", {
traceId: "trc_fact_registry_1",
provider: "codex-readonly-runner",
backend: "hwlab-cloud-api/codex-readonly-runner",
workspace: "/workspace/hwlab",
sandbox: "read-only",
session: acquired.session,
sessionMode: "controlled-readonly-session-registry",
capabilityLevel: "read-only-session-tools",
runner: {
kind: "hwlab-readonly-runner",
sessionId: "ses_fact_registry"
},
runnerTrace: {
traceId: "trc_fact_registry_1",
runnerKind: "hwlab-readonly-runner",
sessionMode: "controlled-readonly-session-registry",
events: ["tool:pwd:completed"],
valuesPrinted: false
},
toolCalls: [{
name: "pwd",
status: "completed",
stdout: "must-not-store-output sk-secret-value OPENAI_API_KEY=secret",
traceId: "trc_fact_registry_1"
}],
skills: {
status: "ready",
items: [
{ name: "alpha" },
{ name: "beta" },
{ name: "gamma" }
],
count: 3,
totalCount: 3
}
}, { now: "2026-05-23T00:00:00.020Z" });
registry.recordFact("cnv_fact_registry", {
traceId: "trc_fact_registry_2",
workspace: "/workspace/hwlab",
runner: { kind: "hwlab-readonly-runner" },
toolCalls: [{
name: "skills.discover",
status: "completed",
stdout: "must-not-store-output sk-secret-value OPENAI_API_KEY=secret"
}],
skills: {
status: "ready",
items: [
{ name: "alpha" },
{ name: "beta" },
{ name: "gamma" }
],
count: 3,
totalCount: 3
}
}, { now: "2026-05-23T00:00:00.030Z" });
registry.recordFact("cnv_fact_registry", {
traceId: "trc_fact_registry_3",
workspace: "/workspace/hwlab",
runner: { kind: "hwlab-readonly-runner" },
toolCalls: [{ name: "session.context", status: "completed" }]
}, { now: "2026-05-23T00:00:00.040Z" });
const facts = registry.getConversationFacts("cnv_fact_registry");
assert.equal(facts.conversationId, "cnv_fact_registry");
assert.equal(facts.sessionId, "ses_fact_registry");
assert.equal(facts.workspace, "/workspace/hwlab");
assert.equal(facts.turnCount, 2);
assert.deepEqual(facts.traceIds, ["trc_fact_registry_2", "trc_fact_registry_3"]);
assert.equal(facts.valuesRedacted, true);
assert.equal(facts.secretMaterialStored, false);
assert.deepEqual(facts.latestSkills.names, ["alpha", "beta"]);
assert.equal(facts.latestSkills.truncated, true);
assert.equal(JSON.stringify(facts).includes("must-not-store-output"), false);
assert.equal(JSON.stringify(facts).includes("sk-secret-value"), false);
assert.equal(JSON.stringify(facts).includes("OPENAI_API_KEY=secret"), false);
assert.match(facts.summary, /workspace=\/workspace\/hwlab/u);
});
test("code agent session registry reports busy sessions as structured blocker", () => {
const registry = createCodeAgentSessionRegistry({
idFactory: () => "ses_busy"
});
const first = registry.acquire({
conversationId: "cnv_busy",
workspace: "/workspace/hwlab",
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_busy_1",
now: "2026-05-23T00:00:00.000Z"
});
assert.equal(first.ok, true);
const busy = registry.acquire({
conversationId: "cnv_busy",
workspace: "/workspace/hwlab",
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_busy_2",
now: "2026-05-23T00:00:01.000Z"
});
assert.equal(busy.ok, false);
assert.equal(busy.code, "session_busy");
assert.equal(busy.session.status, "busy");
assert.equal(busy.session.currentTraceId, "trc_busy_1");
});
test("code agent session registry blocks failed and interrupted sessions without reusing them", () => {
const failedRegistry = createCodeAgentSessionRegistry({
idFactory: () => "ses_failed"
});
const failedSeed = failedRegistry.acquire({
conversationId: "cnv_failed",
workspace: "/workspace/hwlab",
traceId: "trc_failed_1",
now: "2026-05-23T00:00:00.000Z"
});
assert.equal(failedSeed.ok, true);
failedRegistry.fail(failedSeed.session.sessionId, {
conversationId: "cnv_failed",
traceId: "trc_failed_1",
now: "2026-05-23T00:00:00.010Z"
});
const failed = failedRegistry.acquire({
conversationId: "cnv_failed",
workspace: "/workspace/hwlab",
traceId: "trc_failed_2",
now: "2026-05-23T00:00:01.000Z"
});
assert.equal(failed.ok, false);
assert.equal(failed.code, "session_failed");
assert.equal(failed.session.status, "failed");
assert.equal(failed.blocker.sourceIssue, "pikasTech/HWLAB#317");
const interruptedRegistry = createCodeAgentSessionRegistry({
idFactory: () => "ses_interrupted"
});
const interruptedSeed = interruptedRegistry.acquire({
conversationId: "cnv_interrupted",
workspace: "/workspace/hwlab",
traceId: "trc_interrupted_1",
now: "2026-05-23T00:00:00.000Z"
});
assert.equal(interruptedSeed.ok, true);
interruptedRegistry.interrupt(interruptedSeed.session.sessionId, {
conversationId: "cnv_interrupted",
traceId: "trc_interrupted_1",
now: "2026-05-23T00:00:00.010Z"
});
const interrupted = interruptedRegistry.acquire({
conversationId: "cnv_interrupted",
workspace: "/workspace/hwlab",
traceId: "trc_interrupted_2",
now: "2026-05-23T00:00:01.000Z"
});
assert.equal(interrupted.ok, false);
assert.equal(interrupted.code, "session_interrupted");
assert.equal(interrupted.session.status, "interrupted");
});
test("read-only runner returns session registry evidence and blocked long-lived gate", async () => {
const registry = createCodeAgentSessionRegistry({
idFactory: () => "ses_chat_registry"
});
const first = await handleCodeAgentChat(
{
conversationId: "cnv_chat_registry",
traceId: "trc_chat_registry_pwd",
message: "pwd"
},
{
now: () => "2026-05-23T00:01:00.000Z",
sessionRegistry: registry,
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
}
}
);
validateCodeAgentChatSchema(first);
assert.equal(first.status, "completed");
assert.equal(first.sessionId, "ses_chat_registry");
assert.equal(first.session.sessionId, "ses_chat_registry");
assert.equal(first.session.status, "idle");
assert.equal(first.session.sessionMode, "controlled-readonly-session-registry");
assert.equal(first.session.workspace, process.cwd());
assert.equal(first.session.lastTraceId, "trc_chat_registry_pwd");
assert.equal(first.capabilityLevel, "read-only-session-tools");
assert.equal(first.session.longLivedSession, true);
assert.equal(first.runner.longLivedSession, true);
assert.equal(first.longLivedSessionGate.status, "blocked");
assert.equal(first.longLivedSessionGate.pass, false);
assert.ok(first.longLivedSessionGate.blockers.some((blocker) => blocker.code === "codex_stdio_blocked_readonly_session_available"));
const second = await handleCodeAgentChat(
{
conversationId: "cnv_chat_registry",
traceId: "trc_chat_registry_ls",
message: "ls ."
},
{
now: () => "2026-05-23T00:01:01.000Z",
sessionRegistry: registry,
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
}
}
);
assert.equal(second.status, "completed");
assert.equal(second.sessionId, first.sessionId);
assert.equal(second.sessionReuse.reused, true);
assert.equal(second.sessionReuse.turn, 2);
assert.equal(second.session.lastTraceId, "trc_chat_registry_ls");
assert.equal(classifyCodexRunnerCapability(second, { httpStatus: 200 }).capabilityPass, false);
});
test("expired, busy, and failed session states are surfaced as structured blockers", async () => {
const expiredRegistry = createCodeAgentSessionRegistry({
idleTimeoutMs: 10,
idFactory: () => "ses_expired_chat"
});
const first = expiredRegistry.acquire({
conversationId: "cnv_expired_chat",
workspace: process.cwd(),
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_expired_seed",
now: "2026-05-23T00:02:00.000Z"
});
expiredRegistry.release(first.session.sessionId, {
conversationId: "cnv_expired_chat",
traceId: "trc_expired_seed",
now: "2026-05-23T00:02:00.001Z"
});
expiredRegistry.recordFact("cnv_expired_chat", {
traceId: "trc_expired_seed",
workspace: process.cwd(),
session: first.session,
runner: { kind: "hwlab-readonly-runner" },
sessionMode: "controlled-readonly-session-registry",
capabilityLevel: "read-only-session-tools",
toolCalls: [{ name: "pwd", status: "completed" }]
}, { now: "2026-05-23T00:02:00.002Z" });
const expired = await handleCodeAgentChat(
{
conversationId: "cnv_expired_chat",
traceId: "trc_expired_request",
message: "pwd"
},
{
now: () => "2026-05-23T00:02:01.000Z",
sessionRegistry: expiredRegistry,
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
}
}
);
assert.equal(expired.status, "failed");
assert.equal(expired.error.code, "session_expired");
assert.equal(expired.error.layer, "session");
assert.equal(expired.error.retryable, true);
assert.match(expired.error.userMessage, /session 已过期/u);
assert.equal(expired.error.blocker.code, "session_expired");
assert.equal(expired.error.blocker.traceId, "trc_expired_request");
assert.equal(expired.session.status, "expired");
assert.equal(expired.capabilityLevel, "blocked");
assert.equal(expired.longLivedSessionGate.status, "blocked");
assert.equal(expired.conversationFacts.conversationId, "cnv_expired_chat");
assert.equal(expired.conversationFacts.sessionId, "ses_expired_chat");
assert.equal(expired.conversationFacts.turnCount, 1);
assert.equal(expired.conversationFacts.recentToolCalls[0].name, "pwd");
assert.equal(expired.error.blocker.conversationFacts.sessionId, "ses_expired_chat");
const busyRegistry = createCodeAgentSessionRegistry({
idFactory: () => "ses_busy_chat"
});
busyRegistry.acquire({
conversationId: "cnv_busy_chat",
workspace: process.cwd(),
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_busy_seed",
now: "2026-05-23T00:03:00.000Z"
});
busyRegistry.recordFact("cnv_busy_chat", {
traceId: "trc_busy_seed",
workspace: process.cwd(),
sessionId: "ses_busy_chat",
runner: { kind: "hwlab-readonly-runner" },
sessionMode: "controlled-readonly-session-registry",
capabilityLevel: "read-only-session-tools",
toolCalls: [{ name: "skills.discover", status: "completed" }],
skills: {
status: "ready",
items: [{ name: "alpha-skill" }],
count: 1,
totalCount: 1
}
}, { now: "2026-05-23T00:03:00.001Z" });
const busy = await handleCodeAgentChat(
{
conversationId: "cnv_busy_chat",
traceId: "trc_busy_request",
message: "pwd"
},
{
now: () => "2026-05-23T00:03:01.000Z",
sessionRegistry: busyRegistry,
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
}
}
);
assert.equal(busy.status, "failed");
assert.equal(busy.error.code, "session_busy");
assert.equal(busy.error.layer, "session");
assert.equal(busy.error.retryable, true);
assert.match(busy.error.userMessage, /正在处理上一轮/u);
assert.equal(busy.error.blocker.code, "session_busy");
assert.equal(busy.session.status, "busy");
assert.equal(busy.capabilityLevel, "blocked");
assert.equal(busy.error.blockers[0].sourceIssue, "pikasTech/HWLAB#317");
assert.equal(busy.conversationFacts.conversationId, "cnv_busy_chat");
assert.equal(busy.conversationFacts.sessionId, "ses_busy_chat");
assert.equal(busy.conversationFacts.latestSkills.names[0], "alpha-skill");
const failedRegistry = createCodeAgentSessionRegistry({
idFactory: () => "ses_failed_chat"
});
const failedSeed = failedRegistry.acquire({
conversationId: "cnv_failed_chat",
workspace: process.cwd(),
sandbox: "read-only",
runnerKind: "hwlab-readonly-runner",
capabilityLevel: "read-only-session-tools",
traceId: "trc_failed_seed",
now: "2026-05-23T00:04:00.000Z"
});
failedRegistry.fail(failedSeed.session.sessionId, {
conversationId: "cnv_failed_chat",
traceId: "trc_failed_seed",
now: "2026-05-23T00:04:00.001Z"
});
failedRegistry.recordFact("cnv_failed_chat", {
traceId: "trc_failed_seed",
workspace: process.cwd(),
session: failedSeed.session,
runner: { kind: "hwlab-readonly-runner" },
sessionMode: "controlled-readonly-session-registry",
capabilityLevel: "read-only-session-tools",
toolCalls: [{ name: "pwd", status: "completed" }]
}, { now: "2026-05-23T00:04:00.002Z" });
const failed = await handleCodeAgentChat(
{
conversationId: "cnv_failed_chat",
traceId: "trc_failed_request",
message: "pwd"
},
{
now: () => "2026-05-23T00:04:01.000Z",
sessionRegistry: failedRegistry,
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
}
}
);
assert.equal(failed.status, "failed");
assert.equal(failed.error.code, "session_failed");
assert.equal(failed.error.layer, "session");
assert.equal(failed.error.retryable, true);
assert.match(failed.error.userMessage, /session 已失败/u);
assert.equal(failed.error.blocker.code, "session_failed");
assert.equal(failed.session.status, "failed");
assert.equal(failed.capabilityLevel, "blocked");
assert.equal(failed.conversationFacts.conversationId, "cnv_failed_chat");
assert.equal(failed.conversationFacts.sessionId, "ses_failed_chat");
assert.equal(failed.conversationFacts.recentToolCalls[0].name, "pwd");
});
test("OpenAI fallback and codex one-shot do not pass the long-lived session gate", async () => {
const fallback = await handleCodeAgentChat(
{
conversationId: "cnv_fallback_gate",
traceId: "trc_fallback_gate",
message: "你好"
},
{
now: () => "2026-05-23T00:04:00.000Z",
callProvider: async ({ providerPlan }) => ({
provider: providerPlan.provider,
model: providerPlan.model,
backend: providerPlan.backend,
content: "普通文本回复。",
usage: null,
providerTrace: {
source: "test-provider"
}
})
}
);
assert.equal(fallback.status, "completed");
assert.equal(fallback.runner.kind, "openai-responses-fallback");
assert.equal(fallback.capabilityLevel, "text-chat-only");
assert.deepEqual(fallback.runnerLimitations, [
"openai-responses-fallback",
"text-chat-only",
"not-workspace-tools",
"not-session-runner"
]);
assert.equal(fallback.runnerLimitations.includes("not-codex-stdio"), false);
assert.equal(fallback.blocker.code, "text_chat_only_fallback");
assert.equal(fallback.blocker.layer, "provider");
assert.equal(fallback.blocker.retryable, false);
assert.match(fallback.blocker.userMessage, /文本 fallback/u);
assert.equal(fallback.longLivedSessionGate.status, "blocked");
assert.ok(fallback.longLivedSessionGate.blockers.some((blocker) => blocker.code === "openai_responses_fallback_not_session"));
assert.equal(classifyCodexRunnerCapability(fallback, { httpStatus: 200 }).capabilityPass, false);
const oneShot = {
...fallback,
provider: "codex-cli",
backend: "hwlab-cloud-api/codex-cli",
runner: {
kind: "codex-cli-one-shot-ephemeral",
codexStdio: false,
writeCapable: false,
durableSession: false,
longLivedSession: false
},
sessionMode: "ephemeral-one-shot",
implementationType: "codex-cli-one-shot-ephemeral",
longLivedSessionGate: {
status: "blocked",
pass: false,
blockers: [{ code: "one_shot_runner_not_long_lived", sourceIssue: "pikasTech/HWLAB#317" }]
}
};
assert.equal(classifyCodexRunnerCapability(oneShot, { httpStatus: 200 }).capabilityPass, false);
assert.equal(classifyCodeAgentChatReadiness(oneShot, { realDevLive: true, httpStatus: 200 }).devLiveReplyPass, true);
});
test("Code Agent M3 DO write uses Skill CLI to call only HWLAB API /v1/m3/io", async () => {
const calls = [];
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_skill_write",
traceId: "trc_m3_skill_write",
message: "请通过 M3 DO1 写入 true,并回读 DI1"
},
{
now: () => "2026-05-23T00:05:00.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667",
OPENAI_API_KEY: "must-not-be-used"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 IO control");
},
m3IoSkillRequestJson: async (url, request) => {
calls.push({ url, request });
const parsed = new URL(url);
assert.equal(parsed.pathname, HWLAB_M3_IO_API_ROUTE);
assert.equal(parsed.hostname, "hwlab-cloud-api.hwlab-dev.svc.cluster.local");
assert.equal(request.method, "POST");
assert.equal(request.body.action, "do.write");
assert.equal(request.body.resourceId, "res_boxsimu_1");
assert.equal(request.body.port, "DO1");
assert.equal(request.body.value, true);
assert.equal(request.body.source, "hwlab-agent-runtime.m3-io");
return {
ok: true,
status: 200,
body: {
serviceId: "hwlab-cloud-api",
contractVersion: "m3-io-control-v1",
status: "succeeded",
accepted: true,
action: "do.write",
traceId: "trc_m3_skill_write",
operationId: "op_m3_do_write_skill",
auditId: "aud_m3_do_write_skill_succeeded",
evidenceId: "evd_m3_do_write_skill_succeeded",
auditState: {
status: "written_non_durable",
durableStatus: {
status: "degraded"
}
},
evidenceState: {
status: "blocked",
sourceKind: "BLOCKED",
blocker: "runtime_durable_not_green",
writeStatus: "written_non_durable"
},
durableStatus: {
status: "degraded",
durable: false,
blocker: "runtime_durable_not_green"
},
result: {
value: true,
targetReadback: {
status: "succeeded",
value: true,
resourceId: "res_boxsimu_2",
port: "DI1"
}
},
controlPath: {
status: "succeeded",
cloudApi: true,
gatewaySimu: true,
boxSimu: true,
patchPanel: true,
frontendBypass: false
}
}
};
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.backend, "hwlab-cloud-api/hwlab-agent-runtime-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.ready);
assert.equal(payload.session.sessionMode, "controlled-m3-io-skill-cli");
assert.equal(payload.toolCalls.length, 1);
assert.equal(payload.toolCalls[0].name, "hwlab-agent-runtime.m3-io");
assert.equal(payload.toolCalls[0].status, "completed");
assert.equal(payload.toolCalls[0].route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.toolCalls[0].method, "POST");
assert.equal(payload.toolCalls[0].capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.ready);
assert.equal(payload.toolCalls[0].controlReady, true);
assert.equal(payload.toolCalls[0].accepted, true);
assert.equal(payload.toolCalls[0].operationId, "op_m3_do_write_skill");
assert.equal(payload.toolCalls[0].traceId, "trc_m3_skill_write");
assert.equal(payload.toolCalls[0].auditId, "aud_m3_do_write_skill_succeeded");
assert.equal(payload.toolCalls[0].evidenceId, "evd_m3_do_write_skill_succeeded");
assert.equal(payload.toolCalls[0].readback.value, true);
assert.equal(payload.runnerTrace.route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.runnerTrace.method, "POST");
assert.equal(payload.runnerTrace.accepted, true);
assert.equal(payload.runnerTrace.auditId, "aud_m3_do_write_skill_succeeded");
assert.equal(payload.runnerTrace.evidenceId, "evd_m3_do_write_skill_succeeded");
assert.equal(payload.runnerTrace.readback.value, true);
assert.equal(payload.runnerTrace.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.ready);
assert.equal(payload.runnerTrace.controlReady, true);
assert.equal(payload.runner.writeCapable, true);
assert.deepEqual(payload.runner.toolPolicy.allowed, [`POST ${HWLAB_M3_IO_API_ROUTE}`]);
assert.equal(payload.runner.safety.directGatewayCallsAllowed, false);
assert.equal(payload.runner.safety.directBoxSimuCallsAllowed, false);
assert.equal(payload.runner.safety.directPatchPanelCallsAllowed, false);
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.equal(payload.providerTrace.method, "POST");
assert.equal(payload.providerTrace.auditId, "aud_m3_do_write_skill_succeeded");
assert.equal(payload.providerTrace.evidenceId, "evd_m3_do_write_skill_succeeded");
assert.equal(payload.providerTrace.readback.value, true);
assert.equal(payload.providerTrace.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.ready);
assert.equal(payload.providerTrace.controlReady, true);
assert.equal(payload.skills.blockers.some((blocker) => blocker.code === "runtime_durable_not_green"), true);
assert.match(payload.reply.content, /Skill CLI -> HWLAB API \/v1\/m3\/io/u);
assert.equal(calls.length, 1);
assert.equal(calls[0].url, `http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667${HWLAB_M3_IO_API_ROUTE}`);
assert.equal(calls[0].url.includes("gateway"), false);
assert.equal(calls[0].url.includes("box-simu"), false);
assert.equal(calls[0].url.includes("patch-panel"), false);
});
test("Code Agent M3 DI read returns structured blocker from HWLAB API without fallback", async () => {
const calls = [];
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_skill_read_blocked",
traceId: "trc_m3_skill_read_blocked",
message: "读取 M3 DI1 readback"
},
{
now: () => "2026-05-23T00:06:00.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667",
OPENAI_API_KEY: "must-not-be-used"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 IO read");
},
m3IoSkillRequestJson: async (url, request) => {
calls.push({ url, request });
return {
ok: true,
status: 200,
body: {
serviceId: "hwlab-cloud-api",
contractVersion: "m3-io-control-v1",
status: "blocked",
accepted: false,
action: "di.read",
traceId: "trc_m3_skill_read_blocked",
operationId: "op_m3_di_read_blocked",
auditId: "aud_m3_di_read_blocked_failed",
evidenceId: "evd_m3_di_read_blocked_failed",
blocker: {
code: "m3_gateway_session_unavailable",
message: "gateway unavailable",
zh: "gateway 未注册/不可用"
},
blockerClassification: {
category: "gateway_session"
},
auditState: {
status: "written_non_durable"
},
evidenceState: {
status: "blocked",
sourceKind: "BLOCKED",
blocker: "runtime_durable_not_green",
writeStatus: "written_non_durable"
},
durableStatus: {
status: "degraded",
durable: false,
blocker: "runtime_durable_not_green"
},
controlPath: {
cloudApi: true,
gatewaySimu: false,
boxSimu: false,
patchPanel: false,
frontendBypass: false
}
}
};
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.blocker.code, "m3_readiness_blocked");
assert.equal(payload.blocker.layer, "m3-readiness");
assert.equal(payload.blocker.retryable, false);
assert.equal(payload.blocker.route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.blocker.toolName, "hwlab-agent-runtime.m3-io");
assert.equal(payload.toolCalls[0].status, "blocked");
assert.equal(payload.toolCalls[0].route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.toolCalls[0].capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.toolCalls[0].controlReady, false);
assert.equal(payload.toolCalls[0].blocker.code, "m3_gateway_session_unavailable");
assert.equal(payload.skills.blockers[0].code, "m3_gateway_session_unavailable");
assert.equal(payload.runnerTrace.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.runnerTrace.blocker.code, "m3_gateway_session_unavailable");
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.equal(payload.providerTrace.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.providerTrace.controlReady, false);
assert.equal(calls.length, 1);
assert.equal(new URL(calls[0].url).pathname, HWLAB_M3_IO_API_ROUTE);
assert.equal(calls[0].request.body.action, "di.read");
assert.equal(calls[0].request.body.resourceId, "res_boxsimu_2");
assert.equal(calls[0].request.body.port, "DI1");
});
test("Code Agent M3 Skill CLI missing API base returns structured config blocker", async () => {
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_skill_api_base_missing",
traceId: "trc_m3_skill_api_base_missing",
message: "读取 M3 DI1 readback"
},
{
now: () => "2026-05-23T00:06:30.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_REQUIRE_HWLAB_API_BASE_URL: "1"
},
m3IoSkillRequestJson: async () => {
throw new Error("missing API base must not issue a network request");
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.blocker.code, "m3_readiness_blocked");
assert.equal(payload.toolCalls[0].method, "POST");
assert.equal(payload.toolCalls[0].blocker.code, "skill_cli_api_base_missing");
assert.equal(payload.toolCalls[0].blocker.layer, "skill-cli-config");
assert.equal(payload.toolCalls[0].blocker.retryable, false);
assert.deepEqual(payload.toolCalls[0].blocker.missingConfig, [
"HWLAB_CODE_AGENT_HWLAB_API_BASE_URL",
"HWLAB_API_BASE_URL",
"HWLAB_CLOUD_API_BASE_URL",
"contract:hwlab-agent-runtime.m3-io.apiBaseUrl"
]);
assert.equal(JSON.stringify(payload).includes("://"), false);
});
test("Code Agent M3 Skill CLI preserves slow structured blocker beyond legacy 4500ms window", async () => {
const startedAt = Date.now();
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_skill_slow_blocker",
traceId: "trc_m3_skill_slow_blocker",
message: "通过 HWLAB API 读取 M3 DI1 并返回结构化 blocker"
},
{
now: () => "2026-05-23T00:06:30.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667",
OPENAI_API_KEY: "must-not-be-used"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 IO blockers");
},
m3IoSkillRequestJson: async (url, request) => {
assert.equal(new URL(url).pathname, HWLAB_M3_IO_API_ROUTE);
assert.equal(request.timeoutMs, 30000);
await delay(4600);
return {
ok: true,
status: 200,
body: {
serviceId: "hwlab-cloud-api",
contractVersion: "m3-io-control-v1",
status: "blocked",
accepted: false,
action: "di.read",
traceId: "trc_m3_skill_slow_blocker",
operationId: "op_m3_di_read_slow_blocker",
auditId: "aud_m3_di_read_slow_blocker_failed",
evidenceId: "evd_m3_di_read_slow_blocker_failed",
blocker: {
code: "runtime_durable_not_green",
message: "runtime durable evidence is not green",
zh: "runtime durable evidence 尚未为 green"
},
blockerClassification: {
category: "runtime_durability",
reason: "durable evidence is blocked"
},
durableStatus: {
status: "degraded",
durable: false,
blocker: "runtime_durable_not_green"
},
evidenceState: {
status: "blocked",
sourceKind: "BLOCKED",
blocker: "runtime_durable_not_green",
writeStatus: "not_written"
},
controlPath: {
cloudApi: true,
gatewaySimu: false,
boxSimu: false,
patchPanel: false,
frontendBypass: false
}
}
};
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(Date.now() - startedAt >= 4500, true);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.backend, "hwlab-cloud-api/hwlab-agent-runtime-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.sessionMode, "controlled-m3-io-skill-cli");
assert.equal(payload.runner.kind, "hwlab-m3-io-skill-cli");
assert.equal(payload.blocker.code, "m3_readiness_blocked");
assert.equal(payload.blocker.layer, "m3-readiness");
assert.equal(payload.blocker.retryable, false);
assert.equal(payload.toolCalls[0].status, "blocked");
assert.equal(payload.toolCalls[0].blocker.code, "runtime_durable_not_green");
assert.equal(payload.toolCalls[0].traceId, "trc_m3_skill_slow_blocker");
assert.equal(payload.skills.status, "used");
assert.equal(payload.skills.blockers[0].code, "runtime_durable_not_green");
assert.equal(payload.runnerTrace.traceId, "trc_m3_skill_slow_blocker");
assert.equal(payload.runnerTrace.blocker.code, "runtime_durable_not_green");
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.equal(payload.providerTrace.traceId, "trc_m3_skill_slow_blocker");
assert.equal(Object.hasOwn(payload, "error"), false);
assert.match(payload.reply.content, /Blocker: runtime_durable_not_green/u);
});
test("Code Agent M3 Skill CLI HWLAB API unavailable returns retryable structured blocker", async () => {
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_skill_hwlab_api_unavailable",
traceId: "trc_m3_skill_hwlab_api_unavailable",
message: "读取 M3 DI1 readback"
},
{
now: () => "2026-05-23T00:06:40.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667"
},
m3IoSkillRequestJson: async () => ({
ok: false,
status: 503,
body: null,
error: "service unavailable"
})
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.blocker.code, "hwlab_api_unavailable");
assert.equal(payload.blocker.layer, "hwlab-api");
assert.equal(payload.blocker.retryable, true);
assert.equal(payload.blocker.route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.toolCalls[0].method, "POST");
assert.equal(payload.toolCalls[0].blocker.code, "hwlab_api_unavailable");
assert.equal(payload.toolCalls[0].blocker.retryable, true);
});
test("Code Agent M3 DO false write exposes readback and identifier contract", async () => {
const calls = [];
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_skill_write_false",
traceId: "trc_m3_skill_write_false",
message: "通过 HWLAB API 把 res_boxsimu_1 DO1 写成 false 并读取 res_boxsimu_2 DI1"
},
{
now: () => "2026-05-23T00:06:30.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667",
OPENAI_API_KEY: "must-not-be-used"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 IO control");
},
m3IoSkillRequestJson: async (url, request) => {
calls.push({ url, request });
return {
ok: true,
status: 200,
body: {
serviceId: "hwlab-cloud-api",
contractVersion: "m3-io-control-v1",
status: "succeeded",
accepted: true,
action: "do.write",
traceId: "trc_m3_skill_write_false",
operationId: "op_m3_do_write_skill_false",
auditId: "aud_m3_do_write_skill_false_succeeded",
evidenceId: "evd_m3_do_write_skill_false_succeeded",
auditState: {
status: "written_non_durable"
},
evidenceState: {
status: "blocked",
sourceKind: "BLOCKED",
blocker: "runtime_durable_not_green",
writeStatus: "written_non_durable"
},
durableStatus: {
status: "degraded",
durable: false,
blocker: "runtime_durable_not_green"
},
result: {
value: false,
targetReadback: {
status: "succeeded",
value: false,
resourceId: "res_boxsimu_2",
port: "DI1"
}
},
controlPath: {
status: "succeeded",
cloudApi: true,
gatewaySimu: true,
boxSimu: true,
patchPanel: true,
frontendBypass: false
}
}
};
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.ready);
const tool = payload.toolCalls[0];
assert.equal(tool.route, HWLAB_M3_IO_API_ROUTE);
assert.equal(tool.method, "POST");
assert.equal(tool.accepted, true);
assert.equal(tool.status, "completed");
assert.equal(tool.operationId, "op_m3_do_write_skill_false");
assert.equal(tool.traceId, "trc_m3_skill_write_false");
assert.equal(tool.auditId, "aud_m3_do_write_skill_false_succeeded");
assert.equal(tool.evidenceId, "evd_m3_do_write_skill_false_succeeded");
assert.equal(tool.readback.value, false);
assert.equal(tool.readback.resourceId, "res_boxsimu_2");
assert.equal(tool.readback.port, "DI1");
assert.equal(payload.runnerTrace.method, "POST");
assert.equal(payload.runnerTrace.readback.value, false);
assert.equal(payload.providerTrace.readback.value, false);
assert.equal(calls.length, 1);
assert.equal(calls[0].url, `http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667${HWLAB_M3_IO_API_ROUTE}`);
assert.equal(calls[0].request.body.action, "do.write");
assert.equal(calls[0].request.body.value, false);
assert.equal(calls[0].url.includes("gateway"), false);
assert.equal(calls[0].url.includes("box-simu"), false);
assert.equal(calls[0].url.includes("patch-panel"), false);
});
test("Code Agent M3 true/false/read requests bypass Codex stdio and use Skill CLI /v1/m3/io", async () => {
const calls = [];
const codexStdioCalls = [];
const forbiddenCodexStdioManager = {
describe() {
codexStdioCalls.push("describe");
throw new Error("M3 IO must not inspect Codex stdio before Skill CLI control");
},
async probe() {
codexStdioCalls.push("probe");
throw new Error("M3 IO must not probe Codex stdio before Skill CLI control");
},
async chat() {
codexStdioCalls.push("chat");
throw new Error("M3 IO must not enter Codex stdio chat");
},
cancel() {
codexStdioCalls.push("cancel");
},
reapIdle() {
codexStdioCalls.push("reapIdle");
}
};
let sessionSeq = 0;
const sessionRegistry = createCodeAgentSessionRegistry({
idFactory: () => `ses_m3_stdio_bypass_${sessionSeq += 1}`
});
const cases = [
{
suffix: "true",
message: "通过 HWLAB API 把 DO1 置为 true,然后读取 DI1。",
action: "do.write",
value: true,
operationId: "op_m3_do_true_bypass",
auditId: "aud_m3_do_true_bypass",
evidenceId: "evd_m3_do_true_bypass"
},
{
suffix: "false",
message: "把 res_boxsimu_1 的 DO1 置为 false,并回读 res_boxsimu_2 的 DI1。",
action: "do.write",
value: false,
operationId: "op_m3_do_false_bypass",
auditId: "aud_m3_do_false_bypass",
evidenceId: "evd_m3_do_false_bypass"
},
{
suffix: "read",
message: "读取 DI1 的当前状态,走 HWLAB API 控制 M3 IO。",
action: "di.read",
value: false,
operationId: "op_m3_di_read_bypass",
auditId: "aud_m3_di_read_bypass",
evidenceId: "evd_m3_di_read_bypass"
}
];
for (const item of cases) {
const payload = await handleCodeAgentChat(
{
conversationId: `cnv_m3_stdio_bypass_${item.suffix}`,
traceId: `trc_m3_stdio_bypass_${item.suffix}`,
message: item.message
},
{
now: () => "2026-05-23T00:08:40.000Z",
codexStdioManager: forbiddenCodexStdioManager,
sessionRegistry,
env: {
PATH: process.env.PATH,
OPENAI_API_KEY: "test-openai-key-material",
HWLAB_CODE_AGENT_PROVIDER: "codex-stdio",
HWLAB_CODE_AGENT_MODEL: "gpt-test",
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 IO control");
},
m3IoSkillRequestJson: async (url, request) => {
calls.push({ suffix: item.suffix, url, request });
assert.equal(new URL(url).pathname, HWLAB_M3_IO_API_ROUTE);
assert.equal(request.method, "POST");
assert.equal(request.body.action, item.action);
assert.equal(request.body.source, "hwlab-agent-runtime.m3-io");
if (item.action === "do.write") {
assert.equal(request.body.resourceId, "res_boxsimu_1");
assert.equal(request.body.port, "DO1");
assert.equal(request.body.value, item.value);
} else {
assert.equal(request.body.resourceId, "res_boxsimu_2");
assert.equal(request.body.port, "DI1");
}
return {
ok: true,
status: 200,
body: {
serviceId: "hwlab-cloud-api",
contractVersion: "m3-io-control-v1",
status: "succeeded",
accepted: true,
action: item.action,
traceId: `trc_m3_stdio_bypass_${item.suffix}`,
operationId: item.operationId,
auditId: item.auditId,
evidenceId: item.evidenceId,
auditState: {
status: item.action === "do.write" ? "written_non_durable" : "read"
},
evidenceState: {
status: "blocked",
sourceKind: "BLOCKED",
blocker: "runtime_durable_not_green",
writeStatus: item.action === "do.write" ? "written_non_durable" : "not_written"
},
durableStatus: {
status: "degraded",
durable: false,
blocker: "runtime_durable_not_green"
},
result: item.action === "do.write"
? {
value: item.value,
targetReadback: {
status: "succeeded",
value: item.value,
resourceId: "res_boxsimu_2",
port: "DI1"
}
}
: {
status: "succeeded",
value: item.value,
resourceId: "res_boxsimu_2",
port: "DI1"
},
controlPath: {
status: "succeeded",
cloudApi: true,
gatewaySimu: true,
boxSimu: true,
patchPanel: true,
frontendBypass: false
}
}
};
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.runner.kind, "hwlab-m3-io-skill-cli");
assert.equal(payload.sessionMode, "controlled-m3-io-skill-cli");
assert.equal(payload.codexStdioFeasibility.skipped, true);
assert.equal(payload.codexStdioFeasibility.reason, "m3_io_skill_cli_deterministic_route");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.ready);
assert.equal(payload.toolCalls.length, 1);
const skillTool = payload.toolCalls[0];
assert.equal(skillTool.type, "skill-cli");
assert.equal(skillTool.name, "hwlab-agent-runtime.m3-io");
assert.equal(skillTool.status, "completed");
assert.equal(skillTool.route, HWLAB_M3_IO_API_ROUTE);
assert.equal(skillTool.method, "POST");
assert.equal(skillTool.accepted, true);
assert.equal(skillTool.operationId, item.operationId);
assert.equal(skillTool.auditId, item.auditId);
assert.equal(skillTool.evidenceId, item.evidenceId);
assert.equal(skillTool.readback.value, item.value);
assert.equal(payload.runnerTrace.route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.runnerTrace.method, "POST");
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.equal(payload.providerTrace.readback.value, item.value);
assert.ok(payload.runner.toolPolicy.allowed.includes(`POST ${HWLAB_M3_IO_API_ROUTE}`));
assert.equal(JSON.stringify(payload.toolCalls).includes("gateway-simu"), false);
assert.equal(JSON.stringify(payload.toolCalls).includes("box-simu"), false);
assert.equal(JSON.stringify(payload.toolCalls).includes("patch-panel"), false);
}
assert.equal(codexStdioCalls.length, 0);
assert.equal(calls.length, cases.length);
});
test("Code Agent ambiguous M3 IO control request returns structured blocker without Codex stdio", async () => {
const codexStdioCalls = [];
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_ambiguous_stdio_bypass",
traceId: "trc_m3_ambiguous_stdio_bypass",
message: "通过 HWLAB API 控制 M3 IO。"
},
{
now: () => "2026-05-23T00:08:50.000Z",
codexStdioManager: {
describe() {
codexStdioCalls.push("describe");
throw new Error("ambiguous M3 IO must not inspect Codex stdio");
},
async probe() {
codexStdioCalls.push("probe");
throw new Error("ambiguous M3 IO must not probe Codex stdio");
},
async chat() {
codexStdioCalls.push("chat");
throw new Error("ambiguous M3 IO must not enter Codex stdio chat");
},
cancel() {
codexStdioCalls.push("cancel");
},
reapIdle() {
codexStdioCalls.push("reapIdle");
}
},
env: {
PATH: process.env.PATH,
OPENAI_API_KEY: "test-openai-key-material",
HWLAB_CODE_AGENT_PROVIDER: "codex-stdio",
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for ambiguous M3 IO");
},
m3IoSkillRequestJson: async () => {
throw new Error("ambiguous M3 IO must block before any HWLAB API mutation request");
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.blocker.code, "m3_readiness_blocked");
assert.equal(payload.toolCalls.length, 1);
assert.equal(payload.toolCalls[0].status, "blocked");
assert.equal(payload.toolCalls[0].route, HWLAB_M3_IO_API_ROUTE);
assert.equal(payload.toolCalls[0].blocker.code, "m3_io_intent_action_missing");
assert.equal(payload.toolCalls[0].hwlabApi.source, "intent-validation");
assert.equal(payload.toolCalls[0].accepted, false);
assert.equal(payload.toolCalls[0].operationId, null);
assert.equal(payload.codexStdioFeasibility.skipped, true);
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.match(payload.reply.content, /Blocker: m3_io_intent_action_missing/u);
assert.equal(codexStdioCalls.length, 0);
});
test("Code Agent M3 status uses Skill CLI GET /v1/m3/status and keeps route evidence", async () => {
const calls = [];
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_status_skill",
traceId: "trc_m3_status_skill",
message: "读取 M3 status 聚合状态"
},
{
now: () => "2026-05-23T00:09:10.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667",
OPENAI_API_KEY: "must-not-be-used"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 status");
},
m3IoSkillRequestJson: async (url, request) => {
calls.push({ url, request });
assert.equal(new URL(url).pathname, HWLAB_M3_STATUS_API_ROUTE);
assert.equal(request.method, "GET");
return {
ok: true,
status: 200,
body: {
status: "live",
sourceKind: "DEV-LIVE",
traceId: "trc_m3_status_skill",
boxes: [{
id: "boxsimu_2",
resourceId: "res_boxsimu_2",
online: true,
ports: {
DI1: {
value: false
}
}
}],
patchPanel: {
serviceId: "hwlab-patch-panel",
observable: true,
connectionActive: true
},
trust: {
operationId: "op_m3_status_skill",
traceId: "trc_m3_status_skill",
auditId: "aud_m3_status_skill",
evidenceId: "evd_m3_status_skill",
durableStatus: "green",
blocker: null,
readStatus: {
audit: "read",
evidence: "read"
},
runtime: {
durable: true
}
}
}
};
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.readonly);
assert.equal(payload.toolCalls[0].route, HWLAB_M3_STATUS_API_ROUTE);
assert.equal(payload.toolCalls[0].method, "GET");
assert.equal(payload.toolCalls[0].accepted, true);
assert.equal(payload.toolCalls[0].operationId, "op_m3_status_skill");
assert.equal(payload.toolCalls[0].readback.value, false);
assert.equal(payload.runnerTrace.route, HWLAB_M3_STATUS_API_ROUTE);
assert.equal(payload.providerTrace.method, "GET");
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.equal(calls.length, 1);
});
test("Code Agent blocks direct gateway or patch-panel requests instead of using M3 Skill CLI", async () => {
const direct = await handleCodeAgentChat(
{
conversationId: "cnv_m3_direct_blocked",
traceId: "trc_m3_direct_blocked",
message: "请直接调用 gateway-simu /invoke 写入 M3 DO1 true"
},
{
now: () => "2026-05-23T00:07:00.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
HWLAB_CODE_AGENT_HWLAB_API_BASE_URL: "http://hwlab-cloud-api.hwlab-dev.svc.cluster.local:6667"
},
m3IoSkillRequestJson: async () => {
throw new Error("direct gateway request must not reach the M3 skill CLI");
}
}
);
validateCodeAgentChatSchema(direct);
assert.equal(direct.status, "failed");
assert.equal(direct.error.code, "security_blocked");
assert.equal(direct.error.layer, "security");
assert.equal(direct.error.retryable, false);
assert.match(direct.error.userMessage, /安全边界/u);
assert.equal(direct.error.blocker.toolName, "security.hardware-boundary");
assert.equal(direct.toolCalls[0].name, "security.hardware-boundary");
assert.match(direct.error.message, /Skill CLI -> HWLAB API \/v1\/m3\/io/u);
});
function delay(ms) {
return new Promise((resolve) => setTimeout(resolve, ms));
}
async function createFakeCodexCommand() {
const root = await mkdtemp(path.join(os.tmpdir(), "hwlab-fake-codex-"));
const packageRoot = path.join(root, "node_modules", "@openai", "codex");
const command = path.join(packageRoot, "bin", "codex.js");
const native = path.join(root, "node_modules", "@openai", "codex-linux-x64", "vendor", "x86_64-unknown-linux-musl", "codex", "codex");
await mkdir(path.dirname(command), { recursive: true });
await mkdir(path.dirname(native), { recursive: true });
await writeFile(path.join(packageRoot, "package.json"), JSON.stringify({ name: "@openai/codex", version: "0.128.0" }), "utf8");
await writeFile(command, "#!/usr/bin/env node\nconsole.log('codex 0.128.0');\n", "utf8");
await writeFile(native, "#!/bin/sh\nexit 0\n", "utf8");
await chmod(command, 0o755);
await chmod(native, 0o755);
return { root, command };
}
test("Code Agent M3 Skill CLI blocks missing service-local HWLAB API base URL before loopback fallback", async () => {
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_m3_missing_api_base",
traceId: "trc_m3_missing_api_base",
message: "通过 HWLAB API 把 res_boxsimu_1 的 DO1 写成 true,然后读取 res_boxsimu_2 的 DI1。"
},
{
now: () => "2026-05-23T00:07:30.000Z",
env: {
PATH: process.env.PATH,
HWLAB_CODE_AGENT_WORKSPACE: process.cwd(),
OPENAI_API_KEY: "must-not-be-used"
},
callProvider: async () => {
throw new Error("OpenAI fallback must not be used for M3 IO");
},
m3IoSkillRequestJson: async () => {
throw new Error("missing HWLAB API base URL must block before any HTTP request");
}
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "completed");
assert.equal(payload.provider, "hwlab-skill-cli");
assert.equal(payload.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.toolCalls.length, 1);
assert.equal(payload.toolCalls[0].status, "blocked");
assert.equal(payload.toolCalls[0].blocker.code, "skill_cli_api_base_missing");
assert.equal(payload.toolCalls[0].blocker.layer, "skill-cli-config");
assert.equal(payload.toolCalls[0].blocker.retryable, false);
assert.deepEqual(payload.toolCalls[0].blocker.missingConfig, [
"HWLAB_CODE_AGENT_HWLAB_API_BASE_URL",
"HWLAB_API_BASE_URL",
"HWLAB_CLOUD_API_BASE_URL",
"contract:hwlab-agent-runtime.m3-io.apiBaseUrl"
]);
assert.equal(payload.toolCalls[0].hwlabApi.source, "missing-config");
assert.equal(payload.toolCalls[0].hwlabApi.redactedUrl, null);
assert.equal(payload.toolCalls[0].command.includes("127.0.0.1:6667"), false);
assert.equal(payload.toolCalls[0].accepted, false);
assert.equal(payload.toolCalls[0].operationId, null);
assert.equal(payload.session.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.runnerTrace.blocker.code, "skill_cli_api_base_missing");
assert.equal(payload.runnerTrace.capabilityLevel, HWLAB_M3_IO_CAPABILITY_LEVELS.blocked);
assert.equal(payload.providerTrace.fallbackUsed, false);
assert.equal(payload.skills.blockers[0].code, "skill_cli_api_base_missing");
});
test("Codex stdio manager reports concrete blockers without falling back to readonly", async () => {
const manager = createCodexStdioSessionManager({
idFactory: () => "ses_stdio_blocked"
});
const blocked = await handleCodeAgentChat(
{
conversationId: "cnv_stdio_blocked",
traceId: "trc_stdio_blocked",
message: "请用同一个 Code Agent session 继续"
},
{
now: () => "2026-05-23T00:05:00.000Z",
codexStdioManager: manager,
env: {
PATH: "",
HWLAB_CODE_AGENT_PROVIDER: "codex-stdio",
HWLAB_CODE_AGENT_MODEL: "gpt-test",
HWLAB_CODE_AGENT_CODEX_COMMAND: "/tmp/hwlab-missing-codex",
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
}
}
);
assert.equal(blocked.status, "failed");
assert.equal(blocked.provider, "codex-stdio");
assert.equal(blocked.backend, "hwlab-cloud-api/codex-mcp-stdio");
assert.equal(blocked.error.code, "codex_stdio_blocked");
assert.equal(blocked.error.layer, "runner");
assert.equal(blocked.error.retryable, false);
assert.match(blocked.error.userMessage, /Codex stdio/u);
assert.equal(blocked.error.blocker.runner, "codex-mcp-stdio-runner");
assert.ok(blocked.error.blockers.some((blocker) => blocker.code === "codex_cli_binary_missing"));
assert.ok(blocked.error.blockers.some((blocker) => blocker.code === "provider_token_boundary"));
assert.equal(blocked.capabilityLevel, "blocked");
assert.equal(blocked.longLivedSessionGate.status, "blocked");
assert.equal(blocked.availability.ready, false);
assert.equal(blocked.availability.codexStdio.status, "blocked");
assert.equal(Object.hasOwn(blocked, "reply"), false);
});
test("Codex stdio feasibility verifies writable workspace, CODEX_HOME, and version-probed command", async () => {
const workspace = await mkdtemp(path.join(os.tmpdir(), "hwlab-stdio-workspace-"));
const codexHome = await mkdtemp(path.join(os.tmpdir(), "hwlab-codex-home-"));
const fakeCodex = await createFakeCodexCommand();
try {
const manager = createCodexStdioSessionManager();
const availability = manager.describe({
env: {
PATH: process.env.PATH,
OPENAI_API_KEY: "test-openai-key-material",
CODEX_HOME: codexHome,
HWLAB_CODE_AGENT_CODEX_COMMAND: fakeCodex.command,
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_CODEX_WORKSPACE: workspace,
HWLAB_CODE_AGENT_CODEX_SANDBOX: "workspace-write",
HWLAB_CODE_AGENT_OPENAI_BASE_URL: "http://172.26.26.227:17680/v1/responses"
}
});
assert.equal(availability.command, fakeCodex.command);
assert.equal(availability.binary.present, true);
assert.equal(availability.binary.versionDetected, true);
assert.equal(availability.workspace, workspace);
assert.equal(availability.workspaceState.exists, true);
assert.equal(availability.workspaceState.readable, true);
assert.equal(availability.workspaceState.writable, true);
assert.equal(availability.runtimeContract.workspaceMount.status, "ready");
assert.equal(availability.codexHome, codexHome);
assert.equal(availability.codexHomeState.exists, true);
assert.equal(availability.codexHomeState.readable, true);
assert.equal(availability.codexHomeState.writable, true);
assert.equal(availability.runtimeContract.codexHome.status, "ready");
assert.equal(availability.startupReady, true);
assert.equal(availability.canStartLongLivedCodexStdio, true);
assert.equal(availability.ready, false);
assert.ok(availability.blockerCodes.includes("stdio_protocol_not_wired"));
assert.equal(availability.blockerCodes.includes("workspace_mount_missing"), false);
assert.equal(availability.blockerCodes.includes("workspace_write_boundary_blocked"), false);
assert.equal(availability.blockerCodes.includes("codex_home_missing"), false);
assert.equal(availability.blockerCodes.includes("codex_home_write_blocked"), false);
} finally {
await rm(workspace, { recursive: true, force: true });
await rm(codexHome, { recursive: true, force: true });
await rm(fakeCodex.root, { recursive: true, force: true });
}
});
test("repo-owned Codex stdio manager creates and reuses long-lived sessions with trace evidence", async () => {
const calls = [];
const fakeCodex = await createFakeCodexCommand();
const manager = createCodexStdioSessionManager({
idFactory: () => "ses_stdio_ready",
createRpcClient: async () => ({
async initialize() {
return { tools: ["codex", "codex-reply"] };
},
async listTools() {
return ["codex", "codex-reply"];
},
async callTool(name, args) {
calls.push({ name, args });
return {
structuredContent: {
threadId: "thread_stdio_ready",
content: name === "codex" ? "first stdio reply" : "second stdio reply"
}
};
},
close() {}
})
});
const env = {
PATH: process.env.PATH,
OPENAI_API_KEY: "test-openai-key-material",
CODEX_HOME: process.cwd(),
HWLAB_CODE_AGENT_PROVIDER: "codex-stdio",
HWLAB_CODE_AGENT_MODEL: "gpt-test",
HWLAB_CODE_AGENT_CODEX_COMMAND: fakeCodex.command,
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
};
const first = await handleCodeAgentChat(
{
conversationId: "cnv_stdio_ready",
traceId: "trc_stdio_ready_1",
message: "first"
},
{
now: () => "2026-05-23T00:06:00.000Z",
codexStdioManager: manager,
env
}
);
validateCodeAgentChatSchema(first);
assert.equal(first.status, "completed");
assert.equal(first.provider, "codex-stdio");
assert.equal(first.runner.kind, "codex-mcp-stdio-runner");
assert.equal(first.sessionMode, "codex-mcp-stdio-long-lived");
assert.equal(first.implementationType, "repo-owned-codex-mcp-stdio-session");
assert.equal(first.session.longLivedSession, true);
assert.equal(first.session.codexStdio, true);
assert.equal(first.runner.writeCapable, true);
assert.equal(first.runner.durableSession, true);
assert.equal(first.capabilityLevel, "long-lived-codex-stdio-session");
assert.equal(first.longLivedSessionGate.status, "pass");
assert.equal(first.longLivedSessionGate.pass, true);
assert.deepEqual(first.runnerLimitations, ["hardware-control-via-cloud-api-only", "secret-values-redacted"]);
assert.equal(first.runnerLimitations.includes("not-codex-stdio"), false);
assert.equal(first.runnerLimitations.includes("not-write-capable"), false);
assert.equal(first.runnerLimitations.includes("process-local-session-registry"), false);
assert.equal(classifyCodexRunnerCapability(first, { httpStatus: 200 }).capabilityPass, true);
const second = await handleCodeAgentChat(
{
conversationId: "cnv_stdio_ready",
traceId: "trc_stdio_ready_2",
message: "second"
},
{
now: () => "2026-05-23T00:06:01.000Z",
codexStdioManager: manager,
env
}
);
assert.equal(second.status, "completed");
assert.equal(second.sessionId, first.sessionId);
assert.equal(second.conversationId, first.conversationId);
assert.equal(second.sessionReuse.reused, true);
assert.equal(second.sessionReuse.turn, 2);
assert.equal(second.runnerTrace.sessionId, first.sessionId);
assert.equal(second.runnerTrace.sessionReused, true);
assert.equal(second.workspace, first.workspace);
assert.equal(second.sandbox, first.sandbox);
assert.equal(second.toolCalls[0].name, "codex-reply");
assert.equal(second.reply.content, "second stdio reply");
assert.equal(calls[0].name, "codex");
assert.equal(calls[1].name, "codex-reply");
assert.equal(calls[1].args.threadId, "thread_stdio_ready");
await rm(fakeCodex.root, { recursive: true, force: true });
});
test("Codex stdio runner startup failure marks session failed and blocks long-lived gate", async () => {
const fakeCodex = await createFakeCodexCommand();
const manager = createCodexStdioSessionManager({
idFactory: () => "ses_stdio_start_failed",
createRpcClient: async () => ({
async initialize() {
throw new Error("MCP initialize failed with sk-test-secret and TOKEN=abc123");
},
close() {}
})
});
const env = {
PATH: process.env.PATH,
OPENAI_API_KEY: "test-openai-key-material",
CODEX_HOME: process.cwd(),
HWLAB_CODE_AGENT_PROVIDER: "codex-stdio",
HWLAB_CODE_AGENT_MODEL: "gpt-test",
HWLAB_CODE_AGENT_CODEX_COMMAND: fakeCodex.command,
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
};
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_stdio_start_failed",
traceId: "trc_stdio_start_failed",
message: "启动后回答"
},
{
now: () => "2026-05-23T00:08:00.000Z",
codexStdioManager: manager,
env
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "failed");
assert.equal(payload.provider, "codex-stdio");
assert.equal(payload.error.code, "codex_stdio_failed");
assert.equal(payload.session.sessionId, "ses_stdio_start_failed");
assert.equal(payload.session.status, "failed");
assert.equal(payload.session.sessionMode, "codex-mcp-stdio-long-lived");
assert.equal(payload.session.statusReason, "codex_stdio_failed");
assert.equal(payload.longLivedSessionGate.status, "blocked");
assert.equal(payload.longLivedSessionGate.pass, false);
assert.ok(payload.longLivedSessionGate.blockers.some((blocker) => blocker.code === "session_failed"));
assert.equal(payload.capabilityLevel, "blocked");
assert.equal(payload.runnerTrace.sessionStatus, "failed");
assert.equal(payload.runnerTrace.valuesPrinted, false);
const serialized = JSON.stringify(payload);
assert.equal(serialized.includes("sk-test-secret"), false);
assert.equal(serialized.includes("TOKEN=abc123"), false);
assert.equal(classifyCodexRunnerCapability(payload, { httpStatus: 200 }).capabilityPass, false);
await rm(fakeCodex.root, { recursive: true, force: true });
});
test("Codex stdio protocol missing tools is a structured blocker, not a passed session", async () => {
const fakeCodex = await createFakeCodexCommand();
const manager = createCodexStdioSessionManager({
idFactory: () => "ses_stdio_missing_tool",
createRpcClient: async () => ({
async initialize() {
return { tools: ["codex"] };
},
async listTools() {
return ["codex"];
},
close() {}
})
});
const env = {
PATH: process.env.PATH,
OPENAI_API_KEY: "test-openai-key-material",
CODEX_HOME: process.cwd(),
HWLAB_CODE_AGENT_PROVIDER: "codex-stdio",
HWLAB_CODE_AGENT_MODEL: "gpt-test",
HWLAB_CODE_AGENT_CODEX_COMMAND: fakeCodex.command,
HWLAB_CODE_AGENT_CODEX_STDIO_ENABLED: "1",
HWLAB_CODE_AGENT_CODEX_STDIO_SUPERVISOR: "repo-owned",
HWLAB_CODE_AGENT_WORKSPACE: process.cwd()
};
const payload = await handleCodeAgentChat(
{
conversationId: "cnv_stdio_missing_tool",
traceId: "trc_stdio_missing_tool",
message: "需要长会话"
},
{
now: () => "2026-05-23T00:08:30.000Z",
codexStdioManager: manager,
env
}
);
validateCodeAgentChatSchema(payload);
assert.equal(payload.status, "failed");
assert.equal(payload.error.code, "codex_stdio_protocol_blocked");
assert.deepEqual(payload.error.missingTools, ["codex-reply"]);
assert.equal(payload.session.status, "failed");
assert.equal(payload.longLivedSessionGate.status, "blocked");
assert.ok(payload.longLivedSessionGate.blockers.some((blocker) => blocker.code === "session_failed"));
assert.equal(payload.availability.codexStdio.ready, false);
assert.ok(payload.availability.codexStdio.blockerCodes.includes("stdio_protocol_not_wired"));
assert.deepEqual(payload.codexStdioFeasibility.protocol.toolsObserved, ["codex"]);
assert.equal(classifyCodexRunnerCapability(payload, { httpStatus: 200 }).capabilityPass, false);
await rm(fakeCodex.root, { recursive: true, force: true });
});