fix(qa-lab): harden telegram qa artifacts

extensions/qa-lab/src/cli.runtime.test.ts

@@ -57,6 +57,7 @@ vi.mock("./docker-up.runtime.js", () => ({
 }));
 
 import {
+  __testing,
   runQaLabSelfCheckCommand,
   runQaDockerBuildImageCommand,
   runQaDockerScaffoldCommand,
@@ -185,6 +186,15 @@ describe("qa cli runtime", () => {
     });
   });
 
+  it("rejects output dirs that escape the repo root", () => {
+    expect(() =>
+      __testing.resolveRepoRelativeOutputDir("/tmp/openclaw-repo", "../outside"),
+    ).toThrow("--output-dir must stay within the repo root.");
+    expect(() =>
+      __testing.resolveRepoRelativeOutputDir("/tmp/openclaw-repo", "/tmp/outside"),
+    ).toThrow("--output-dir must be a relative path inside the repo root.");
+  });
+
   it("defaults telegram qa runs onto the live provider lane", async () => {
     await runQaTelegramCommand({
       repoRoot: "/tmp/openclaw-repo",

extensions/qa-lab/src/cli.runtime.ts

@@ -22,6 +22,21 @@ type InterruptibleServer = {
   stop(): Promise<void>;
 };
 
+function resolveRepoRelativeOutputDir(repoRoot: string, outputDir?: string) {
+  if (!outputDir) {
+    return undefined;
+  }
+  if (path.isAbsolute(outputDir)) {
+    throw new Error("--output-dir must be a relative path inside the repo root.");
+  }
+  const resolved = path.resolve(repoRoot, outputDir);
+  const relative = path.relative(repoRoot, resolved);
+  if (relative.startsWith("..") || path.isAbsolute(relative)) {
+    throw new Error("--output-dir must stay within the repo root.");
+  }
+  return resolved;
+}
+
 function resolveQaManualLaneModels(opts: {
   providerMode: QaProviderMode;
   primaryModel?: string;
@@ -242,7 +257,7 @@ export async function runQaSuiteCommand(opts: {
   if (runner === "multipass") {
     const result = await runQaMultipass({
       repoRoot,
-      outputDir: opts.outputDir ? path.resolve(repoRoot, opts.outputDir) : undefined,
+      outputDir: resolveRepoRelativeOutputDir(repoRoot, opts.outputDir),
       providerMode,
       primaryModel: opts.primaryModel,
       alternateModel: opts.alternateModel,
@@ -265,7 +280,7 @@ export async function runQaSuiteCommand(opts: {
   }
   const result = await runQaSuiteFromRuntime({
     repoRoot,
-    outputDir: opts.outputDir ? path.resolve(repoRoot, opts.outputDir) : undefined,
+    outputDir: resolveRepoRelativeOutputDir(repoRoot, opts.outputDir),
     providerMode,
     primaryModel: opts.primaryModel,
     alternateModel: opts.alternateModel,
@@ -296,7 +311,7 @@ export async function runQaTelegramCommand(opts: {
     opts.providerMode === undefined ? "live-frontier" : normalizeQaProviderMode(opts.providerMode);
   const result = await runTelegramQaLive({
     repoRoot,
-    outputDir: opts.outputDir ? path.resolve(repoRoot, opts.outputDir) : undefined,
+    outputDir: resolveRepoRelativeOutputDir(repoRoot, opts.outputDir),
     providerMode,
     primaryModel: opts.primaryModel,
     alternateModel: opts.alternateModel,
@@ -328,7 +343,7 @@ export async function runQaCharacterEvalCommand(opts: {
   const judges = parseQaModelSpecs("--judge-model", opts.judgeModel);
   const result = await runQaCharacterEval({
     repoRoot,
-    outputDir: opts.outputDir ? path.resolve(repoRoot, opts.outputDir) : undefined,
+    outputDir: resolveRepoRelativeOutputDir(repoRoot, opts.outputDir),
     models: candidates.models,
     scenarioId: opts.scenario,
     candidateFastMode: opts.fast,
@@ -420,7 +435,10 @@ export async function runQaDockerScaffoldCommand(opts: {
   bindUiDist?: boolean;
 }) {
   const repoRoot = path.resolve(opts.repoRoot ?? process.cwd());
-  const outputDir = path.resolve(repoRoot, opts.outputDir);
+  const outputDir = resolveRepoRelativeOutputDir(repoRoot, opts.outputDir);
+  if (!outputDir) {
+    throw new Error("--output-dir is required.");
+  }
   const result = await writeQaDockerHarnessFiles({
     outputDir,
     repoRoot,
@@ -457,7 +475,7 @@ export async function runQaDockerUpCommand(opts: {
   const repoRoot = path.resolve(opts.repoRoot ?? process.cwd());
   const result = await runQaDockerUp({
     repoRoot,
-    outputDir: opts.outputDir ? path.resolve(repoRoot, opts.outputDir) : undefined,
+    outputDir: resolveRepoRelativeOutputDir(repoRoot, opts.outputDir),
     gatewayPort: Number.isFinite(opts.gatewayPort) ? opts.gatewayPort : undefined,
     qaLabPort: Number.isFinite(opts.qaLabPort) ? opts.qaLabPort : undefined,
     providerBaseUrl: opts.providerBaseUrl,
@@ -479,3 +497,7 @@ export async function runQaMockOpenAiCommand(opts: { host?: string; port?: numbe
   });
   await runInterruptibleServer("QA mock OpenAI", server);
 }
+
+export const __testing = {
+  resolveRepoRelativeOutputDir,
+};

extensions/qa-lab/src/gateway-child.ts

@@ -678,7 +678,10 @@ export async function startQaGatewayChild(params: {
     controlUiEnabled: params.controlUiEnabled,
   });
   const cfg = params.mutateConfig ? params.mutateConfig(baseCfg) : baseCfg;
-  await fs.writeFile(configPath, `${JSON.stringify(cfg, null, 2)}\n`, "utf8");
+  await fs.writeFile(configPath, `${JSON.stringify(cfg, null, 2)}\n`, {
+    encoding: "utf8",
+    mode: 0o600,
+  });
   const allowedPluginIds = [...(cfg.plugins?.allow ?? []), "openai"].filter(
     (pluginId, index, array): pluginId is string => {
       return (

extensions/qa-lab/src/telegram-live.runtime.test.ts

@@ -166,6 +166,43 @@ describe("telegram live qa runtime", () => {
     ).toBe("match");
   });
 
+  it("redacts observed message content by default in artifacts", () => {
+    expect(
+      __testing.buildObservedMessagesArtifact({
+        includeContent: false,
+        observedMessages: [
+          {
+            updateId: 1,
+            messageId: 9,
+            chatId: -100123,
+            senderId: 42,
+            senderIsBot: true,
+            senderUsername: "driver_bot",
+            text: "secret text",
+            caption: "secret caption",
+            replyToMessageId: 8,
+            timestamp: 1_700_000_000_000,
+            inlineButtons: ["Approve"],
+            mediaKinds: ["photo"],
+          },
+        ],
+      }),
+    ).toEqual([
+      {
+        updateId: 1,
+        messageId: 9,
+        chatId: -100123,
+        senderId: 42,
+        senderIsBot: true,
+        senderUsername: "driver_bot",
+        replyToMessageId: 8,
+        timestamp: 1_700_000_000_000,
+        inlineButtons: ["Approve"],
+        mediaKinds: ["photo"],
+      },
+    ]);
+  });
+
   it("formats phase-specific canary diagnostics with context", () => {
     const error = new Error(
       "SUT bot did not send any group reply after the canary command within 30s.",

extensions/qa-lab/src/telegram-live.runtime.ts

@@ -44,6 +44,11 @@ type TelegramObservedMessage = {
   mediaKinds: string[];
 };
 
+type TelegramObservedMessageArtifact = Omit<TelegramObservedMessage, "text" | "caption"> & {
+  text?: string;
+  caption?: string;
+};
+
 type TelegramQaScenarioResult = {
   id: string;
   title: string;
@@ -425,6 +430,28 @@ function renderTelegramQaMarkdown(params: {
   return lines.join("\n");
 }
 
+function buildObservedMessagesArtifact(params: {
+  observedMessages: TelegramObservedMessage[];
+  includeContent: boolean;
+}) {
+  return params.observedMessages.map<TelegramObservedMessageArtifact>((message) =>
+    params.includeContent
+      ? { ...message }
+      : {
+          updateId: message.updateId,
+          messageId: message.messageId,
+          chatId: message.chatId,
+          senderId: message.senderId,
+          senderIsBot: message.senderIsBot,
+          senderUsername: message.senderUsername,
+          replyToMessageId: message.replyToMessageId,
+          timestamp: message.timestamp,
+          inlineButtons: message.inlineButtons,
+          mediaKinds: message.mediaKinds,
+        },
+  );
+}
+
 function findScenario(ids?: string[]) {
   if (!ids || ids.length === 0) {
     return [...TELEGRAM_QA_SCENARIOS];
@@ -628,6 +655,7 @@ export async function runTelegramQaLive(params: {
   const sutAccountId = params.sutAccountId?.trim() || "sut";
   const scenarios = findScenario(params.scenarioIds);
   const observedMessages: TelegramObservedMessage[] = [];
+  const includeObservedMessageContent = process.env.OPENCLAW_QA_TELEGRAM_CAPTURE_CONTENT === "1";
   const startedAt = new Date().toISOString();
 
   const driverIdentity = await getBotIdentity(runtimeEnv.driverToken);
@@ -755,13 +783,23 @@ export async function runTelegramQaLive(params: {
       finishedAt,
       scenarios: scenarioResults,
     })}\n`,
-    "utf8",
+    { encoding: "utf8", mode: 0o600 },
   );
-  await fs.writeFile(summaryPath, `${JSON.stringify(summary, null, 2)}\n`, "utf8");
+  await fs.writeFile(summaryPath, `${JSON.stringify(summary, null, 2)}\n`, {
+    encoding: "utf8",
+    mode: 0o600,
+  });
   await fs.writeFile(
     observedMessagesPath,
-    `${JSON.stringify(observedMessages, null, 2)}\n`,
-    "utf8",
+    `${JSON.stringify(
+      buildObservedMessagesArtifact({
+        observedMessages,
+        includeContent: includeObservedMessageContent,
+      }),
+      null,
+      2,
+    )}\n`,
+    { encoding: "utf8", mode: 0o600 },
   );
   if (canaryFailure) {
     throw new Error(
@@ -781,6 +819,7 @@ export async function runTelegramQaLive(params: {
 export const __testing = {
   TELEGRAM_QA_SCENARIOS,
   buildTelegramQaConfig,
+  buildObservedMessagesArtifact,
   canaryFailureMessage,
   classifyCanaryReply,
   normalizeTelegramObservedMessage,