fix: harden routed preview recovery boundaries (#2103)

* fix: harden routed preview recovery boundaries

* fix: secure routed browser command boundaries

* fix: complete browser cleanup security boundaries

* fix: preserve routed security across browser sessions

* fix: preserve canonical routing in shared scenarios

Author: chubes4

.github/workflows/agent-task-contracts.yml

@@ -14,6 +14,9 @@ on:
       - "fixtures/agent-task-runtime-paths-run-29305012941.json"
       - "package-lock.json"
       - "package.json"
+      - "packages/runtime-playground/src/**"
+      - "tests/browser-*.test.ts"
+      - "tests/editor-*.test.ts"
       - "tests/agent-task-*.test.ts"
       - "tests/runtime-sources-materialization.test.ts"
       - "tests/runtime-sources-playground-integration.test.ts"
@@ -58,6 +61,9 @@ on:
       - "fixtures/agent-task-runtime-paths-run-29305012941.json"
       - "package-lock.json"
       - "package.json"
+      - "packages/runtime-playground/src/**"
+      - "tests/browser-*.test.ts"
+      - "tests/editor-*.test.ts"
       - "tests/agent-task-*.test.ts"
       - "tests/runtime-sources-materialization.test.ts"
       - "tests/runtime-sources-playground-integration.test.ts"
@@ -129,6 +135,8 @@ jobs:
       - run: npm run test:trusted-apply-artifact-channel
       - run: npm run test:runtime-command-artifact-bounds
       - run: npm run test:redaction
+      - run: npm run test:browser-preview-routing
+      - run: npm run test:browser-routed-command-security
       - run: npm run test:production-boundary-enforcement
       - run: npm run test:runtime-tool-policy
 

package.json

@@ -109,6 +109,7 @@
     "smoke": "tsx scripts/run-smoke.ts",
     "test:redaction": "tsx tests/redaction.test.ts",
     "test:browser-preview-routing": "tsx --test tests/browser-preview-routing.test.ts",
+    "test:browser-routed-command-security": "tsx --test tests/browser-routed-command-security.test.ts",
     "test:cloudflare-runtime": "node --test tests/cloudflare-d1-provisioner.test.mjs && tsx tests/cloudflare-site-context.test.ts && tsx tests/cloudflare-coordinator-site-partitioning.test.ts && tsx tests/cloudflare-d1-operation-repository.test.ts && tsx tests/cloudflare-provisioning-api.test.ts && tsx tests/cloudflare-runtime.test.ts && node ./node_modules/typescript/bin/tsc -p packages/runtime-cloudflare --noEmit",
     "test:cloudflare-administrator-claim": "tsx tests/cloudflare-provisioning-api.test.ts",
     "test:cloudflare-wordpress-auth": "tsx tests/cloudflare-wordpress-auth.test.ts",

packages/runtime-playground/src/browser-actions-runner.ts

@@ -12,7 +12,8 @@ import { browserAssertionsSummary, browserStepRecord, executeBrowserInteractionS
 import { browserCommandLivenessPolicy, isBrowserCommandLivenessError, withBrowserCommandLiveness } from "./browser-liveness.js"
 import { serializeBrowserError } from "./browser-metrics.js"
 import { executeBrowserObservationAssertion } from "./browser-observation-assertions.js"
-import { browserPreviewNetworkPolicyIsActive, browserPreviewNetworkPolicySummary, browserPreviewNeedsContextRouting, browserPreviewReadinessError, browserPreviewTopology, resolveBrowserPreviewUrl, routeBrowserPreviewContextNetwork } from "./browser-preview-routing.js"
+import { browserPreviewNetworkPolicyIsActive, browserPreviewNetworkPolicySummary, browserPreviewNeedsContextRouting, browserPreviewReadinessError, browserPreviewTopology, closeBrowserAndDrainPreviewRoutes, createBrowserPreviewRouteTracker, resolveBrowserPreviewUrl, routeBrowserPreviewContextNetwork } from "./browser-preview-routing.js"
+import { browserCommandResult } from "./browser-result-sanitization.js"
 import { BROWSER_PROBE_STATE_INIT_SCRIPT, browserProbeReplayability, browserProbeViewport } from "./browser-probe.js"
 import { runBrowserProbeCommand, type BrowserProbeRunPlan } from "./browser-probe-runner.js"
 import { browserActionTargetUrls, browserAuthRequest, browserProbeWaterfallArtifact, browserProbeWebSocketArtifact, browserProbeWebSocketSummary, browserRedirectDiagnosticsArtifact, browserRequestCoverageArtifact, browserStorageStateAuthSummary, browserStorageStateImportFromArgs, browserWordPressDiagnosticsArtifact, createBrowserProbeProgressTracker, fileSha256, installBrowserWordPressDiagnostics, installWordPressAdminAuthCookies, livenessRemainingWallTimeMs, normalizeBrowserProbeScriptCheckpoint, type BrowserCommandProgressEvent, type BrowserStorageStateImport } from "./browser-probe-support.js"
@@ -133,6 +134,7 @@ export async function runBrowserActionsCommand({
   const browser = session?.browser ?? await launchChromiumBrowser()
   const topology = browserPreviewTopology(args, runtimeSpec, server.serverUrl, server.previewProxyDiagnostics?.targetOrigin)
   const { preview, networkPolicy } = topology
+  const routeTracker = session?.routeTracker ?? createBrowserPreviewRouteTracker()
   let requestedUrl = initialUrl ? topology.resolveUrl(initialUrl) : preview.effectiveOrigin
   let finalUrl = requestedUrl
   let htmlSha256: string | undefined
@@ -173,7 +175,7 @@ export async function runBrowserActionsCommand({
     }) : undefined)
     const context = environmentRuntime?.context ?? null
     if (context && !session) {
-      await routeBrowserPreviewContextNetwork(context, networkPolicy, topology.origins.localProxyOrigin)
+      await routeBrowserPreviewContextNetwork(context, networkPolicy, topology.origins.localProxyOrigin, routeTracker)
     }
     const page = activePage = environmentRuntime?.page ?? await browser.newPage()
     if (onProgress) {
@@ -437,11 +439,12 @@ export async function runBrowserActionsCommand({
     pendingError = error instanceof Error ? error : new Error(String(error))
     errors.push(serializeBrowserError("probe-error", error))
   } finally {
-    await settleBrowserNetworkTasks(networkTasks, livenessPolicy.networkSettleTimeoutMs)
+    await settleBrowserNetworkTasks(networkTasks, livenessPolicy.networkSettleTimeoutMs).catch((error) => errors.push(serializeBrowserError("probe-error", error)))
     if (activePage && resolvedEnvironment) environmentEvidence = await observePlaywrightBrowserEnvironment(activePage, requestedEnvironment, resolvedEnvironment).catch(() => environmentEvidence)
-    if (!session) {
-      await environmentRuntime?.close().catch(() => undefined)
-      await browser.close()
+    const cleanupBrowser = session ? { close: async () => {} } : { close: async () => { await environmentRuntime?.close(); await browser.close() } }
+    for (const routeError of await closeBrowserAndDrainPreviewRoutes(cleanupBrowser, routeTracker)) {
+      errors.push(serializeBrowserError("probe-error", routeError))
+      pendingError ??= routeError
     }
     if (capture.has("steps")) {
       await artifactSession.writeJsonLines("steps", "steps.jsonl", stepRecords)
@@ -580,9 +583,7 @@ export async function runBrowserActionsCommand({
     throw new Error("wordpress.browser-actions did not produce a browser artifact")
   }
 
-  return {
-    artifact,
-    output: `${JSON.stringify({
+  return browserCommandResult(artifact, {
       command: "wordpress.browser-actions",
       requestedUrl,
       preview,
@@ -591,8 +592,7 @@ export async function runBrowserActionsCommand({
       files: artifact.files,
       summary: artifact.summary,
       steps: stepRecords,
-    }, null, 2)}\n`,
-  }
+  })
 }
 
 export function browserToolVerifierUnsupportedResult(step: BrowserInteractionStep, index: number, startedAt: string): BrowserToolVerifierResult {
@@ -967,6 +967,7 @@ export async function runBrowserScenarioCommand({
   let pendingError: Error | undefined
   let scenarioSession: PlaywrightBrowserEnvironmentSession | undefined
   let scenarioBrowser: Awaited<ReturnType<typeof launchChromiumBrowser>> | undefined
+  let scenarioRouteTracker: ReturnType<typeof createBrowserPreviewRouteTracker> | undefined
 
   try {
     if (runPlan.probe && runPlan.actions) {
@@ -976,12 +977,16 @@ export async function runBrowserScenarioCommand({
       if (unsupported.length > 0) {
         throw new Error(`wordpress.browser-scenario browser environment is unsupported: ${unsupported.join(", ")}`)
       }
+      const topology = browserPreviewTopology(args, runtimeSpec, server.serverUrl, server.previewProxyDiagnostics?.targetOrigin)
       const runtime = await createPlaywrightBrowserEnvironmentContext(browser, resolved, {
-        ...(runPlan.actions.storageStateImport ? { contextOptions: { storageState: runPlan.actions.storageStateImport.storageState } } : {}),
+        contextOptions: {
+          ...topology.contextOptions(),
+          ...(runPlan.actions.storageStateImport ? { storageState: runPlan.actions.storageStateImport.storageState } : {}),
+        },
       })
-      const topology = browserPreviewTopology(args, runtimeSpec, server.serverUrl)
-      if (browserPreviewNeedsContextRouting(topology.networkPolicy)) await routeBrowserPreviewContextNetwork(runtime.context, topology.networkPolicy, topology.preview.effectiveOrigin)
-      scenarioSession = { browser, requested: requestedEnvironment, resolved, runtime }
+      const routeTracker = scenarioRouteTracker = createBrowserPreviewRouteTracker()
+      if (browserPreviewNeedsContextRouting(topology.networkPolicy)) await routeBrowserPreviewContextNetwork(runtime.context, topology.networkPolicy, topology.origins.localProxyOrigin, routeTracker)
+      scenarioSession = { browser, requested: requestedEnvironment, resolved, routeTracker, runtime }
     }
 
     if (runPlan.probe) {
@@ -1006,8 +1011,15 @@ export async function runBrowserScenarioCommand({
       }
     }
   } finally {
-    await scenarioSession?.runtime.close().catch(() => undefined)
-    await scenarioBrowser?.close().catch(() => undefined)
+    if (scenarioSession && scenarioBrowser && scenarioRouteTracker) {
+      const activeSession = scenarioSession
+      const activeBrowser = scenarioBrowser
+      const cleanupBrowser = { close: async () => { await activeSession.runtime.close(); await activeBrowser.close() } }
+      const routeErrors = await closeBrowserAndDrainPreviewRoutes(cleanupBrowser, scenarioRouteTracker)
+      pendingError ??= routeErrors[0]
+    } else {
+      await scenarioBrowser?.close().catch(() => undefined)
+    }
   }
 
   const primaryArtifact = actionsResult?.artifact ?? probeResult?.artifact
@@ -1067,17 +1079,14 @@ export async function runBrowserScenarioCommand({
     throw new BrowserCommandArtifactError(`wordpress.browser-scenario failed: ${pendingError.message}`, artifact)
   }
 
-  return {
-    artifact,
-    output: `${JSON.stringify({
+  return browserCommandResult(artifact, {
       command: "wordpress.browser-scenario",
       requestedUrl: artifact.requestedUrl,
       finalUrl,
       files: artifact.files,
       summary: artifact.summary,
       scenario: scenarioSummary,
-    }, null, 2)}\n`,
-  }
+  })
 }
 
 async function browserScenarioFromArgs(args: string[]): Promise<BrowserScenarioInput> {

packages/runtime-playground/src/browser-artifact-session.ts

@@ -3,6 +3,7 @@ import { basename, join } from "node:path"
 import type { ArtifactProvenanceMetadata } from "@automattic/wp-codebox-core"
 import { ArtifactBundleWriter } from "./artifact-bundle-writer.js"
 import { browserArtifactFileManifest, type BrowserArtifactFiles } from "./browser-artifacts.js"
+import { sanitizeBrowserResultValue } from "./browser-result-sanitization.js"
 
 export class BrowserArtifactSession {
   readonly writer: ArtifactBundleWriter
@@ -32,11 +33,11 @@ export class BrowserArtifactSession {
   }
 
   async writeJson(key: keyof BrowserArtifactFiles, fileName: string, value: unknown): Promise<void> {
-    await this.writer.writeJson(this.path(fileName), value, this.manifestWithoutContentType(key))
+    await this.writer.writeJson(this.path(fileName), sanitizeBrowserResultValue(value), this.manifestWithoutContentType(key))
   }
 
   async writeJsonLines(key: keyof BrowserArtifactFiles, fileName: string, records: unknown[]): Promise<void> {
-    await this.writer.writeJsonLines(this.path(fileName), records, this.manifestWithoutContentType(key))
+    await this.writer.writeJsonLines(this.path(fileName), sanitizeBrowserResultValue(records), this.manifestWithoutContentType(key))
   }
 
   async writeGenerated(key: keyof BrowserArtifactFiles, fileName: string, write: (absolutePath: string) => Promise<void>): Promise<void> {

packages/runtime-playground/src/browser-command-artifact-error.ts

@@ -1,9 +1,14 @@
 import type { BrowserArtifact } from "./browser-artifacts.js"
+import { sanitizeBrowserArtifact, sanitizeBrowserResultValue } from "./browser-result-sanitization.js"
 
 export class BrowserCommandArtifactError extends Error {
-  constructor(message: string, readonly artifact: BrowserArtifact, readonly artifactRoot?: string) {
-    super(message)
+  readonly artifact: BrowserArtifact
+
+  constructor(message: string, artifact: BrowserArtifact, readonly artifactRoot?: string) {
+    super(sanitizeBrowserResultValue(message, "message"))
     this.name = "BrowserCommandArtifactError"
+    Object.assign(artifact, sanitizeBrowserArtifact(artifact))
+    this.artifact = artifact
   }
 }
 

packages/runtime-playground/src/browser-environment-matrix.ts

@@ -13,6 +13,7 @@ import {
 } from "@automattic/wp-codebox-core"
 import type { Browser, BrowserContext, BrowserContextOptions, Page } from "playwright"
 import type { BrowserArtifactSummary } from "./browser-artifacts.js"
+import type { BrowserPreviewRouteTracker } from "./browser-preview-routing.js"
 
 export const PLAYWRIGHT_BROWSER_ENVIRONMENT_CAPABILITIES = [
   "browser.environment.viewport",
@@ -75,6 +76,7 @@ export interface PlaywrightBrowserEnvironmentSession {
   browser: Browser
   requested: BrowserEnvironment
   resolved: ResolvedBrowserEnvironment
+  routeTracker?: BrowserPreviewRouteTracker
   runtime: PlaywrightBrowserEnvironmentRuntime
 }
 

packages/runtime-playground/src/browser-metrics.ts

@@ -527,7 +527,7 @@ export async function serializeBrowserFinishedRequest(request: Request, timestam
   if (!response) {
     return {
       type: "response",
-      url: request.url(),
+      url: redactBrowserNetworkUrl(request.url()),
       method: request.method(),
       resourceType: request.resourceType(),
       timestamp,
@@ -545,7 +545,7 @@ export async function serializeBrowserResponse(response: Response, timestamp = n
   const responseTextPreview = await browserDocument5xxResponsePreview(response)
   return {
     type: "response",
-    url: response.url(),
+    url: redactBrowserNetworkUrl(response.url()),
     method: request.method(),
     resourceType: request.resourceType(),
     status: response.status(),
@@ -596,17 +596,22 @@ function redactBrowserResponseText(body: string): string {
 }
 
 export function serializeBrowserRequestFailure(request: Request, timestamp = now()): BrowserProbeNetworkRecord {
+  const failure = request.failure()
   return {
     type: "requestfailed",
-    url: request.url(),
+    url: redactBrowserNetworkUrl(request.url()),
     method: request.method(),
     resourceType: request.resourceType(),
     timing: browserRequestTiming(request),
-    failure: request.failure(),
+    failure: failure ? { errorText: redactString(failure.errorText, { redactAllUrlQueryValues: true, redactUrlHash: true, redactQueryAssignments: true }) } : null,
     timestamp,
   }
 }
 
+function redactBrowserNetworkUrl(url: string): string {
+  return redactString(url, { redactAllUrlQueryValues: true, redactUrlHash: true, redactQueryAssignments: true })
+}
+
 function browserRequestTiming(request: Request): Record<string, number> {
   return Object.fromEntries(
     Object.entries(request.timing()).filter((entry): entry is [string, number] => typeof entry[1] === "number" && Number.isFinite(entry[1])),

packages/runtime-playground/src/browser-multi-actor-scenario-runner.ts

@@ -6,7 +6,8 @@ import { BrowserCommandArtifactError } from "./browser-command-artifact-error.js
 import { attachBrowserCaptureListeners, launchChromiumBrowser, settleBrowserNetworkTasks } from "./browser-capture-session.js"
 import { executeBrowserInteractionStep } from "./browser-interactions.js"
 import { browserProbeReplayability } from "./browser-probe.js"
-import { browserPreviewReadinessError, browserPreviewTopology, routeBrowserPreviewContextNetwork } from "./browser-preview-routing.js"
+import { browserPreviewReadinessError, browserPreviewTopology, closeBrowserAndDrainPreviewRoutes, createBrowserPreviewRouteTracker, routeBrowserPreviewContextNetwork } from "./browser-preview-routing.js"
+import { browserCommandResult } from "./browser-result-sanitization.js"
 import { installWordPressAdminAuthCookies } from "./browser-probe-support.js"
 import { bootstrapPhpCode } from "./php-bootstrap.js"
 import { assertPlaygroundResponseOk, type PlaygroundRunResponse } from "./playground-command-errors.js"
@@ -33,6 +34,7 @@ export async function runBrowserMultiActorScenarioCommand(input: {
   const browserArgs = routeHost && !scenario.browserArgs.some((arg) => arg.startsWith("route-host=")) ? [...scenario.browserArgs, `route-host=${routeHost}`] : scenario.browserArgs
   const topology = browserPreviewTopology(browserArgs, runtimeSpec, server.serverUrl, server.previewProxyDiagnostics?.targetOrigin)
   const browser = await launchChromiumBrowser()
+  const routeTracker = createBrowserPreviewRouteTracker()
   const evidence: Record<string, ActorEvidence> = {}
   let result: BrowserMultiActorScenarioResult | undefined
   let failure: Error | undefined
@@ -56,7 +58,7 @@ export async function runBrowserMultiActorScenarioCommand(input: {
       const userId = await actorUserId(actor.name, session.user.userId, session.user, runtimeSpec, runPlaygroundCommand, server)
       const environmentRuntime = await createPlaywrightBrowserEnvironmentContext(browser, resolvedEnvironment, { contextOptions: topology.contextOptions() })
       const context = environmentRuntime.context
-      await routeBrowserPreviewContextNetwork(context, topology.networkPolicy, topology.origins.localProxyOrigin)
+      await routeBrowserPreviewContextNetwork(context, topology.networkPolicy, topology.origins.localProxyOrigin, routeTracker)
       const page = environmentRuntime.page
       await context.tracing.start({ screenshots: true, snapshots: true })
       await installWordPressAdminAuthCookies({ command: "wordpress.browser-scenario", cookieUrls: topology.authCookieUrls([topology.resolveUrl(scenario.url)]), page, runPlaygroundCommand, runtimeSpec, server, userId })
@@ -73,7 +75,8 @@ export async function runBrowserMultiActorScenarioCommand(input: {
     failure = error instanceof Error ? error : new Error(String(error))
     if (error instanceof BrowserMultiActorScenarioError) result = error.result
   } finally {
-    await browser.close()
+    const routeErrors = await closeBrowserAndDrainPreviewRoutes(browser, routeTracker)
+    failure ??= routeErrors[0]
   }
 
   const replay = result?.replay ?? { schema: "wp-codebox/browser-multi-actor-replay/v1", seed: scenario.seed, scenario, schedule: [] }
@@ -91,7 +94,7 @@ export async function runBrowserMultiActorScenarioCommand(input: {
   const traces = Object.values(evidence).map((actor) => actor.files.trace).filter((path): path is string => Boolean(path))
   const artifact = { artifactType: "scenario" as const, requestedUrl: target, url: target, preview: topology.preview, ...topology.origins, files: { summary: "files/browser/multi-actor-scenario-summary.json", steps: "files/browser/multi-actor-events.json", network: "files/browser/multi-actor-network.json", requestCoverage: "files/browser/multi-actor-request-coverage.json", waterfall: "files/browser/multi-actor-waterfall.json", ...(traces.length > 0 ? { traces } : {}) }, summary: { actions: scenario.actions.length, steps: scenario.actions.length, consoleMessages: Object.values(evidence).reduce((total, actor) => total + actor.console.length, 0), errors: Object.values(evidence).reduce((total, actor) => total + actor.errors.length, 0), finalUrl: target, htmlSnapshot: false, networkEvents: network.length, replayability: browserProbeReplayability(captures), screenshot: captures.has("screenshot"), viewport: null, environment: Object.values(evidence)[0]?.environment, multiActor: { seed: scenario.seed, finalState: result?.finalState ?? "failed", actors: Object.keys(evidence), replay: "files/browser/multi-actor-replay.json" } } } satisfies BrowserArtifact
   if (failure) throw new BrowserCommandArtifactError(`wordpress.browser-scenario failed: ${failure.message}`, artifact)
-  return { artifact, output: `${JSON.stringify({ command: "wordpress.browser-scenario", files: artifact.files, summary: artifact.summary, scenario: summary }, null, 2)}\n` }
+  return browserCommandResult(artifact, { command: "wordpress.browser-scenario", files: artifact.files, summary: artifact.summary, scenario: summary })
 }
 
 export async function navigateBrowserMultiActorPages(