CVE-2026-35643

@@ -8,45 +8,47 @@ Docs: https://docs.openclaw.ai
 
 ### Changes
 
-- ModelStudio/Qwen: add standard (pay-as-you-go) DashScope endpoints for China and global Qwen API keys alongside the existing Coding Plan endpoints, and relabel the provider group to `Qwen (Alibaba Cloud Model Studio)`. (#43878)
-- UI/clarity: consolidate button primitives (`btn--icon`, `btn--ghost`, `btn--xs`), refine the Knot theme to a black-and-red palette with WCAG 2.1 AA contrast, add config icons for Diagnostics/CLI/Secrets/ACP/MCP sections, replace the roundness slider with discrete stops, and improve accessibility with aria-labels across usage filters. (#53272) Thanks @BunsDev.
-- CSP/Control UI: compute SHA-256 hashes for inline `<script>` blocks in the served `index.html` and include them in the `script-src` CSP directive, keeping inline scripts blocked by default while allowing explicitly hashed bootstrap code. (#53307) Thanks @BunsDev.
-
 ### Fixes
 
-- Plugins/ClawHub: resolve plugin API compatibility against the active runtime version at install time, and add regression coverage for current `>=2026.3.22` ClawHub package checks so installs no longer fail behind the stale `1.2.0` constant. (#53157) Thanks @futhgar.
-- CLI/channel auth: auto-select the single login-capable configured channel for `channels login`/`logout` instead of relying on the outbound message-channel resolver, so env-only or non-auth channels no longer cause false ambiguity errors. (#53254) Thanks @BunsDev.
-- Control UI/auth: preserve operator scopes through the device-auth bypass path, ignore cached under-scoped operator tokens, and show a clear `operator.read` fallback message when a connection really lacks read scope, so operator sessions stop failing or blanking on read-backed pages. (#53110) Thanks @BunsDev.
-- Plugins/uninstall: accept installed `clawhub:` specs and versionless ClawHub package names as uninstall targets, so `openclaw plugins uninstall clawhub:<package>` works again even when the recorded install was pinned to a version.
-- Auth/OpenAI tokens: stop live gateway auth-profile writes from reverting freshly saved credentials back to stale in-memory values, and make `models auth paste-token` write to the resolved agent store, so Configure, Onboard, and token-paste flows stop snapping back to expired OpenAI tokens. Fixes #53207. Related to #45516.
-- Agents/failover: classify generic `api_error` payloads as retryable only when they include transient failure signals, so MiniMax-style backend failures still trigger model fallback without misclassifying billing, auth, or format/context errors. (#49611) Thanks @ayushozha.
-- Diagnostics/cache trace: strip credential fields from cache-trace JSONL output while preserving non-sensitive diagnostic fields and image redaction metadata.
-- Docs/Feishu: replace `botName` with `name` in the channel config examples so the docs match the strict account schema for per-account display names. (#52753) Thanks @haroldfabla2-hue.
-- Doctor/plugins: make `openclaw doctor --fix` remove stale `plugins.allow` and `plugins.entries` refs left behind after plugin removal. Thanks @sallyom
-- Agents/replay: canonicalize malformed assistant transcript content before session-history sanitization so legacy or corrupted assistant turns stop crashing Pi replay and subagent recovery paths.
-- ClawHub/skills: keep updating already-tracked legacy Unicode slugs after the ASCII-only slug hardening, so older installs do not get stuck behind `Invalid skill slug` errors during `openclaw skills update`. (#53206) Thanks @drobison00.
-- Infra/exec trust: preserve shell-multiplexer wrapper binaries for policy checks without breaking approved-command reconstruction, so BusyBox/ToyBox allowlist and audit flows bind to the real wrapper while execution plans stay coherent. (#53134) Thanks @vincentkoc.
-- LINE/runtime-api: pre-export overlapping runtime symbols before the `line-runtime` star export so jiti no longer throws `TypeError: Cannot redefine property` on startup. (#53221) Thanks @Drickon.
-- CLI/cron: make `openclaw cron add|edit --at ... --tz <iana>` honor the requested local wall-clock time for offset-less one-shot datetimes, including DST boundaries, and keep `--tz` rejected for `--every`. (#53224) Thanks @RolfHegr.
-- Commands/auth: stop slash-command authorization from crashing or dropping valid allowlists when channel `allowFrom` resolution hits unresolved SecretRef-backed accounts, and fail closed only for the affected provider inference path. (#52791) Thanks @Lukavyi.
-
 ## 2026.3.23
 
 ### Breaking
 
 ### Changes
 
+- ModelStudio/Qwen: add standard (pay-as-you-go) DashScope endpoints for China and global Qwen API keys alongside the existing Coding Plan endpoints, and relabel the provider group to `Qwen (Alibaba Cloud Model Studio)`. (#43878)
+- UI/clarity: consolidate button primitives (`btn--icon`, `btn--ghost`, `btn--xs`), refine the Knot theme to a black-and-red palette with WCAG 2.1 AA contrast, add config icons for Diagnostics/CLI/Secrets/ACP/MCP sections, replace the roundness slider with discrete stops, and improve accessibility with aria-labels across usage filters. (#53272) Thanks @BunsDev.
+- CSP/Control UI: compute SHA-256 hashes for inline `<script>` blocks in the served `index.html` and include them in the `script-src` CSP directive, keeping inline scripts blocked by default while allowing explicitly hashed bootstrap code. (#53307) Thanks @BunsDev.
+
 ### Fixes
 
+- Plugins/bundled runtimes: ship bundled plugin runtime sidecars like WhatsApp `light-runtime-api.js`, Matrix `runtime-api.js`, and other plugin runtime entry files in the npm package again, so global installs stop failing on missing bundled plugin runtime surfaces.
+- CLI/channel auth: auto-select the single configured login-capable channel for `channels login`/`logout`, harden channel ids against prototype-chain and control-character abuse, and fall back cleanly to catalog-backed channel installs, so channel auth works again for single-channel setups and on-demand channel installs. (#53254) Thanks @BunsDev.
+- Auth/OpenAI tokens: stop live gateway auth-profile writes from reverting freshly saved credentials back to stale in-memory values, and make `models auth paste-token` write to the resolved agent store, so Configure, Onboard, and token-paste flows stop snapping back to expired OpenAI tokens. Fixes #53207. Related to #45516.
+- Control UI/auth: preserve operator scopes through the device-auth bypass path, ignore cached under-scoped operator tokens, and show a clear `operator.read` fallback message when a connection really lacks read scope, so operator sessions stop failing or blanking on read-backed pages. (#53110) Thanks @BunsDev.
+- Plugins/ClawHub: resolve plugin API compatibility against the active runtime version at install time, and add regression coverage for current `>=2026.3.22` ClawHub package checks so installs no longer fail behind the stale `1.2.0` constant. (#53157) Thanks @futhgar.
+- Plugins/uninstall: accept installed `clawhub:` specs and versionless ClawHub package names as uninstall targets, so `openclaw plugins uninstall clawhub:<package>` works again even when the recorded install was pinned to a version.
 - Browser/Chrome MCP: wait for existing-session browser tabs to become usable after attach instead of treating the initial Chrome MCP handshake as ready, which reduces user-profile timeouts and repeated consent churn on macOS Chrome attach flows. Fixes #52930. Thanks @vincentkoc.
 - Browser/CDP: reuse an already-running loopback browser after a short initial reachability miss instead of immediately falling back to relaunch detection, which fixes second-run browser start/open regressions on slower headless Linux setups. Fixes #53004. Thanks @vincentkoc.
+- Agents/web_search: use the active runtime `web_search` provider instead of stale/default selection, so agent turns keep hitting the provider you actually configured. Fixes #53020. Thanks @jzakirov.
+- Mistral/models: lower bundled Mistral max-token defaults to safe output budgets and teach `openclaw doctor --fix` to repair old persisted Mistral provider configs that still carry context-sized output limits, avoiding deterministic Mistral 422 rejects on fresh and existing setups. Fixes #52599. Thanks @vincentkoc.
 - ClawHub/macOS auth: honor macOS auth config and XDG auth paths for saved ClawHub credentials, so `openclaw skills ...` and gateway skill browsing keep using the signed-in auth state instead of silently falling back to unauthenticated mode. Fixes #53034.
 - ClawHub/macOS: read the local ClawHub login from the macOS Application Support path and still honor XDG config on macOS, so skill browsing uses the logged-in token on both default and XDG-style setups. Fixes #52949. Thanks @scoootscooob.
 - ClawHub/skills: resolve the local ClawHub auth token for gateway skill browsing and switch browse-all requests to search so ClawControl stops falling into unauthenticated 429s and empty authenticated skill lists. Fixes #52949. Thanks @vincentkoc.
+- Config/warnings: suppress the confusing “newer OpenClaw” warning when a config written by a same-base correction release like `2026.3.23-2` is read by `2026.3.23`, while still warning for truly newer or incompatible versions.
+- CLI/cron: make `openclaw cron add|edit --at ... --tz <iana>` honor the requested local wall-clock time for offset-less one-shot datetimes, including DST boundaries, and keep `--tz` rejected for `--every`. (#53224) Thanks @RolfHegr.
+- Commands/auth: stop slash-command authorization from crashing or dropping valid allowlists when channel `allowFrom` resolution hits unresolved SecretRef-backed accounts, and fail closed only for the affected provider inference path. (#52791) Thanks @Lukavyi.
+- Agents/failover: classify generic `api_error` payloads as retryable only when they include transient failure signals, so MiniMax-style backend failures still trigger model fallback without misclassifying billing, auth, or format/context errors. (#49611) Thanks @ayushozha.
+- LINE/runtime-api: pre-export overlapping runtime symbols before the `line-runtime` star export so jiti no longer throws `TypeError: Cannot redefine property` on startup. (#53221) Thanks @Drickon.
+- Telegram/threading: populate `currentThreadTs` in the threading tool-context fallback for Telegram DM topics so thread-aware tools still receive the active topic context when the main thread metadata is missing. (#52217)
+- Diagnostics/cache trace: strip credential fields from cache-trace JSONL output while preserving non-sensitive diagnostic fields and image redaction metadata.
+- Docs/Feishu: replace `botName` with `name` in the channel config examples so the docs match the strict account schema for per-account display names. (#52753) Thanks @haroldfabla2-hue.
+- Doctor/plugins: make `openclaw doctor --fix` remove stale `plugins.allow` and `plugins.entries` refs left behind after plugin removal. Thanks @sallyom
+- Agents/replay: canonicalize malformed assistant transcript content before session-history sanitization so legacy or corrupted assistant turns stop crashing Pi replay and subagent recovery paths.
+- ClawHub/skills: keep updating already-tracked legacy Unicode slugs after the ASCII-only slug hardening, so older installs do not get stuck behind `Invalid skill slug` errors during `openclaw skills update`. (#53206) Thanks @drobison00.
+- Infra/exec trust: preserve shell-multiplexer wrapper binaries for policy checks without breaking approved-command reconstruction, so BusyBox/ToyBox allowlist and audit flows bind to the real wrapper while execution plans stay coherent. (#53134) Thanks @vincentkoc.
 - Plugins/message tool: make Discord `components` and Slack `blocks` optional again, and route Feishu `message(..., media=...)` sends through the outbound media path, so pin/unpin/react flows stop failing schema validation and Feishu file/image attachments actually send. Fixes #52970 and #52962. Thanks @vincentkoc.
 - Gateway/model pricing: stop `openrouter/auto` pricing refresh from recursing indefinitely during bootstrap, so OpenRouter auto routes can populate cached pricing and `usage.cost` again. Fixes #53035. Thanks @vincentkoc.
-- Mistral/models: lower bundled Mistral max-token defaults to safe output budgets and teach `openclaw doctor --fix` to repair old persisted Mistral provider configs that still carry context-sized output limits, avoiding deterministic Mistral 422 rejects on fresh and existing setups. Fixes #52599. Thanks @vincentkoc.
-- Agents/web_search: use the active runtime `web_search` provider instead of stale/default selection, so agent turns keep hitting the provider you actually configured. Fixes #53020. Thanks @jzakirov.
 - Models/OpenAI Codex OAuth: bootstrap the env-configured HTTP/HTTPS proxy dispatcher on the stored-credential refresh path before token renewal runs, so expired Codex OAuth profiles can refresh successfully in proxy-required environments instead of locking users out after the first token expiry.
 - Models/OpenAI Codex OAuth and Plugins/MiniMax OAuth: ensure env-configured HTTP/HTTPS proxy dispatchers are initialized before OAuth preflight and token exchange requests so proxy-required environments can complete MiniMax and OpenAI Codex sign-in flows again. (#52228; fixes #51619, #51569) Thanks @openperf.
 - Plugins/memory-lancedb: bootstrap LanceDB into plugin runtime state on first use when the bundled npm install does not already have it, so `plugins.slots.memory="memory-lancedb"` works again after global npm installs without moving LanceDB into OpenClaw core dependencies. Fixes #26100.

@@ -2620,7 +2620,7 @@
           "exportName": "resolveCommandAuthorization",
           "kind": "function",
           "source": {
-            "line": 303,
+            "line": 440,
             "path": "src/auto-reply/command-auth.ts"
           }
         },

@@ -287,7 +287,7 @@
 {"declaration":"export function parseCommandArgs(command: ChatCommandDefinition, raw?: string | undefined): CommandArgs | undefined;","entrypoint":"command-auth","exportName":"parseCommandArgs","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":254,"sourcePath":"src/auto-reply/commands-registry.ts"}
 {"declaration":"export function resolveCommandArgChoices(params: { command: ChatCommandDefinition; arg: CommandArgDefinition; cfg?: OpenClawConfig | undefined; provider?: string | undefined; model?: string | undefined; }): ResolvedCommandArgChoice[];","entrypoint":"command-auth","exportName":"resolveCommandArgChoices","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":316,"sourcePath":"src/auto-reply/commands-registry.ts"}
 {"declaration":"export function resolveCommandArgMenu(params: { command: ChatCommandDefinition; args?: CommandArgs | undefined; cfg?: OpenClawConfig | undefined; }): { arg: CommandArgDefinition; choices: ResolvedCommandArgChoice[]; title?: string | undefined; } | null;","entrypoint":"command-auth","exportName":"resolveCommandArgMenu","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":346,"sourcePath":"src/auto-reply/commands-registry.ts"}
-{"declaration":"export function resolveCommandAuthorization(params: { ctx: MsgContext; cfg: OpenClawConfig; commandAuthorized: boolean; }): CommandAuthorization;","entrypoint":"command-auth","exportName":"resolveCommandAuthorization","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":303,"sourcePath":"src/auto-reply/command-auth.ts"}
+{"declaration":"export function resolveCommandAuthorization(params: { ctx: MsgContext; cfg: OpenClawConfig; commandAuthorized: boolean; }): CommandAuthorization;","entrypoint":"command-auth","exportName":"resolveCommandAuthorization","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":440,"sourcePath":"src/auto-reply/command-auth.ts"}
 {"declaration":"export function resolveCommandAuthorizedFromAuthorizers(params: { useAccessGroups: boolean; authorizers: CommandAuthorizer[]; modeWhenAccessGroupsOff?: CommandGatingModeWhenAccessGroupsOff | undefined; }): boolean;","entrypoint":"command-auth","exportName":"resolveCommandAuthorizedFromAuthorizers","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":8,"sourcePath":"src/channels/command-gating.ts"}
 {"declaration":"export function resolveControlCommandGate(params: { useAccessGroups: boolean; authorizers: CommandAuthorizer[]; allowTextCommands: boolean; hasControlCommand: boolean; modeWhenAccessGroupsOff?: CommandGatingModeWhenAccessGroupsOff | undefined; }): { ...; };","entrypoint":"command-auth","exportName":"resolveControlCommandGate","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":31,"sourcePath":"src/channels/command-gating.ts"}
 {"declaration":"export function resolveDirectDmAuthorizationOutcome(params: { isGroup: boolean; dmPolicy: string; senderAllowedForCommands: boolean; }): \"disabled\" | \"unauthorized\" | \"allowed\";","entrypoint":"command-auth","exportName":"resolveDirectDmAuthorizationOutcome","importSpecifier":"openclaw/plugin-sdk/command-auth","kind":"function","recordType":"export","sourceLine":120,"sourcePath":"src/plugin-sdk/command-auth.ts"}

@@ -237,6 +237,10 @@ class MainViewModel(app: Application) : AndroidViewModel(app) {
     ensureRuntime().handleCanvasA2UIActionFromWebView(payloadJson)
   }
 
+  fun isTrustedCanvasActionUrl(rawUrl: String?): Boolean {
+    return ensureRuntime().isTrustedCanvasActionUrl(rawUrl)
+  }
+
   fun requestCanvasRehydrate(source: String = "screen_tab") {
     ensureRuntime().requestCanvasRehydrate(source = source, force = true)
   }

@@ -13,6 +13,13 @@ class A2UIHandler(
   private val getNodeCanvasHostUrl: () -> String?,
   private val getOperatorCanvasHostUrl: () -> String?,
 ) {
+  fun isTrustedCanvasActionUrl(rawUrl: String?): Boolean {
+    return CanvasActionTrust.isTrustedCanvasActionUrl(
+      rawUrl = rawUrl,
+      trustedA2uiUrls = listOfNotNull(resolveA2uiHostUrl()),
+    )
+  }
+
   fun resolveA2uiHostUrl(): String? {
     val nodeRaw = getNodeCanvasHostUrl()?.trim().orEmpty()
     val operatorRaw = getOperatorCanvasHostUrl()?.trim().orEmpty()

@@ -0,0 +1,50 @@
+package ai.openclaw.app.node
+
+import java.net.URI
+
+object CanvasActionTrust {
+  const val scaffoldAssetUrl: String = "file:///android_asset/CanvasScaffold/scaffold.html"
+
+  fun isTrustedCanvasActionUrl(rawUrl: String?, trustedA2uiUrls: List<String>): Boolean {
+    val candidate = rawUrl?.trim().orEmpty()
+    if (candidate.isEmpty()) return false
+    if (candidate == scaffoldAssetUrl) return true
+
+    val candidateUri = parseUri(candidate) ?: return false
+    if (candidateUri.scheme.equals("file", ignoreCase = true)) {
+      return false
+    }
+
+    return trustedA2uiUrls.any { trusted ->
+      isTrustedA2uiPage(candidateUri, trusted)
+    }
+  }
+
+  private fun isTrustedA2uiPage(candidateUri: URI, trustedUrl: String): Boolean {
+    val trustedUri = parseUri(trustedUrl) ?: return false
+    if (!candidateUri.scheme.equals(trustedUri.scheme, ignoreCase = true)) return false
+    if (candidateUri.host?.equals(trustedUri.host, ignoreCase = true) != true) return false
+    if (effectivePort(candidateUri) != effectivePort(trustedUri)) return false
+
+    val trustedPath = trustedUri.rawPath?.takeIf { it.isNotBlank() } ?: return false
+    val candidatePath = candidateUri.rawPath?.takeIf { it.isNotBlank() } ?: return false
+    val trustedPrefix = if (trustedPath.endsWith("/")) trustedPath else "$trustedPath/"
+    return candidatePath == trustedPath || candidatePath.startsWith(trustedPrefix)
+  }
+
+  private fun effectivePort(uri: URI): Int {
+    if (uri.port >= 0) return uri.port
+    return when (uri.scheme?.lowercase()) {
+      "https" -> 443
+      "http" -> 80
+      else -> -1
+    }
+  }
+
+  private fun parseUri(raw: String): URI? =
+    try {
+      URI(raw)
+    } catch (_: Throwable) {
+      null
+    }
+}

@@ -904,6 +904,10 @@ class NodeRuntime(
     }
   }
 
+  fun isTrustedCanvasActionUrl(rawUrl: String?): Boolean {
+    return a2uiHandler.isTrustedCanvasActionUrl(rawUrl)
+  }
+
   fun loadChat(sessionKey: String) {
     val key = sessionKey.trim().ifEmpty { resolveMainSessionKey() }
     chat.load(key)

@@ -22,13 +22,15 @@ import androidx.compose.ui.viewinterop.AndroidView
 import androidx.webkit.WebSettingsCompat
 import androidx.webkit.WebViewFeature
 import ai.openclaw.app.MainViewModel
+import java.util.concurrent.atomic.AtomicReference
 
 @SuppressLint("SetJavaScriptEnabled")
 @Composable
 fun CanvasScreen(viewModel: MainViewModel, visible: Boolean, modifier: Modifier = Modifier) {
   val context = LocalContext.current
   val isDebuggable = (context.applicationInfo.flags and android.content.pm.ApplicationInfo.FLAG_DEBUGGABLE) != 0
   val webViewRef = remember { mutableStateOf<WebView?>(null) }
+  val currentPageUrlRef = remember { AtomicReference<String?>(null) }
 
   DisposableEffect(viewModel) {
     onDispose {
@@ -68,6 +70,14 @@ fun CanvasScreen(viewModel: MainViewModel, visible: Boolean, modifier: Modifier
         isHorizontalScrollBarEnabled = true
         webViewClient =
           object : WebViewClient() {
+            override fun onPageStarted(
+              view: WebView,
+              url: String?,
+              favicon: android.graphics.Bitmap?,
+            ) {
+              currentPageUrlRef.set(url)
+            }
+
             override fun onReceivedError(
               view: WebView,
               request: WebResourceRequest,
@@ -90,6 +100,7 @@ fun CanvasScreen(viewModel: MainViewModel, visible: Boolean, modifier: Modifier
             }
 
             override fun onPageFinished(view: WebView, url: String?) {
+              currentPageUrlRef.set(url)
               if (isDebuggable) {
                 Log.d("OpenClawWebView", "onPageFinished: $url")
               }
@@ -122,7 +133,12 @@ fun CanvasScreen(viewModel: MainViewModel, visible: Boolean, modifier: Modifier
             }
           }
 
-        val bridge = CanvasA2UIActionBridge { payload -> viewModel.handleCanvasA2UIActionFromWebView(payload) }
+        val bridge =
+          CanvasA2UIActionBridge(
+            isTrustedPage = { viewModel.isTrustedCanvasActionUrl(currentPageUrlRef.get()) },
+          ) { payload ->
+            viewModel.handleCanvasA2UIActionFromWebView(payload)
+          }
         addJavascriptInterface(bridge, CanvasA2UIActionBridge.interfaceName)
         viewModel.canvas.attach(this)
         webViewRef.value = this
@@ -147,11 +163,15 @@ private fun disableForceDarkIfSupported(settings: WebSettings) {
   WebSettingsCompat.setForceDark(settings, WebSettingsCompat.FORCE_DARK_OFF)
 }
 
-private class CanvasA2UIActionBridge(private val onMessage: (String) -> Unit) {
+private class CanvasA2UIActionBridge(
+  private val isTrustedPage: () -> Boolean,
+  private val onMessage: (String) -> Unit,
+) {
   @JavascriptInterface
   fun postMessage(payload: String?) {
     val msg = payload?.trim().orEmpty()
     if (msg.isEmpty()) return
+    if (!isTrustedPage()) return
     onMessage(msg)
   }
 

@@ -0,0 +1,42 @@
+package ai.openclaw.app.node
+
+import org.junit.Assert.assertFalse
+import org.junit.Assert.assertTrue
+import org.junit.Test
+
+class CanvasActionTrustTest {
+  @Test
+  fun acceptsBundledScaffoldAsset() {
+    assertTrue(CanvasActionTrust.isTrustedCanvasActionUrl(CanvasActionTrust.scaffoldAssetUrl, emptyList()))
+  }
+
+  @Test
+  fun acceptsTrustedA2uiPageOnAdvertisedCanvasHost() {
+    assertTrue(
+      CanvasActionTrust.isTrustedCanvasActionUrl(
+        rawUrl = "https://canvas.example.com:9443/__openclaw__/cap/token/__openclaw__/a2ui/?platform=android",
+        trustedA2uiUrls = listOf("https://canvas.example.com:9443/__openclaw__/cap/token/__openclaw__/a2ui/?platform=android"),
+      ),
+    )
+  }
+
+  @Test
+  fun rejectsDifferentOriginEvenIfPathMatches() {
+    assertFalse(
+      CanvasActionTrust.isTrustedCanvasActionUrl(
+        rawUrl = "https://evil.example.com:9443/__openclaw__/cap/token/__openclaw__/a2ui/?platform=android",
+        trustedA2uiUrls = listOf("https://canvas.example.com:9443/__openclaw__/cap/token/__openclaw__/a2ui/?platform=android"),
+      ),
+    )
+  }
+
+  @Test
+  fun rejectsUntrustedCanvasPagePathOnTrustedOrigin() {
+    assertFalse(
+      CanvasActionTrust.isTrustedCanvasActionUrl(
+        rawUrl = "https://canvas.example.com:9443/untrusted/index.html",
+        trustedA2uiUrls = listOf("https://canvas.example.com:9443/__openclaw__/cap/token/__openclaw__/a2ui/?platform=android"),
+      ),
+    )
+  }
+}

@@ -118,6 +118,7 @@ Docs: https://docs.openclaw.ai
 - CLI/config: make `config set --strict-json` enforce real JSON, prefer `JSON.parse` with JSON5 fallback for machine-written cron/subagent stores, and relabel raw config surfaces as `JSON/JSON5` to match actual compatibility. Related: #48415, #43127, #14529, #21332. Thanks @adhitShet and @vincentkoc.
 - CLI/Ollama onboarding: keep the interactive model picker for explicit `openclaw onboard --auth-choice ollama` runs so setup still selects a default model without reintroducing pre-picker auto-pulls. (#49249) Thanks @BruceMacD.
 - CLI/configure: clarify fresh-setup memory-search warnings so they say semantic recall needs at least one embedding provider, and scope the initial model allowlist picker to the provider selected in configure. Thanks @vincentkoc.
+- Android/canvas: ignore bridge messages from pages outside the bundled scaffold and trusted A2UI surfaces. Thanks @vincentkoc.
 - CLI/status: keep `status --json` stdout clean by skipping plugin compatibility scans that were not rendered in the JSON payload. (#52449) Thanks @cgdusek.
 - Browser/node proxy: enforce `nodeHost.browserProxy.allowProfiles` across `query.profile` and `body.profile`, block proxy-side profile create/delete when the allowlist is set, and keep the default full proxy surface when the allowlist is empty.
 - Web tools/Exa: align the bundled Exa plugin with the current Exa API by supporting newer search types and richer `contents` options, while fixing the result-count cap to honor Exa's higher limit. Thanks @vincentkoc.