OpenClaw v2026.4.22 Hardens Android Intents — Auto-Send Blocked for Injected Prompts

External Android apps can no longer auto-dispatch prompts to OpenClaw via ASK_OPENCLAW intents. Actions now only prefill the draft, requiring user confirmation before sending.

OpenClaw v2026.4.22 Hardens Android Intents — Auto-Send Blocked for Injected Prompts

OpenClaw has changed how it handles Android intents so external apps can no longer send prompts into the assistant without a user confirming the action first. In the v2026.4.22 release, the app now uses ASK_OPENCLAW intents to prefill a draft prompt, but it does not auto-dispatch the message on its own.

The update was published on GitHub on April 22, 2026. According to the release notes, the change hardens the Android integration by removing the automatic send behavior that previously allowed an outside app to inject a prompt directly into OpenClaw.

⚡ New to this?

This is a security-related change in an Android app. An intent is a standard Android mechanism that lets one app ask another app to do something, and in this case it was being used to pass prompts into OpenClaw. Non-experts should care because automatic sending can let another app trigger actions without an obvious review step, which is a common trust and safety issue on mobile devices.

🦞 OpenClaw angle

If you use OpenClaw on Android, external apps could previously inject prompts without your knowledge. This closes that vector.

On Android, an intent is a standard way for one app to ask another app to do something, such as open a screen, share data, or start a task. Intents are a normal part of the platform, but they can become a security concern if an app accepts data and acts on it too quickly, especially when the sender is not fully trusted.

That is the risk OpenClaw is addressing here. With the earlier behavior, another Android app could trigger OpenClaw with a pre-filled prompt and have it sent automatically. The new release changes that flow so the prompt is only prepared in the draft field, and the user must review it and tap send themselves.

The distinction matters because assistant apps often sit between user input and external tools, APIs, or connected services. If a prompt arrives from another app and is executed without confirmation, it can create a path for unexpected actions. Even when the prompt itself is harmless, automatic dispatch removes a layer of user control that many security-conscious Android users and administrators expect.

The release note frames the change as a hardening measure. In security terms, hardening means reducing the ways software can be abused, usually by tightening defaults, limiting automatic behavior, or requiring explicit approval for sensitive actions. Here, the main control is simple: OpenClaw will still accept the intent, but it will not treat the incoming text as ready-to-send instruction.

For developers, this also reflects a broader pattern in mobile app design. Features that bridge apps, such as sharing, deep links, or custom intents, can be useful for automation, but they need careful guardrails. A design that works well for convenience can also create a hidden trust boundary if one app can cause another to act on its behalf.

The release is part of OpenClaw’s Android build and is listed as version v2026.4.22 in the project’s GitHub releases. The core behavioral change is narrow but clear: ASK_OPENCLAW intents still open a path into the assistant, but they now stop at draft population instead of sending the prompt automatically.

Source: GitHub ↗

More from OpenClaw Releases