OpenClaw v2026.4.2 Release: Task Flow Returns, Android Assistant, and Plugin Hooks

OpenClaw v2026.4.2 dropped on April 2, 2026, bringing back one of the most requested features — Task Flow — along with Android integration, new plugin capabilities, and tighter security controls.

Task Flow Is Back

Task Flow was temporarily removed in an earlier version due to stability concerns. v2026.4.2 restores it with a cleaner architecture and two distinct sync modes:

• Managed mode — OpenClaw owns the task state. Tasks are created, tracked, and completed entirely within the framework. Best for self-contained agent workflows.
• Mirrored mode — Task state syncs with an external system (Jira, Linear, Notion, etc.). Changes in either direction propagate automatically. Best for teams that already have a project management tool.

Both modes support durable state tracking, meaning tasks survive gateway restarts and agent crashes. If an agent goes down mid-task, it picks up exactly where it left off when it comes back online.

Android Assistant Entrypoints

OpenClaw agents can now be triggered through Google Assistant on Android devices. The release adds App Actions metadata that maps voice commands to agent invocations.

Say "Hey Google, ask my research agent to summarize today's news" and it routes to your OpenClaw agent. The response comes back through the Assistant interface. This makes OpenClaw agents accessible on every Android phone without installing a separate app.

Combined with the macOS Voice Wake from v2026.4.1, OpenClaw now has native voice entry points on both major desktop and mobile platforms.

before_agent_reply Plugin Hook

Plugin developers get a powerful new hook: before_agent_reply. It fires after an inline action completes but before the agent sends its response to the user.

Use cases include:

• Injecting synthetic replies (e.g., adding a disclaimer or translation)
• Logging agent responses to external analytics systems
• Modifying responses based on user permissions or context
• Triggering side effects like notifications or webhook calls

This hook complements the existing after_agent_reply and gives plugin authors full control over the response lifecycle.

Enhanced Feishu Integration

The Feishu (Lark) integration gets a dedicated Drive comment-event flow. When someone comments on a document in Feishu Drive, your OpenClaw agent can automatically respond, summarize the thread, or take action based on the comment content.

This is particularly valuable for teams in Asia-Pacific markets where Feishu is the primary collaboration tool. Agents can now participate in document reviews alongside human team members.

Centralized Provider Transport Policy

Managing multiple LLM providers used to mean configuring transport settings (timeouts, retries, proxy rules) for each one separately. v2026.4.2 introduces a centralized transport policy that applies across all providers.

Set your retry strategy, timeout limits, and proxy configuration once. Every provider — OpenAI, Anthropic, Google, Ollama, or any custom endpoint — inherits the same policy. You can still override per provider when needed.

Improved Exec Approval Security

Exec approvals now support granular scoping. Instead of approving all shell commands or none, you can define patterns:

• Allow git * but block rm -rf *
• Allow npm run * but block npm publish
• Allow read-only commands but block anything that writes

This is a meaningful security improvement for production deployments where agents need some system access but should not have unlimited shell permissions.

How to Update

Pull the latest Docker image or update your installation. If you were using Task Flow before it was removed, your existing task configurations should migrate automatically. Check the full changelog on GitHub for migration notes.