OpenClaw v2026.5.16-beta.5: Plugin SDK, New Skills, and a Security Fix

OpenClaw shipped v2026.5.16-beta.5 on Sunday evening, May 17, and it is one of the more feature-dense beta drops the project has had in weeks. The headline is a proper typed Plugin SDK — but there are also three new bundled skills, a Slack integration upgrade, and a QA-Lab security fix worth knowing about.

A First-Class Plugin SDK

The biggest structural change in beta.5 is defineToolPlugin, the new entry point for building simple typed tool plugins for OpenClaw. Along with it come three new CLI subcommands:

• openclaw plugins build — compiles your plugin to a distributable artifact
• openclaw plugins validate — checks your plugin's manifest and tool contracts
• openclaw plugins init — scaffolds a new plugin with generated manifest metadata, optional tool declarations, and context factories

Before this, writing a plugin meant navigating the full extension API without guardrails. defineToolPlugin gives you typed schemas, generated manifests, and context factories out of the box — a much lower floor for community authors. The SDK also now bundles openclaw/plugin-sdk/zod in the published package artifact, fixing a long-standing install issue where pnpm global install failed due to a missing zod symlink (#78398).

Three New Built-in Skills

Meme Maker is the most visible addition. It supports curated template search, local SVG/PNG rendering, Imgflip hosted rendering, and Know Your Meme provenance links. Useful both for actual meme creation and as a reference example of what a media-generation skill looks like.

Python Debugger adds pdb, breakpoint(), post-mortem inspection, and debugpy remote attach directly to your agent's toolkit. If you use OpenClaw for developer-facing workflows, this one is worth installing.

Node Inspector rounds out the debugging trio with Node.js process inspection and fused diagram generation. There is also a throwaway spike workflow skill for quick one-off explorations.

Slack Gets Assistant Thread Lifecycle Support

Slack's assistant thread API lets apps participate in the native Slack AI experience — dedicated thread contexts, suggested prompts, and thread-scoped state. Beta.5 wires OpenClaw into this properly: you can now configure your bot to receive assistant_thread_started and assistant_thread_context_changed events, and respond inside those threads with the full OpenClaw agent runtime. This is a meaningful upgrade for teams running OpenClaw as a Slack bot.

QA-Lab Security Fix

PR #66355 patches two related issues in the Docker QA harness:

1. Bearer tokens were returned in unauthenticated bootstrap payloads. The bootstrap response used to include Control UI tokens without requiring authentication first. Those tokens are now withheld from unauthenticated callers.
2. Docker harness ports were not bound to loopback. Harness HTTP listeners now bind to 127.0.0.1 only, so they are not reachable from other containers or the host network by default.

Neither issue affects standard Gateway deployments — this is isolated to the openclaw qa Docker test harness. But if you run QA infrastructure in shared CI environments, updating is the right call.

Mac App Settings Redesign

The macOS app received a significant settings UI overhaul: consistent card layouts, cached navigation, and cleaner panes for permissions, voice, skills, cron, exec, and debug. It also fixes a SwiftUI metadata crash that could occur when rendering the Cron Jobs pane. The Settings sidebar toggle moves into the native titlebar, and visited panes stay mounted so switching tabs no longer triggers a full reload.

Other Notable Fixes

• Group chat ambient turns: messages.groupChat.unmentionedInbound: "room_event" is now a supported opt-in mode, letting you run always-on room context without the agent responding visibly to every message.
• GPT-5 brevity cap removed: OpenClaw was post-processing GPT-5 replies with a hardcoded brevity cap that appended ellipses to long responses. That is gone (#82910).
• Gemini 3 thought signatures preserved: Tool-call function calls on Gemini 3 were failing with missing thought_signature 400 errors after context replay. Fixed (#80358).
• Memory-core startup scan: The memory index now incrementally syncs on startup rather than rescanning all sessions, improving cold-start time on large workspaces.

How to Update

openclaw update

The full changelog is on the v2026.5.16-beta.5 release page. As with all beta releases, production deployments may want to wait for the stable tag.