🤖 Ghostwritten by Claude Opus 4.6 · Fact-checked & edited by GPT 5.4 · Curated by Tom Hundley
OpenClaw v2026.3.22 introduces four headline changes: ClawHub, a built-in plugin registry; bundle discovery, which installs curated plugin packs for tools like Codex, Claude, and Cursor; a new /btw slash command for adding follow-up context; and Anthropic Vertex support for teams using Claude through Google Cloud. The practical benefit is simple: plugin discovery and setup should be easier, more consistent, and less dependent on manually searching npm.
If you already use OpenClaw, this release appears designed to reduce setup friction. If you're new to it, ClawHub is the feature most likely to matter first because it moves plugin search and installation into the product itself. Below, we walk through what changed, how to use it, and where to be careful—especially with community plugins.
TL;DR: ClawHub is OpenClaw's built-in plugin registry, intended to replace the old process of manually finding plugins through npm or web searches.
Before v2026.3.22, installing a plugin often meant finding a package name elsewhere, installing it manually, and then checking whether it actually worked with your OpenClaw version. ClawHub brings that workflow into OpenClaw itself.
Instead of searching npm or forum threads, you browse plugins from inside the product. That should make discovery faster and reduce guesswork around compatibility and maintenance.
| Feature | Old Way (npm Hunting) | New Way (ClawHub) |
|---|---|---|
| Discovery | Web search and package lookup | Built-in search and browse |
| Compatibility | Manual checking | Registry can surface version-relevant results |
| Installation | Terminal commands | In-product command flow |
| Security signals | Package page and author reputation | Registry metadata such as ratings or review signals |
| Configuration | Often manual | More guided setup for supported plugins |
| Updates | Manual checks | In-product update visibility |
According to the article's source material, OpenClaw's changelog says plugin operations now route through a unified openclaw plugins command. If accurate, that gives users one consistent entry point for install, update, removal, and configuration.
From the OpenClaw interface, type:
/plugins search [what you need]Examples:
/plugins search slack/plugins search databaseTo install a plugin:
/plugins install [plugin-name]That keeps the workflow inside OpenClaw rather than sending users out to npm first.
TL;DR: Bundles group related plugins together so you can install a ready-made setup for a specific tool or workflow.
Bundles are curated collections of plugins meant to work together. In this release, the article claims OpenClaw includes bundle discovery for:
The value here is speed. Instead of installing and wiring up each plugin one by one, you install a bundle and let OpenClaw handle more of the setup.
The release notes described in the draft suggest that bundle installs automatically map plugins into OpenClaw's AgentSkills system. If you've already set up workspace files, that means less manual editing after installation.
Previously, adding a capability could require:
With bundles, the intended flow is closer to:
Use:
/plugins bundle listThen install one:
/plugins bundle install cursorAfter that, check your skills list with:
/skillsThe original draft cited JetBrains' 2025 Developer Ecosystem Survey to support the broader point that AI coding tools are mainstream. That directional claim is plausible, but the specific "over 70%" figure was removed here because it was not verifiable from the provided source text alone.
For teams standardizing AI-assisted workflows, bundles could make OpenClaw a more practical coordination layer across tools.
TL;DR: /btw appears to let you add follow-up context to your last instruction without rewriting the whole prompt.
This is a small feature, but potentially a useful one. The idea is to append clarifying context after you've already sent a message.
Example:
Summarize today's Slack messages from the design channelThen refine it with:
/btw only messages after 2pmIf OpenClaw handles this as described, it should reduce prompt rewrites and make iterative workflows feel more natural.
That matters most for users who work conversationally: send an instruction, inspect the result, then tighten the scope. /btw fits that pattern well.
TL;DR: OpenClaw now appears to support Anthropic models through Google Cloud Vertex AI, which may help teams that centralize billing and access controls in GCP.
If your organization uses Claude through Vertex AI rather than directly through Anthropic, this release adds a native path for that setup.
Verify your OpenClaw version — type /version and confirm you see v2026.3.22 or later. If not, upgrade following the security fix guide first.
Open model provider settings — navigate to Settings → Model Providers, or use /settings providers if supported in your environment.
Add Anthropic Vertex as a provider — select it from the available provider list.
Enter your configuration:
provider: anthropic-vertex
project_id: "your-gcp-project-id"
region: "us-central1"
model: "claude-sonnet-4"Authenticate — complete the Google Cloud authentication flow presented by OpenClaw.
Test the connection — send a simple prompt and confirm the model responds.
A note on the model name: the original draft used a dated model string. Because model naming and availability can change, this version uses a more general placeholder. Check current Vertex AI documentation and the OpenClaw release notes before publishing a specific model identifier.
If the provider consolidation described in the draft is accurate, this release also simplifies model management by putting multiple providers in one place.
TL;DR: ClawHub should improve plugin discovery, but you still need to review permissions, trust signals, and update practices before installing community plugins.
A registry is safer than random package hunting only if users still apply basic judgment. Community plugins can request broad permissions, and any plugin with access to files, tokens, or network calls deserves scrutiny.
Use this checklist before installing a community plugin:
The draft also argues that routing plugin operations through a centralized openclaw plugins command creates a cleaner security chokepoint. Architecturally, that is a reasonable claim: centralization can make auditing and policy enforcement easier, assuming the implementation is sound.
Not necessarily. If OpenClaw preserves backward compatibility, older plugins may continue to work while you transition. The practical advantage of moving to ClawHub-managed plugins is better visibility into compatibility, updates, and installation flow.
Probably yes, if OpenClaw handles overlapping dependencies cleanly. In practice, the important question is how it resolves duplicate plugins, conflicting skill mappings, or version mismatches. Test bundle combinations in a non-critical workspace first.
The draft says OpenClaw treats it like a normal new instruction rather than throwing an error. That behavior is plausible, but you should verify it against the current release before publishing as a guaranteed behavior.
The integration itself is typically a product feature, but model usage through Vertex AI is billed through Google Cloud. Your actual cost depends on the Claude model, region, and your cloud agreement.
No single signal is enough. Check permissions, maintainer identity, update cadence, user feedback, and whether the plugin's requested access matches its purpose. If a simple integration asks for broad file-system or network access, treat that as a warning sign.
/btw supports iterative prompting by adding follow-up context to the previous instructionOpenClaw v2026.3.22 looks like a meaningful usability release. ClawHub is the biggest improvement because it addresses the most frustrating part of plugin adoption: finding, trusting, and installing the right package. Bundles, /btw, and Vertex support all build on that same theme of reducing friction.
If you're already on OpenClaw, start by testing ClawHub search, installing one bundle, and reviewing your plugin permissions model. If you're documenting this release for your team, link them to your internal setup standards alongside the official release notes so they adopt the new workflow safely.
Want help turning fast-moving platform updates into clear internal documentation or production-ready developer guides? Elegant Software Solutions helps teams turn technical change into usable systems and content.
Discover more content: