ClawHub: A Registry for OpenClaw Skills With Soft Delete and Vector Search

What ClawHub Solves for OpenClaw Users

ClawHub addresses a specific gap: OpenClaw agents need a discoverable, versioned place to publish text-based skills, which are a SKILL.md file plus supporting files. Without a registry, sharing a skill means passing around a tarball or a git URL, and there is no way to search across the community. ClawHub also hosts native code plugins, bundle plugins, and experimental whole-agent Claw packages. The intended audience is both skill authors who want to publish with changelogs and tags, and agent operators who want to browse, install, pin, and update skills from a CLI. The registry is built for fast browsing and a CLI-friendly API, which suggests the primary consumer is a developer or an automated workflow, not a casual web user.

How the Registry Works Under the Hood

The architecture is clear from the repository layout. The web app is built with TanStack Start, which is React with Vite and Nitro. The backend is Convex, which provides the database, file storage, and HTTP actions. Convex Auth handles GitHub OAuth for login. Search is not keyword-based; it uses OpenAI embeddings with the text-embedding-3-small model and Convex vector search. The shared API schema and routes live in packages/schema, published as clawhub-schema, so the CLI and the web app use the same types. The data flow is straightforward: a skill is published via the CLI or API, stored in Convex, indexed for embeddings, and then searchable. The README mentions moderation hooks and admin approval, which implies a curation layer beyond simple upload.

Getting Started With the CLI

The CLI is the main entry point. You authenticate with clawhub login, or clawhub login --device for remote or headless environments. To discover skills, you run clawhub search or clawhub explore. The unified catalog for skills and plugins is accessed with clawhub package explore and clawhub package inspect <name>. To install a skill without installing it, use clawhub inspect @openclaw/demo, then clawhub install @openclaw/demo. Local install management includes clawhub pin <skill> to freeze a version, clawhub unpin, clawhub uninstall, clawhub list, and clawhub update --all. Publishing a skill is clawhub skill publish <path>. For plugins, you use clawhub package publish <source>, and the manifest must include openclaw.compat.pluginApi and openclaw.build.openclawVersion. The README points to docs/cli.md for a minimal example. Local development requires Bun, and you run bunx convex dev for the backend and bun run dev for the web app.

Soft Delete and Restore: The Core Design Choice

The most interesting mechanism is soft delete. When you remove an uploaded registry skill, clawhub delete <skill> marks it as deleted but does not remove it from the database. clawhub undelete <skill> restores it. The same applies to packages with clawhub package delete and clawhub package undelete. This is a deliberate choice to avoid breaking old links and installs. The README states that soft-delete/restore is allowed for the skill or package owner, publisher owner/admin, moderators, and admins. Hard delete is admin-only, used for ban flows and management tools. This is a trade-off: it protects against accidental removal and preserves history, but it means a regular user cannot permanently remove a skill they uploaded. If you publish something you later regret, you can only hide it, not erase it from the registry, unless an admin steps in.

Renaming and Merging Without Breaking Links

ClawHub handles the problem of skill identity with rename and merge operations. clawhub skill rename <skill> <new-name> keeps the old slug as a redirect alias, so existing links and installs continue to work. clawhub skill merge <source> <target> hides the source listing and redirects the old slug to the canonical target. This is useful when a skill evolves into a different name or when duplicates appear. The same pattern appears for packages. This approach is better than a simple rename that breaks URLs, but it also means the registry accumulates redirect aliases over time. Operators should be aware that the database will contain stale slugs, though they are hidden from listings.

Nix Plugin Support and Configuration Requirements

ClawHub can store a nix-clawdbot plugin pointer in SKILL.md frontmatter. This is not a regular skill pack; it bundles the skill pack, the CLI binary, and its config flags and requirements. The frontmatter includes a metadata block with a clawdbot key, a nix object that has a plugin source like github:clawdbot/nix-steipete-tools?dir=tools/peekaboo, and a systems list such as aarch64-darwin. The same metadata can declare required environment variables, state directories, and an example config snippet. The README recommends including cli --help output for these nix plugins. This feature makes ClawHub more than a skill store; it becomes a distribution mechanism for full agent tooling. However, it is specific to nix-clawdbot users, so it adds complexity for anyone not on that stack.

Telemetry and the Cost of Install Counts

ClawHub tracks install telemetry when you run clawhub install while logged in. The purpose is to compute install counts, which are likely shown on skill pages as a popularity metric. You can disable this with the environment variable CLAWHUB_DISABLE_TELEMETRY=1. This is a privacy consideration that the project addresses explicitly. The trade-off is that install counts will be inaccurate for users who disable telemetry, but that is an acceptable cost for user trust. The README points to docs/telemetry.md for details, but the key fact is that telemetry is opt-out, not opt-in. If you are privacy-sensitive, you need to set the flag in your environment before running any install command.

Limitations and the Case Against ClawHub

The most obvious limitation is the reliance on Convex and OpenAI embeddings. Running your own instance requires a Convex deployment, an OpenAI API key, and the associated costs. The README lists OPENAI_API_KEY as an environment variable, so search and indexing will not work without it. This is a real operational burden. Another limitation is that hard delete is admin-only, which may not fit a moderation model where users need to retract content permanently. Also, the registry is tightly coupled to OpenClaw; it is not a general-purpose skill registry. If you are not using OpenClaw, this project is irrelevant. The CLI is the primary interface, and the web app is secondary, so users who prefer a GUI may find the experience thin.

Alternatives and How They Differ

A direct alternative is a plain git repository or a package manager like npm for skills. A git repo gives you versioning, branching, and hard deletion via history rewrite, but it lacks search, moderation, and a unified catalog. ClawHub adds a vector search index and a curated approval flow, which a git repo does not provide. Another alternative is a custom registry built on a general-purpose package manager like npm or PyPI, but those are not designed for SKILL.md files and do not have the soft-delete/restore semantics that ClawHub offers. The key difference is that ClawHub is purpose-built for OpenClaw, so it understands skill metadata, nix plugin pointers, and the concept of pinning local installs. If you need those features, a generic registry will require significant glue code.

Maintenance and License Considerations

The project is MIT licensed, which means you can use, modify, and distribute it freely, including for commercial purposes, with attribution. The repository has recent releases, with v0.23.3 pushed on 2026-08-04, so it is actively maintained. The docs directory includes operator docs for deploy operators, which suggests that self-hosting is a supported path, though it is not trivial given the Convex and OpenAI dependencies. The maintenance cost is tied to keeping up with Convex schema changes and OpenAI API updates. The repo layout includes specs and design history, which is a plus for understanding the evolution of the project. You should verify the current state of the docs, especially the CONTRIBUTING.md for environment setup, before committing to a deployment.

Editorial conclusion

Adopt ClawHub if you publish or consume OpenClaw skills and plugins and want a registry with versioning, rename aliases, and vector search. Skip it if you need hard deletion by regular users or if you cannot host a Convex backend and an OpenAI API key. Before adopting, verify the soft-delete semantics for your moderation workflow, test the CLI login flow with GitHub OAuth, and confirm that the telemetry disable flag works in your environment. The project is MIT licensed and actively maintained, but you must weigh the operational cost of running Convex and OpenAI embeddings against the convenience of a unified catalog.

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes