Using ClawHub

Publishing

Publishing

Publishing sends a skill folder or plugin package to ClawHub under the owner you choose. ClawHub checks that your token can publish for that owner, validates the metadata, name, version, files, and source information, then stores the release and starts automated security checks.

If validation fails, nothing is published. New releases may also stay out of normal install and download surfaces until review finishes.

Skills

The simplest publishing path is the CLI. Sign in, then publish a local skill folder:

bash
clawhub loginclawhub skill publish ./my-skill \  --slug my-skill \  --name "My Skill" \  --owner <owner>

Use --owner <handle> when publishing to an org owner. Omit it to publish as the authenticated user. Publishing skips unchanged content. A new skill starts at 1.0.0, and later changes automatically publish the next patch version. Pass --version only when you need an explicit version.

Skill catalog metadata

Categories place a skill in the category filters on the ClawHub skills browse page. Topics become the filter chips offered inside a selected category. Set both when you publish:

bash
clawhub skill publish ./my-skill \  --categories development,operations \  --topics "git,worktree,cleanup"

Both flags take comma-separated values. Categories must be slugs from this list, matched exactly, so Development is rejected. Topics are free-form labels; ClawHub stores what you pass and displays the normalized form, so Git Worktree appears as #git-worktree.

Slug Description
integrations Connect services, fetch data, reconcile records, and operate APIs.
automation Build repeatable processes, scheduled jobs, pipelines, and orchestration.
research Search, browse, scrape, summarize, monitor, and extract web information.
development Inspect, edit, test, build, debug, and operate codebases.
productivity Manage tasks, calendars, email, meetings, projects, and business work.
communication Message, publish, and operate social or communication services.
creative Create and edit images, video, audio, music, design, and writing.
knowledge Work with documents, notes, knowledge bases, teaching, and learning.
agents Change how an agent plans, reflects, learns, remembers, or collaborates.
operations Inspect, monitor, deploy, and operate local systems or infrastructure.
security Audit, scan, authenticate, and protect systems or data.
finance Work with payments, budgets, banking, shopping, markets, and commerce.
lifestyle Travel, health, fitness, cooking, sports, home, and daily-life utilities.
other Skills that do not yet fit another browse category.

Rules ClawHub applies to both fields:

  • A skill can carry at most 3 categories and at most 5 topics.
  • An unknown category slug fails the publish. --dry-run does not check slugs; the registry validates them when the publish runs.
  • other is dropped when it is passed alongside a specific category. The 3-category limit is applied after that, so other,development,operations stores two categories rather than failing.
  • Repeats are dropped rather than rejected, and they are matched after normalization, so git,Git is one topic. Both limits count what is left after that, not what you passed.
  • Each topic is at most 48 characters, and topics cannot contain invisible formatting characters.
  • These topic names are reserved by ClawHub and are rejected: approved, audited, certified, clawhub, community, curated, endorsed, featured, official, officials, openclaw, recommended, staff-pick, trusted, trusted-publisher, verified. The check runs on the normalized form, so Official and staff pick are rejected too.
  • A skill first published without --categories is stored as other, so it only appears under the Other category.
  • On a later publish, omitting --categories or --topics keeps the values already stored. Pass the flag again to change them. Passing an empty value clears the field: --categories "" returns the skill to other, and --topics "" removes its topics.
  • Passing either flag publishes even when the files have not changed, so fixing metadata this way creates a new patch version.

Skill owners can also edit categories and topics from the skill's settings page on ClawHub. That is the quickest fix for a skill that was already published into other.

Publishing from a catalog repo

For catalog repos, use ClawHub's reusable skill-publish.yml workflow. It calls skill publish for each immediate skill folder under root (default: skills), or only the folder supplied as skill_path.

yaml
jobs:  publish:    uses: openclaw/clawhub/.github/workflows/skill-publish.yml@main    with:      owner: <owner>      dry_run: false    secrets:      clawhub_token: ${{ secrets.CLAWHUB_TOKEN }}

Use dry_run: true to preview new and changed skills without publishing.

The workflow forwards optional changelog, categories, and topics inputs to skill publish, plus clear_categories and clear_topics for removing metadata a skill already carries. A skill first published without categories is stored as other, the same as clawhub sync; you can also set catalog metadata later from the skill's settings page.

Like tags, categories and topics apply to every skill the run publishes, and supplying them suspends the unchanged-skill skip — the run releases a new patch version of each selected skill, including skills whose files did not change. Pass skill_path to bound that to one skill. See the workflow notes for the full behavior.

Plugins

Plugins use npm-style package names. Scoped package names include the owner in the first part of the name:

text
@owner/package-name

The scope must match the selected publish owner. If your package is named @openclaw/dronzer, it can only be published as @openclaw. If you publish as @vintageayu, rename the package to @vintageayu/dronzer.

This prevents a package from claiming an org namespace that the publisher does not control.

If you are the rightful owner of an org, brand, package scope, owner handle, or namespace that is already claimed or reserved on ClawHub, open an Org / Namespace Claim issue with public, non-sensitive proof. See Org and Namespace Claims for what to include and what to keep out of public issues.

Before Publishing a Plugin

  • Pick an owner that matches the package scope.
  • A code plugin's manifest id must be unique within that publisher's packages. Different publishers can distribute the same runtime id under distinct scoped package names. Choose the explicit scoped name to install a community package; known bare OpenClaw aliases select the official package. One OpenClaw installation still uses one plugin for each runtime id.
  • Include openclaw.plugin.json. Code plugins also need package.json with openclaw.compat.pluginApi and openclaw.build.openclawVersion.
  • To show a custom plugin catalog icon, include assets/icon.png at the package root in the published archive. Use a valid PNG no larger than 512 KiB.
  • Include source repository and exact commit metadata, or use the CLI from a GitHub-backed checkout so it can detect them.
  • Run clawhub package validate <source> before publishing. For package, manifest, SDK import, or artifact findings, see Plugin validation fixes.
  • Run clawhub package publish <source> --dry-run before creating a release.
  • Expect new releases to stay out of public install surfaces until automated security checks and verification finish.

An inspector operational failure reports the failing stage. A temporary-workspace cleanup failure also blocks publication and appears alongside any original inspection findings; it does not erase the primary error. These failures are distinct from plugin policy findings. Report the stage and Convex request id when asking maintainers to investigate; do not include credentials or package contents in diagnostic reports.

Plugin Icons

ClawHub uses the bundled assets/icon.png for plugin catalog artwork. Manifest icon URLs and paths are ignored. If your npm package uses a files allowlist, include the asset there and check npm pack --dry-run to confirm it will ship. Missing or invalid artwork falls back to the plugin category glyph.

Existing releases with only a manifest icon URL also use the category glyph. Publish a new release containing assets/icon.png to restore custom artwork. ClawHub validates and hosts the bundled PNG; an external image URL is not a supported substitute.

Maintainers can repair existing latest releases with maintenance:repairPluginIconsInternal after deploying bundled icon support. The repair uses an asset already in the release, or, for official @openclaw packages owned by the active OpenClaw publisher, the asset at the release's recorded OpenClaw source commit. Run the default dry run for each of family: "code-plugin" and family: "bundle-plugin", following each returned cursor until isDone. Apply from the initial cursor with dryRun: false, then repeat the dry run to verify no remaining matches. URL-only releases without a bundled asset cannot be restored by this repair.

Plugin Catalog Metadata

Choose the single main reason someone installs the plugin. Code plugins and bundle plugins can declare one category in their root openclaw.plugin.json:

json
{  "id": "example-agent-runtime",  "categories": ["agent-runtimes"]}

Merge the categories field into your existing manifest. Use a current plugin category slug, such as context, memory, or developer-tools; explicit declarations take priority over automatic classification. Each new release must declare exactly one category or omit the field.

When categories is omitted, ClawHub generates one category from bounded package metadata and documentation using its configured model (default: GPT-5.6 Luna). It falls back to other when classification is unavailable. A previous release's category is not an author declaration and is not preserved merely by omission.

The package CLI's --categories flag and reusable workflow's categories and clear_categories inputs are deprecated for plugins. They remain accepted but are ignored, with a warning on stderr. Move a chosen category into the manifest; remove the field to return to automatic classification. This also applies to --categories "", which no longer clears plugin categories. Experimental Claw publishes and skill publishing retain their existing category behavior.

Topics remain separate: use --topics or the workflow's topics input to set them, omit the input to preserve existing topics, or use --topics "" / clear_topics: true to clear them. See the package CLI reference and topic rules.

Trusted Publishing for Packages

Package trusted publishing is a two-step setup:

  1. Publish the package once through normal manual or token-authenticated clawhub package publish. This creates the package row and establishes the package managers who can change its trusted publisher config.
  2. A package manager sets the GitHub Actions trusted publisher config:
bash
clawhub package trusted-publisher set @owner/package-name \  --repository owner/repo \  --workflow-filename package-publish.yml

After config is set, future supported GitHub Actions publishes can use OIDC/trusted publishing without storing a long-lived ClawHub token in the repository. The configured repository and workflow filename must match the GitHub Actions OIDC claim. If you also pass --environment <name>, the GitHub Actions environment claim must match that name exactly.

ClawHub verifies the configured GitHub repository when trusted publisher config is set. Public repositories can be verified through public GitHub metadata. Private repositories require ClawHub to have GitHub access to that repository, for example through a future ClawHub GitHub App installation or another authorized GitHub integration.

The current reusable package publish workflow supports secretless trusted publishing for workflow_dispatch publishes when id-token: write is available. Tag-push real publishes still need clawhub_token, so keep CLAWHUB_TOKEN available for tag releases, first publishes, untrusted packages, or break-glass publishes.

Real publishes through the reusable workflow wait for the staged attempt to become public by default. The workflow fails when security checks block or fail the attempt, the attempt expires, or the 30-minute publication deadline is reached. Callers can adjust the deadline with publication_timeout_minutes. The maximum is 40 minutes, leaving 35 minutes of reusable job time for setup, upload, and output capture. Set wait_for_publication: false only for an intentional asynchronous publish.

Inspect or remove the config with:

bash
clawhub package trusted-publisher get @owner/package-nameclawhub package trusted-publisher delete @owner/package-name

Deleting trusted publisher config is the rollback path. It disables future trusted publish token minting until a package manager sets config again.

OpenClaw release recovery

OpenClaw automated releases stay non-public until their exact release-parent attempt succeeds. If that parent fails or is cancelled, the publish attempt fails permanently. A human recovery dispatch requires protected environment approval and a version 2 recovery receipt identifying the original authorized child. Recovery must use the same workflow ref and SHA, candidate, tooling, and package inventory. That workflow route does not accept cancelled parents.

Already-failed staged plugin attempts can instead be recovered under fresh publisher authority, without changing the old workflow or its outcome:

sh
clawhub package recover <attempt-id> \  --manual-override-reason "Retry the retained release artifacts after workflow failure" \  --wait --json

This uses a normal ClawHub user token and current package publish access. It creates a successor with the same retained artifacts and version, runs new security checks, and preserves the failed attempt and original authorization as audit history. Current token or publisher-access revocation still blocks publication. It cannot override moderation or revive an active attempt. Without --wait, the result is explicitly pending. The equivalent HTTP route is POST /api/v1/publish/attempts/{id}/recover.

Operators can preview orphaned package attempts, supplying an exact version, optional slugPrefix or attemptIds, and a reason:

sh
bunx convex run --prod maintenance:discardStalePackagePublishAttemptsInternal \  '{"version":"2026.9.1","slugPrefix":"@openclaw/","reason":"Release parent failed after staging"}'

It defaults to a dry run; add "dryRun":false to discard each pending release and retire its attempt. Signed-in admins can run the same operation as the maintenance:discardStalePackagePublishAttempts action. The reason appears as error at /api/v1/publish/attempts/<id>, so use publisher-facing wording. Published releases and terminal attempts are never discarded by this operation.

FAQ

Package scope must match selected owner

If the package scope and selected owner do not match, ClawHub rejects the publish:

text
Package scope "@openclaw" must match selected owner "@vintageayu".Publish as "@openclaw" or rename this package to "@vintageayu/dronzer".

To fix it, either choose the owner named by the package scope, or rename the package so the scope matches the owner you can publish as.

If the package name already has the right scope but the package is owned by the wrong publisher, transfer ownership instead:

sh
clawhub package transfer @opik/opik-openclaw --to opik

Package transfers require admin access to both the current owner and the destination publisher, unless performed by a platform admin. Use --to <owner> to select an existing, active destination publisher. Scoped package names can transfer only to the publisher matching their scope. See package transfer for details.

Skills use the separate ownership transfer workflow. Transfers to another user normally require recipient acceptance.

If you do not have access to the current owner but believe your org, project, or brand is the rightful namespace owner, open an Org / Namespace Claim issue with public, non-sensitive proof for staff review. See Org and Namespace Claims before filing.

This protects org namespaces. A package named @openclaw/dronzer claims the @openclaw namespace, so only publishers with access to the @openclaw owner can publish it.

Was this useful?
On this page

On this page