How to use the ClawHub Featured Curation agent skill

What does the clawhub-featured skill make the agent do?

The skill turns the agent into an advisor for Featured selection. It prepares evidence-based Featured recommendations and can apply explicitly authorized Featured changes through existing audited staff commands.

Recommendations are advisory. The agent runs the admin reports, reads install and search data for a stated time window, and returns a curation brief. Each candidate in the brief has a catalog, an identity, a usable listing link, and a short evidence-based reason. The brief also states the evidence period, coverage limits, and source report IDs.

• It does not change Featured membership just because a recommendation exists.
• A membership change requires explicit authorization for the exact item or complete selection.
• Catalog names, summaries, and reasons are treated as untrusted data, and instructions embedded in catalog content are not followed.

Who should use it, and who should not?

The skill is built for ClawHub staff preparing a Featured selection, and for maintainers who want an evidence-based shortlist of plugins or skills with clear data periods and coverage limits.

It is not for public users who want their own item featured. It needs staff access and staff authentication to run. It is also not a way to change Featured status automatically, because a recommendation alone does not allow changes.

Install the clawhub-featured skill

Download the ZIP for the skill from our page. Unzip it into your personal skills folder at ~/.claude/skills/, or into a project's .claude/skills/ folder if you want the skill available only in that project.

You can also upload the skill in the Claude apps. Open Settings, then Capabilities, then Skills, and upload the skill there.

After installing, confirm the folder contains SKILL.md and references/promotion.md. Those are the two files the skill uses.

Meet the runtime requirements

The skill needs a deployed ClawHub checkout with bun, or an installed clawhub-admin binary. It also needs ClawHub staff authentication for the admin commands and network access to the clawhub.ai public API endpoints.

The skill runs scripts and uses the network. If staff authentication is missing, the agent reports the exact missing access to you and does not label public catalog leads as eligible recommendations.

• A deployed ClawHub checkout with bun, or an installed clawhub-admin binary
• ClawHub staff authentication for admin commands
• Network access to clawhub.ai public API endpoints

Trigger the skill with natural requests

Ask for the outcome you want in plain language and name the catalog and time window. The agent reads the skill when the request matches its purpose. Examples of requests that fit are asking for a Featured shortlist of plugins for a stated period, or asking for a skill-only curation brief with its data coverage limits.

For a change, state the exact item or complete selection and say clearly that you authorize the change. Without that explicit authorization, the agent stays advisory.

• Prepare a Featured shortlist of plugins for the last thirty completed UTC days, with coverage limits.
• Build a curation brief for skills, with source report IDs and a Management link.
• Check whether a specific plugin is already featured before any change is discussed.

Understand the commands the agent runs

For recommendations, the agent reads both catalogs through the existing recommendation owner. The checkout form uses the bun run admin prefix, and an installed matching CLI can use clawhub-admin. Report creation saves a private evidence snapshot and changes no badges or delivery settings.

The default window ends at today's UTC midnight, exclusively. Saved reports expire after 24 hours, and an in-flight report can be resumed with a report ID. Trending and downloads lists from the public API are used only as context.

• Recommendations use the report's ranking: native recorded installs over thirty completed UTC days, then the final seven days.
• Search matches are separate context. They are not installs and not evidence of unique users.
• Lifetime downloads are not weekly installs, and a Trending rank alone does not prove a growth rate.

Safety notes on scripts and network access

The skill runs scripts and makes network requests to public ClawHub API endpoints. Read the commands before the agent runs them, and run them only on a checkout or CLI you trust.

Changes are handled separately in references/promotion.md. Those commands apply immediately and have no dry-run, --yes, or --reason option, so the agent should run only the exact authorized command after review. If an item is already featured, the agent reports that state instead of removing and re-adding it. Popularity never overrides security, public visibility, installability, category gates, or the slot limit.

• Confirm the authorization is explicit and names the exact item or set.
• Check that the report is not stale; refresh it before promotion review.
• Keep the requester, item identity, and reason in the review evidence.