How to use the ClawHub Staff Moderation agent skill

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file. The file tells an AI agent when to use the skill and how to carry out a task. This skill covers ClawHub staff moderation actions: skills, users, org publishers, plugin packages, trusted publishers, official publishers, and staff email.

The skill does not give the agent new powers. It tells the agent to run the repo-local admin tool with commands such as bun run admin, and it tells the agent to follow strict rules around writes. Server role checks and audit logging still apply to every action.

Install the skill with a ZIP download or the Claude apps

Download the ZIP for this skill from this page and unzip it. For a personal install, place the resulting folder in ~/.claude/skills/. For a project install, place it in the project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/clawhub-moderation/
• Project: .claude/skills/clawhub-moderation/
• Claude apps: Settings → Capabilities → Skills, then upload

Meet the requirements before you start

The skill assumes you already have a checked-out ClawHub repository and the bun runtime, because commands run as bun run admin. You also need a ClawHub admin API token with the staff role, which you obtain with the admin tool's login command. The machine needs network access to the ClawHub server.

Without a verified admin token the skill cannot do anything useful, and the server will not bypass its role checks.

• A checked-out ClawHub repository
• bun runtime installed
• ClawHub admin API token with staff role, obtained via the login command
• Network access to the ClawHub server

Ask for tasks in natural language

You do not need to remember command syntax. Name the target and the action, and include a reason for any destructive, restorative, ownership, or moderation change. The skill requires an explicit target: a skill slug, a user handle, or a user id.

For example, you might ask the agent to show the moderation status of a plugin package, or to unban a specific user handle with a stated reason. The agent will show the exact command and ask you to confirm before any write.

• Good: "Unban handle alice-dev, reason: appeal approved by moderator team."
• Vague: "Fix that user." The agent should ask for the handle and reason first.

Understand the safety flow

Before any write, the agent shows the exact command and asks for confirmation unless you have already said to proceed or supplied --yes. It verifies the state afterwards and reports the result. Email is held to a stricter standard: the agent only sends after you explicitly ask for the email and approve the final recipient, subject, and body. Drafting with a dry run is allowed at any time.

Some commands default to a dry run. For example, skills hard-delete and org delete show what would happen first, and hard-delete only applies with the exact confirmation token returned by that dry run.

• Require a target and a reason for writes
• Show the exact command and wait for confirmation
• Use dry runs where the tool offers them
• Verify the result after the write

Safety notes: scripts and network access

This skill runs scripts and uses the network. Every command goes to the ClawHub server through the admin API, so run it only on a machine you trust and with a token you are authorised to use. Many commands are destructive. Hard-delete, bans, and email sends should always have a human confirming each step.

Do not use this skill for unattended automation. Do not ask the agent to bypass API-token auth, server role checks, or audit logging; the skill itself forbids this.

• Use only with a verified staff token
• Keep a human in the loop for every destructive step
• Ban actions revoke API tokens, mark the user deleted, hide owned skills, and soft-delete comments
• Record the reason in your task record, since rescan has no reason flag

Tips for reliable moderation work

Prefer user handles over ids, and use --id only when you have a user id. Use --fuzzy only when the user has asked for it or the exact handle is ambiguous. When more than one publisher uses the same slug, pass --owner to revoke-version.

There is no standalone skills hide command in clawhub-admin. To hide a skill as part of resolving a report, use report triage with --action hide.

• Start with bun run admin -- --help to see the command groups
• Check status before and after changes
• Use bun run admin -- skills reports --status open to review open reports