ghst CLI Skill

This skill wraps the ghst CLI tool, which allows the agent to interact directly with any Ghost instance via the Ghost Admin API.

Prerequisites & Installation

For this skill to function, the ghst binary must be available in the system's $PATH.

Local Installation

Install the CLI globally from the official npm registry using npm:

npm install -g @tryghost/ghst

Alternatively, you can run it via npx without a permanent installation (though this is slower for repeated agent tasks):

npx @tryghost/ghst --help

Security: Always install @tryghost/ghst from the official npm registry (https://www.npmjs.com/package/@tryghost/ghst). Consider pinning to a specific, reviewed version rather than relying on whatever version resolves at runtime (e.g., npm install -g @tryghost/ghst@1.2.3). Keep the package updated intentionally and review release notes before upgrading.

Docker / Containerized Environments

If you are running OpenClaw in Docker, ensure the ghst binary is included in your container image by adding the installation command to your Dockerfile:

RUN npm install -g @tryghost/ghst

Configuration & Authentication

To interact with a Ghost instance, the agent requires a Ghost API URL and a Ghost Staff Access Token. There are two main ways to achieve this within the ghst skill:

1. Explicit flags: --url and --staff-token
2. Environment variables: GHOST_URL and GHOST_STAFF_ACCESS_TOKEN

Staff Account & Token Security

To minimize risk, create a dedicated Ghost staff account specifically for agent access and grant it only the least privileges required (e.g., Author or Editor rather than Owner). Store the GHOST_STAFF_ACCESS_TOKEN securely—never commit it to version control or expose it in chat logs. If the token is ever leaked, rotate it immediately via the Ghost Admin panel, and disable or remove the token when the agent no longer needs access.

Instructions for Bot Owners

1. Navigate to your Ghost Admin panel.
2. Go to Settings -> Staff (or Users) -> Edit your User Profile.
3. At the bottom, generate or copy a Staff Access Token.

Option 1 — ~/.openclaw/.env (recommended for personal use)

Add the variables directly to your ~/.openclaw/.env file:

GHOST_URL="https://your-blog-url.ghost.io"
GHOST_STAFF_ACCESS_TOKEN="your-staff-access-token-id:secret"

Option 2: openclaw.json skill entry (per-skill injection)

In ~/.openclaw/openclaw.json, add an env block under your skill's entry:

{
  "skills": {
    "entries": {
      "ghst": {
        "enabled": true,
        "env": {
          "GHOST_URL": "https://your-blog-url.ghost.io",
          "GHOST_STAFF_ACCESS_TOKEN": "your-staff-access-token-id:secret"
        }
      }
    }
  }
}

Once configured, restart your agent or wait for the config to be picked up.

Advanced Environment Variables

For complex setups, the following environment variables are supported:

Agent Guidelines & Usage

When operating the ghst skill, the agent must adhere to the following rules to ensure robust and safe usage.

1. Robust Scripting

• Always use --json or --jq: Ensure your commands produce machine-readable JSON output rather than human-readable text.

  ghst post list --json
  ghst post list --json --jq '.posts[].title'
  

• Use --non-interactive: Since you are running in an automated environment, never issue commands that prompt for user input unexpectedly. Use --yes in combination with --non-interactive for any required destructive operations that you have received user approval for.

  ghst comment delete <comment-id> --yes --non-interactive
  

• Non-Interactive Auth: If you need to authenticate a new site programmatically:

  ghst auth login --non-interactive --url "https://blog.com" --staff-token "..."
  

2. Editing Posts and Pages

For detailed instructions on the "Read-Edit-Write" workflow to edit post or page lexical content (including minor rewording and URL changes), see the editing.md reference.

3. Searching and Filtering Posts and Pages

When you need to find specific posts or pages (e.g., by title, status, or tag), refer to the advanced filtering and NQL query examples documented in post.md and page.md.

Command Reference

Detailed documentation for each resource can be found in the references/ directory:

Example Workflows:

• Bulk Post Tagging: ghst post bulk --filter "status:draft" --update --add-tag "Release"
• Member Cleanup: ghst member bulk --filter "status:free" --action delete --yes --non-interactive
• Analytics Export: ghst stats posts --range 30d --csv --output ./report.csv
• Social Interaction: ghst socialweb note --content "Hello from the CLI"

4. Safe Operation & Protections

• Approvals & Notices: You must obtain explicit user confirmation before performing any of the following actions:

  • Publishing, scheduling, or unpublishing posts/pages
  • Deleting posts, pages, members, comments, labels, themes, or any resource
  • Bulk updates or bulk deletes (ghst ... bulk ...)
  • Changing site settings (ghst setting set)
  • Creating, updating, or deleting webhooks (ghst webhook)
  • Importing or exporting data (ghst migrate, ghst member export)
  • Raw API calls (ghst api)

  Note: The CLI emits GHST_AGENT_NOTICE: lines on stderr when a manual confirmation is interrupted. If you see this, you must stop and ask the user for explicit approval.

• Prefer Listing / Fetching First: Before performing any destructive or mutating action, prefer to list or fetch the target resource to verify its identity. Always use exact IDs or exact slugs rather than fuzzy matching or assumptions.

• Destructive Commands: Only use --yes --non-interactive for the following once the user has explicitly approved:

  • ghst member bulk --action delete
  • ghst label bulk --action delete
  • ghst socialweb delete
  • ghst auth logout
  • ghst auth link (when replacing active link)

• File Safety: CLI tools like member export and migrate export will refuse to overwrite existing files. Check for file existence before exporting if necessary.

• Security Check: Never print or output sensitive tokens (e.g., values coming from ghst auth token or config --show-secrets) into the chat unprompted. Treat them as privileged credentials.