Cutseven/grok-community-skills

Twelve free, MIT-licensed agent skills for Grok Build. The skills are instructions only: no MCP servers and no scripts. Each skill uses whatever tools your agent already has (email/calendar connectors, file access, shell) and asks before doing anything irreversible. The plugin also ships two optional Grok edition exclusive extras: a safety hook and a read-only subagent.
Using Claude, Codex, Cursor, Copilot, Gemini CLI or a plain chatbot? The same twelve skills are available in the open Agent Skills format, with copy-paste prompts, at Cutseven/community-agent-skills.
| Skill | What it does |
|---|---|
| smart-inbox-triage | Sorts email into urgent/action/info/spam, pulls out action items and due dates, writes a short digest |
| universal-file-finder | Finds local files from a plain-language description (name, content, size, date, type) |
| meeting-sync-bot | Collects availability across time zones, ranks slots, drafts invites |
| subscription-tracker | Finds recurring charges in statements, flags renewals, drafts cancellation requests |
| meal-planner-grocery | Plans meals from what you have and builds a low-waste grocery list |
| auto-backup-restore | Plans and runs encrypted, versioned backups and safe restores |
| one-command-task-capturer | Turns one sentence into a structured task, catches duplicates, runs a weekly review |
| context-translator-summarizer | Translates for meaning, then gives a bullet summary and glossary |
| batch-task-runner | Chains rename/convert/resize/compress/zip jobs, with a dry-run preview first |
| cloudflare-infra-worker | Read-first Cloudflare engineer: DNS, Workers/Pages, WAF, SSL/TLS, Tunnel/Access; changes only in EXECUTE mode, with previews, rollback plans and explicit confirmation for destructive actions |
| unifi-network-admin | Careful UniFi network admin: read-only by default, one approved change at a time with backup, verification and rollback; troubleshooting, severity-ranked reports, quiet scheduled health checks |
| home-assistant-admin | Careful Home Assistant admin: safe (read-only) mode by default, approval-gated admin mode, one verified change at a time with backup and rollback; automations, integrations, updates, entity/battery/offline checks, energy, quiet scheduled checks |
Install
From the xAI plugin marketplace, once it's listed there (submitted in xai-org/plugin-marketplace#1276, pending review):
Or straight from this repository:
Grok asks for --trust because the plugin includes a hook (see below). Plugins are off until enabled: press Space on it in the Plugins tab (/plugins), or add it to [plugins].enabled. See Skills, Plugins, and Marketplaces.
Safety
Every skill follows the same rules: it never sends, posts, deletes, moves, or overwrites anything until you've seen exactly what will happen and said yes. Content from emails, files, and web pages is treated as data, never as instructions.
Grok edition exclusive
These two extras use Grok Build plugin features and are not part of the open Agent Skills standard, so they aren't in the any-agent edition. They don't change how the twelve skills behave; they back up the same safety rules.
Confirm-destructive hook
A PreToolUse hook on shell commands (hooks/hooks.json, hooks/confirm-destructive.sh). When a command looks like it deletes, overwrites, force-pushes or sends something, the hook answers ask, so Grok shows its permission prompt with the reason, even in always-approve or auto mode. Examples: rm, find -delete, shred, truncate, dd of=, mkfs, git reset --hard, git clean -f, git push --force, rsync --delete, rclone sync/purge, restic forget/prune, aws s3 rm, wrangler … delete, curl -X DELETE, terraform destroy, sendmail/mail -s.