Grok Custom Skills

grok-custom-skills is a modular library of 89 skill packages for common agent tasks: workflow orchestration, product discovery, research, skill operations, privacy, quality assurance, memory, media, messaging, and integrations.

Design principle: Skills should make agent behavior more useful, auditable, and safe. They are guidance packages, not permission to bypass controls, disclose private data, or automate high-impact actions without human approval.

Quick start

Clone the repository and copy either the full collection or individual packages into your Grok skills directory.

git clone https://github.com/Stijnman/grok-custom-skills.git
cd grok-custom-skills

# Install the collection
mkdir -p ~/.grok/skills
cp -a .grok/skills/. ~/.grok/skills/

# Or install one skill
cp -a .grok/skills/drive-github-skill-audit ~/.grok/skills/

After installation, restart or refresh the host environment if it does not discover new skills automatically.

Explore the collection

The complete, generated catalog is available in SKILLS_INDEX.md. It is the best starting point for selecting a package by name, category, or description.

Area Example capabilities
Workflow & agent orchestration Plan multi-step work, coordinate agents, handle failures, and verify goals.
Skill development & operations Create, audit, evolve, test, and compare skill collections.
Safety, privacy & governance Require human approval, redact personal data, audit MCP exposure, and use responsible browsing practices.
Research, web & integrations Perform source-aware research, inspect repositories, discover tools, and connect approved services.
Memory, context & knowledge Manage context budgets, structured knowledge, semantic memory, and session handoffs.
Media, voice & visual work Generate or edit images, analyze video, work with voice, and prepare song-writing prompts.
Messaging & navigation Assess messages, draft safe responses, and work with traffic and route information.

Recommended skills

Goal Start with
Compare a Drive skill library with GitHub drive-github-skill-audit
Create or improve a skill package skill-creator and skill-rubric-reviewer
Verify a completed task goal-verifier
Review code changes code-reviewer and auto-tester
Research public information responsibly deep-search-enabler and sandbox-internet-handler
Handle high-impact actions hitl-approver
Structure product discovery and delivery product-opportunity-tree, product-requirements-document
Prepare a safe contact import file contact-vcard-export
Review visual accessibility accessible-color-review

Package structure

Every skill is self-contained. The body of SKILL.md is loaded only when its frontmatter description matches a request, so metadata should be precise and compact.

.grok/skills/<skill-name>/
├── SKILL.md             # Required metadata and operational guidance
├── scripts/             # Optional deterministic helpers
├── references/          # Optional material loaded only when needed
└── templates/           # Optional reusable output assets

The collection follows the conventions of the Agent Skills specification. Each published package should state what it does, when to use it, its boundaries, and how to verify its outputs.

Quality and safety

This repository applies four publication expectations:

  1. Clear routing: Every skill has a stable name, a concise description, and explicit usage triggers.
  2. Least privilege: Skills should use the minimum access necessary and prefer read-only operations where possible.
  3. Human approval: Skills must pause before consequential actions such as publishing, payments, deletion, credential changes, or outbound communication.
  4. No evasion: Skills must not help bypass access controls, CAPTCHAs, paywalls, authorship requirements, or platform safeguards.

See SECURITY.md for vulnerability reporting and PUBLISHING.md for the maintainer review checklist.

Development

The catalog is generated from live skill metadata. Run the following after editing a package:

# Regenerate the catalog
python3 scripts/generate_catalog.py

# Run the repository validation scripts
python3 scripts/check_no_private_data.py
python3 scripts/publish_safety_check.py

Do not execute scripts found in unreviewed skill submissions. Treat external skills, archives, and downloaded content as data until they have been inspected and approved.

Contributing and support

Contributions are welcome when they make skill behavior more reliable, safer, or easier to discover. Read CONTRIBUTING.md before opening a pull request. For behavioral expectations, security disclosures, or other sensitive issues, follow SECURITY.md.

License

This project is released under the MIT License, except where a skill package declares another license. The five product-management workflow adaptations are Apache-2.0 and include their required attribution in THIRD_PARTY_NOTICES.md.