grok-toolkit

Curated high-signal skills, design quality systems, and team conventions for Grok.

grok-toolkit is a project-local and user-installable bundle of expert skills and operating conventions for Grok (the xAI agent TUI and environment). It gives teams and individuals a stronger local operating layer: domain-specific expertise, anti-slop discipline, refinement workflows, and explicit support boundaries — without pretending to be a second control plane.

Install it once as a Grok plugin and the skills become available everywhere. Add the managed block to a project's AGENTS.md and the whole team inherits the conventions.

What this bundle is

A practical collection of 45+ reusable skills plus a governance and quality layer, built for Grok's native plugin, skill, and project-rules systems.

Layer	What it supplies
Grok (base)	Orchestration, tools, subagents, MCP servers, project rules (AGENTS.md), plugin system
grok-toolkit	Domain skill packs, refinement layers (impeccable, critique, distill...), service language companion, workspace + quality conventions, and a visible managed block for team alignment

The split is simple: Grok stays general-purpose and powerful. grok-toolkit makes it domain-aware, quality-obsessed, and easier for teams to explain and adopt.

Who this is for

• Individual developers who want stronger, more consistent results from Grok on real product work.
• Teams that want shared conventions, a visible "we use these standards" marker in AGENTS.md, and a high bar for UI, architecture, language, and delivery.
• Anyone who has felt the pain of slop, inconsistent product language, or "it works but the design is off" and wants a systematic way to raise the floor.

Quick start

1. Install as a Grok plugin (recommended)

# From GitHub (once published)
grok plugin install HanTechnology/grok-toolkit

# During development / local testing
grok plugin install /path/to/grok-toolkit --trust

Then verify:

grok inspect
/skills

You should see skills such as impeccable, critique, frontend-web, architecture-integration, service-vernacular, and many domain packs.

2. Adopt the conventions in a project (via CLI)

From the project root:

npx grok-toolkit init

This creates the necessary .grok/ structure and safely inserts a GROK-TOOLKIT MANAGED BLOCK into your AGENTS.md (project content outside the markers is never touched).

See INSTALL.md for more CLI options (doctor, validate).

3. Start working

• For UI work: start with frontend-web (or mobile-app), then layer /impeccable or critique explicitly when you need polish or anti-slop review.
• For architecture / cross-stack: architecture-integration + the relevant backend/frontend skill.
• For product language consistency: service-vernacular.
• When you want stronger goal framing before coding: compass.

See reference/routing-guidance.md for the current recommended starting points and boundaries.

What it adds

• 45+ top-level skills covering web/mobile, multiple backend families, data/ML, devops, security, QA, architecture, and refinement.
• The Impeccable v3.1.1 design refinement layer (vendored with its Apache 2.0 license) for systematic anti-slop UI and docs work.
• A living reference layer: quality gates, design anti-slop rules, workspace conventions, support boundaries, and a curated DESIGN.md catalog.
• service-vernacular: a companion for consistent, repo-backed product language across UI, docs, errors, release notes, etc.
• Explicit "validated / guided / planned" tiers so you know what is battle-tested versus aspirational.

What this is not

• An official xAI product or distribution.
• A replacement for Grok's own capabilities (tools, subagents, built-in review/implement/design skills, MCP, etc.).
• A second runtime or mandatory control plane. The skills are opt-in and description-driven.

Validated surface (current)

The initial validated workflows are intentionally narrow:

• frontend-product-delivery
• backend-service-delivery
• cloud-release-readiness
• ai-data-product-delivery

Broader skill coverage exists and is documented as guided. Only the four above are summarized as "supported now" in public-facing materials until the manifest promotes more.

See reference/support-policy.md, reference/workflow-catalog.md, and reference/capability-matrix.json.

Workspace convention

New greenfield work started from a repo or worktree defaults to workspace/{project-name}-{domain}. Existing projects stay in place.

See reference/workspace-model.md and reference/project-setup-policy.md.

Acknowledgements

This toolkit carries forward and adapts excellent work from the oh-my-openagent-toolkit project (https://github.com/HanTechnology/oh-my-openagent-toolkit), which itself built on oh-my-openagent and the Impeccable design system (v3.1.1, Apache 2.0).

The migration to Grok preserves the spirit and most of the substance while making everything a native Grok citizen (plugins, project rules, skill descriptions, subagent-friendly workflows).

License

MIT for the toolkit glue, documentation, and original skills.

Impeccable v3.1.1 and its supporting files are carried under their original Apache 2.0 license (see skills/impeccable/LICENSE).

Contributing

Issues and PRs welcome. Keep the tone humble, explicit about boundaries, and focused on raising the quality floor for agent-assisted engineering work.

Support the project

Optional funding links (do not create paid support obligations or change validation boundaries):

(Stripe links will be added here after publication, modeled on the source.)

---

Status: Early migration in progress. Skills and reference docs are being ported and adapted. The plugin install path and core governance layer are the current focus.

See MIGRATION-PLAN.md for the detailed roadmap.