A unified skill set for AI coding agents combining best practices from the top agent skill repositories:
| Source | Stars | Focus |
|---|---|---|
| mattpocock/skills | ~66k | TDD, debugging, shared language |
| addyosmani/agent-skills | ~35k | Production-grade workflow phases |
| forrestchang/andrej-karpathy-skills | ~120k | Behavioral guidelines |
| VoltAgent/awesome-design-md | ~74k | Design system patterns |
Copy these files to your project root:
project/
├── .grok/ # Grok (recommended for this TUI)
│ └── skills/ # (or symlink the central skills/)
├── .cursor/rules/ # Cursor rules
│ ├── ultimate-dev-workflow.mdc
│ └── ...
├── CLAUDE.md # Global agent guidelines (also loaded by Grok)
├── GROK.md # Primary for Grok (new/optimized)
├── AGENTS.md # Project conventions & skill mapping
├── CONTEXT.md # Domain language glossary (customize per project)
├── docs/adr/ # Architecture Decision Records
│ └── 0000-template.md
└── references/ # Quick-reference checklists
├── testing-patterns.md
├── security-checklist.md
└── performance-checklist.md
# Symlink for Cursor
ln -s ~/agent-skills/.cursor/rules .cursor/rules
# For Grok: the update script installs everything properly (see below)
# Or manually:
mkdir -p ~/.grok/skills
# then copy or junction the skills/ subfolders
ln -s ~/agent-skills/CLAUDE.md CLAUDE.md
ln -s ~/agent-skills/GROK.md GROK.md
ln -s ~/agent-skills/AGENTS.md AGENTS.md
ln -s ~/agent-skills/references referencesBest for Grok: Run update-skills.ps1 (after any edit). It installs:
GROK.md(and compat files) to~/.grok/- All modular skills to
~/.grok/skills/<name>/→ they become/spec-driven-development,/test-driven-development, etc. slash commands and auto-triggerable.
Now any update to this central "Agent Skills" repo can be propagated to all your projects and all your agents (Grok, Cursor, Claude, Gemini).
- State assumptions explicitly
- Present multiple interpretations
- Push back on problematic approaches
- Stop when confused, ask
- No features beyond what was asked
- No premature abstractions
- 100 lines where 1000 would do = failure
- Touch only what you must
- Don't "improve" orthogonal code
- Clean up only your own mess
- "Fix the bug" → Write reproduction test first
- "Add validation" → Write tests for invalid inputs
- Multi-step tasks need explicit verification points
DEFINE ──→ PLAN ──→ BUILD ──→ VERIFY ──→ REVIEW ──→ SHIP
│ │ │ │ │ │
▼ ▼ ▼ ▼ ▼ ▼
Spec Tasks Code Test Quality Deploy
| Phase | Skill | One-Liner |
|---|---|---|
| Define | grill-with-docs |
Refine requirements |
| Define | spec-driven-development |
Write structured spec |
| Plan | planning-and-task-breakdown |
Break into verifiable tasks |
| Build | incremental-implementation |
Build slice by slice |
| Build | test-driven-development |
Red-Green-Refactor |
| Verify | debugging-and-error-recovery |
Reproduce → Hypothesize → Fix |
| Review | code-review-and-quality |
Five-axis quality gates |
| Ship | git-workflow-and-versioning |
Atomic commits |
| Ship | shipping-and-launch |
Pre-launch checklist |
| Intent | Skill to Apply |
|---|---|
| Vague idea | grill-with-docs or start with spec |
| New project | spec-driven-development |
| Need tasks | planning-and-task-breakdown |
| Implementing | test-driven-development (TDD + Prove-It) |
| Bug fix | debugging-and-error-recovery + Prove-It |
| UI work | frontend-design + DESIGN.md |
| Shared language | shared-language |
| Code review | code-review-and-quality |
| Master workflow | ultimate-dev-workflow |
RED: Write ONE failing test
GREEN: Write minimal code to pass
REFACTOR: Clean up while tests pass
1. Write reproduction test (should FAIL)
2. Implement fix
3. Test PASSES
4. Run full suite
1. Reproduce (build fast signal)
2. Hypothesize (3-5 ranked)
3. Instrument (one variable)
4. Fix + regression test
5. Cleanup
- Correctness — Matches spec? Edge cases handled?
- Readability — Can another engineer understand?
- Architecture — Fits system design?
- Security — Input validated? Secrets protected?
- Performance — N+1 patterns? Unbounded ops?
| Prefix | Required? |
|---|---|
| (none) | Yes |
| Critical | Yes (blocks merge) |
| Nit | No |
| FYI | No |
| File | Purpose |
|---|---|
CLAUDE.md |
Global guidelines for all agents |
AGENTS.md |
Project conventions and skill mapping |
CONTEXT.md |
Domain language glossary |
.cursor/rules/ultimate-dev-workflow.mdc |
Cursor rule for development workflow |
docs/adr/0000-template.md |
Template for Architecture Decision Records |
For UI work, create a DESIGN.md file following the Stitch format:
# Design: [Project]
## Visual Theme
Dark-first / Light-first, mood, density
## Colors
- Primary: #hex (role)
- Secondary: #hex (role)
## Typography
- Headings: Font, size, weight
- Body: Font, size, weight
## Components
### Button
- Primary: solid bg, white text
- States: hover, active, disabledCopy from VoltAgent/awesome-design-md for popular design systems.
| "Rationalization" | Reality |
|---|---|
| "Simple, no spec needed" | Even simple needs acceptance criteria |
| "I'll write tests after" | You won't |
| "I know the bug, just fixing it" | 30% of obvious fixes are wrong |
| "It works, good enough" | Working + unreadable + insecure = debt |
| "I'll clean up later" | Later never comes |
- After editing any skill or guideline, run
.\update-skills.ps1from this folder. - Restart Grok or run
grok inspect(or the in-TUI/skillscommand). - In any project, drop
GROK.md(or symlink) + rely on~/.grok/skills/for the modular ones. - Grok will auto-load
GROK.md/AGENTS.md/CLAUDE.mdas Project Rules and discover the installed skills for both slash commands and automatic intent-based invocation.
The individual skills/<name>/SKILL.md files have been tuned with rich description + when-to-use fields and Grok-specific notes (heavy use of todo_write, run_terminal_command for verification, read_file for loading context, etc.).
This skill set combines best practices from:
- mattpocock/skills - ~66k stars
- addyosmani/agent-skills - ~35k stars
- forrestchang/andrej-karpathy-skills - ~120k stars
- VoltAgent/awesome-design-md - ~74k stars
Last updated: 2026-05 (Grok-optimized edition with native ~/.grok/skills/ support)