sickn33/antigravity-awesome-skills: 1,958 Agentic Skills for AI Coding Assistants

sickn33/antigravity-awesome-skills: 1,958 Agentic Skills for AI Coding Assistants

Developers using AI coding assistants face a recurring friction: every session starts from scratch. You re-explain your stack, restate your conventions, and re-negotiate output formats with Claude Code, Cursor, or Gemini CLI. The result is inconsistent quality, wasted context tokens, and mental overhead that compounds across projects. sickn33/antigravity-awesome-skills addresses this by providing an installable, searchable library of 1,958+ reusable SKILL.md playbooks—structured operating instructions that major AI assistants can load and follow. With 43,189 GitHub stars and active maintenance through July 2026, this MIT-licensed Python^{↗ Bright Coding Blog} project has become a practical infrastructure layer for developers who want their agents to start smart, stay consistent, and work across tools.

What is sickn33/antigravity-awesome-skills?

sickn33/antigravity-awesome-skills (also referred to as Agentic Awesome Skills, release 14.4.0) is an open-source GitHub repository that packages reusable skill definitions for AI coding assistants. It is not a model or IDE itself; it is a distribution layer for structured instructions that tools like Claude Code, Cursor, Codex CLI, Autohand Code, Gemini CLI, Antigravity, Kiro, OpenCode, and GitHub Copilot can consume.

The project is maintained by sickn33 as an independent community effort, explicitly unaffiliated with Google or any commercial AI platform. It sits at the intersection of developer tooling, AI agent infrastructure, and prompt engineering governance—a category that has grown rapidly as teams move from ad-hoc LLM prompting to reproducible, versioned agent workflows.

Each skill is a SKILL.md file with structured operating instructions: context, constraints, output formats, and step-by-step guidance for tasks ranging from brainstorming and security auditing to infrastructure deployment and product analytics. The repository also generates specialized plugins (domain-specific distributions), bundles (role-based recommendations), and workflows (ordered execution playbooks) so users can choose breadth or curation as needed.

The project's relevance is straightforward: as AI assistants gain longer context windows and tool-use capabilities, the quality gap between raw models and well-instructed agents widens. This library closes that gap with installable, composable, and version-pinned instructions rather than one-off prompts copied from forums.

Key Features

Installable by design, not just browsable. The core differentiator is the npm-based CLI installer (npx agentic-awesome-skills). Skills land in the exact directory your assistant expects—~/.agents/skills for Antigravity 2.0, ~/.gemini/antigravity-cli/skills/ for Antigravity CLI, .cursor/skills/ for Cursor, and so on. This eliminates manual path configuration and reduces setup errors.

Multi-tool compatibility with explicit paths. The README documents install targets for 11+ tools including Claude Code, Cursor, Codex CLI, Gemini CLI, Autohand Code, Antigravity IDE and CLI, Kiro CLI and IDE, GitHub Copilot, OpenCode, and AdaL CLI. Each has verified paths and invocation syntax, reducing the trial-and-error typically required when adapting community resources across platforms.

Three discovery surfaces for different mental models. The hosted Skill Workbench filters by risk level, provenance, host compatibility, and setup evidence before generating exact install commands. Specialized plugins (13 published, more in roadmap) package 8-10 skills for domains like web apps, security, DevOps^{↗ Bright Coding Blog}, QA, and SaaS launch. Bundles and workflows answer "which skills belong together?" and "what order should the agent run them?" respectively.

Stable discovery contract for integrations. The project publishes skills_index.json with a versioned schema (schemas/skills-index.v1.schema.json), so host integrations can discover skills programmatically without parsing markdown^{↗ Smart Converter} or scraping HTML.

Release-pinned, reproducible installs. The installer supports --tag, --release, and --pin flags; the hosted workbench generates version-locked commands. This matters for teams who need deterministic behavior across developer machines and CI environments.

Use Cases

Cross-tool standardization for platform teams. Engineering organizations using multiple AI assistants (Claude Code for backend, Cursor for frontend, Codex CLI for automation) can distribute the same skill definitions across tools. The npx installer with tool-specific flags (--claude, --cursor, --codex) ensures each assistant receives appropriately formatted instructions without maintaining separate prompt libraries.

Security-conscious development workflows. The AAS Security Engineer and AAS Secure App Builder plugins provide structured audit and hardening guidance. For regulated environments, the --risk safe,none filter and provenance metadata in the Skill Workbench help teams exclude unvetted community contributions.

SaaS MVP shipping with bundled workflows. The repository includes ordered execution playbooks for tasks like "ship a SaaS MVP" that sequence planning, implementation, testing, and security review skills. Teams can start from a workflow rather than assembling individual skills ad hoc.

Context-window management for long sessions. The agent-overload-recovery documentation and --category, --tags installer filters help users install only relevant skills, preventing assistants from loading excessive instructions into limited context windows. This is particularly relevant for OpenCode and similar .agents/skills hosts.

Community skill contribution and governance. Organizations building internal skill libraries can follow the repository's CONTRIBUTING.md, validation scripts (npm run validate), and automated skill-review checks as a template for their own governance.

Installation & Setup

The project supports three primary installation patterns: full library, specialized plugin, or single-skill via GitHub CLI.

Full library install (default)

# Default path: ~/.agents/skills (Antigravity 2.0 global)
npx agentic-awesome-skills

# Antigravity CLI slash commands (agy): ~/.gemini/antigravity-cli/skills/<skill>/SKILL.md
npx agentic-awesome-skills --agy

The installer uses a shallow, release-pinned clone by default. This keeps first-run installs lighter than a full git clone while matching the published npm package version. Use --tag main only when you intentionally want the current repository tip rather than a stable release.

Tool-specific installation

# Claude Code
npx agentic-awesome-skills --claude

# Cursor
npx agentic-awesome-skills --cursor

# Codex CLI
npx agentic-awesome-skills --codex

# Gemini CLI
npx agentic-awesome-skills --gemini

# Kiro CLI
npx agentic-awesome-skills --kiro

# Custom path for Autohand Code or other tools
npx agentic-awesome-skills --path ~/.autohand/skills
npx agentic-awesome-skills --path .autohand/skills

Each flag places skills in the directory convention expected by that tool. Verify the install with:

test -d ~/.agents/skills && echo "Skills installed in ~/.agents/skills"

Single-skill install with GitHub CLI (preview)

For Copilot and other supported hosts, the GitHub CLI preview supports exact-skill installation:

# Preview a skill before installing
gh skill preview sickn33/agentic-awesome-skills skills/brainstorming/SKILL.md

# Install with explicit agent, scope, and version pin
gh skill install sickn33/agentic-awesome-skills skills/brainstorming/SKILL.md \
  --agent github-copilot --scope user --pin v14.2.0

The gh skill interface is explicitly noted as a preview that may change. The README advises against --all unless you genuinely want every discovered skill, as this can overwhelm context-limited hosts.

Reduced install for context-sensitive hosts

# OpenCode or similar: install only development and backend skills, safe risk only
npx agentic-awesome-skills --path .agents/skills --category development,backend --risk safe,none

This pattern prevents runtime overload by filtering before installation rather than after.

Real Code Examples

The README provides concrete invocation patterns for each supported tool. These are not abstract examples—they are the documented syntax for activating installed skills.

Claude Code invocation

>> /brainstorming help me plan a feature

Claude Code uses slash-command syntax with the skill name. The --claude install flag ensures skills land in the directory Claude Code monitors for slash commands.

Cursor invocation

@brainstorming help me plan a feature

Cursor uses @ mentions to invoke skills. The --cursor flag places skills in .cursor/skills/ with the structure Cursor expects.

Antigravity IDE and CLI invocations

# Antigravity IDE
Use @brainstorming to plan a feature

# Antigravity CLI (agy)
/brainstorming help me plan a feature

Note the syntax divergence: the IDE uses @ mentions while the CLI uses / slash commands. The --antigravity and --agy flags handle these path differences automatically.

Reproducible exact-set install with dry-run

npx agentic-awesome-skills@14.3.0 --codex --release 14.3.0 \
  --skills frontend-design,game-development/2d-games --dry-run

This command demonstrates release pinning (@14.3.0), tool targeting (--codex), skill selection (--skills with comma-separated identifiers), and safe preview (--dry-run). Remove --dry-run only after reviewing the install plan. The README notes that unknown or ambiguous skill identifiers fail closed, and metadata filters combine with --skills using AND logic.

Verification command

test -d ~/.agents/skills && echo "Skills installed in ~/.agents/skills"

A simple POSIX check for the default install location. For non-default paths, adjust the directory accordingly.

Advanced Usage & Best Practices

Prefer specialized plugins for focused work. The README is explicit: "Do not install everything first if you already know the work." The AAS Web App Builder, AAS Security Engineer, and similar plugins reduce noise and context consumption. Install the full library only when you need broad discoverability or work across domains in a single session.

Use the Skill Workbench for exact-set composition. The hosted workbench at sickn33.github.io/agentic-awesome-skills/workbench exposes conflicts, risk levels, and host compatibility before generating commands. This is preferable to manual --skills lists for complex selections.

Pin releases in team environments. The --release and --pin flags, plus npm version pinning (npx agentic-awesome-skills@14.3.0), ensure reproducibility. Without pinning, npx resolves to the latest version, which may introduce new skills or structure changes that affect agent behavior.

Monitor for context overload. The agent-overload-recovery.md documentation and --category, --tags, --risk filters exist because too many active skills can degrade performance. For hosts like OpenCode with .agents/skills directories, prefer reduced installs upfront rather than pruning after the fact.

Contribute source-only, not generated artifacts. The CONTRIBUTING.md specifies that community PRs should include only SKILL.md source files and pass npm run validate. Generated files like CATALOG.md and skills_index.json are rebuilt automatically—committing them creates merge conflicts.

Comparison with Alternatives

Awesome Claude Skills (referenced in the README's comparison doc) offers curation over breadth. It suits developers who want a smaller, manually vetted set rather than a searchable catalog. Vexilo serves a different purpose entirely: it indexes Claude Code's built-in primitives (agents, commands, skills, rules) as a navigation layer, not a skill library. Neither provides the installable, multi-tool, version-pinned distribution that sickn33/antigravity-awesome-skills targets.

The trade-off is real: this project's breadth (1,958+ skills) requires more filtering discipline. Users who install everything without --category or --risk filters may experience context degradation. The specialized plugins and Skill Workbench exist specifically to mitigate this.

FAQ

What license covers the skills? MIT License. Individual skills may inherit their source repository's license; check provenance metadata in the Skill Workbench.

Is this officially affiliated with Google, Anthropic, or OpenAI? No. The README explicitly states: "This is an independent community project. It is not affiliated with, sponsored by, endorsed by, or authorized by Google."

How do I know which skills are safe to use? The Skill Workbench filters by risk level (safe, none, etc.) and provenance. The --risk safe,none installer flag restricts to vetted skills.

Why does gh skill install require --pin? Exact-skill installation requires unambiguous source identification in this large, mirrored repository. Version pinning prevents drift.

Can I use this with self-hosted or custom AI assistants? Yes. The --path flag installs to any directory. Ensure your assistant monitors that path for SKILL.md files.

How often is the repository updated? Last commit was 2026-07-14 per repo stats. Release 14.4.0 is current as of that date.

What happens if I install too many skills? Context-sensitive hosts may degrade. Use --category, --tags, or --risk filters, or consult agent-overload-recovery.md.

Conclusion

sickn33/antigravity-awesome-skills solves a concrete infrastructure problem for developers using AI coding assistants: it replaces one-off prompts with installable, versioned, structured instructions that work across tools. With 1,958+ skills, 43,189 GitHub stars, and explicit support for 11+ assistants, it is best suited for teams and individuals who need breadth with governance—the ability to search a large catalog while filtering by risk, domain, and host compatibility.

It is not a replacement for curated lists like Awesome Claude Skills, nor a navigation layer like Vexilo. It is a distribution system for agent instructions, with the complexity that entails. Developers who invest time in understanding the plugin, bundle, and workflow surfaces will get more value than those who run npx agentic-awesome-skills without filters.

If you use Claude Code, Cursor, Codex CLI, Gemini CLI, or any supported assistant, start with a specialized plugin matching your current task, verify the install, and expand from there. Explore the repository at https://github.com/sickn33/antigravity-awesome-skills.