antigravity-awesome-skills — Claude MCP 必备工具中文教程

Installation

For Codex and Claude, start with the AAS Core guide: configure the local MCP, ask the agent to inspect the project and choose exact IDs from the full catalog, review the proposed aas-stack.json, then run CLI validation and planning. The MCP and validation are read-only. Planning writes only the requested plan artifact; it does not materialize skill payloads or AAS managed state in the target.

Use direct installation when your host does not yet have a native AAS Core adapter, when you already know the exact skill IDs, or when you deliberately prefer manual selection:

• Specialized plugins when the job has a clear domain.
• Full library install when you want every skill available in a local skills directory.
• Bundles and workflows when you want role-based recommendations or ordered execution playbooks.

Direct skill install

```bash

Focused single-skill install with GitHub CLI (preview)

GitHub CLI can preview and install one exact skill for Copilot and other supported hosts. Use an exact SKILL.md path in this large, mirrored repository so the selected source is unambiguous and discovery stays fast:

gh skill preview sickn33/agentic-awesome-skills skills/brainstorming/SKILL.md
gh skill install sickn33/agentic-awesome-skills skills/brainstorming/SKILL.md \
  --agent github-copilot --scope user --pin v14.2.0

gh skill support is currently a GitHub CLI preview and may change. Install a focused skill or plugin surface for the job; do not use --all unless you intentionally want every discovered canonical and mirrored skill.

Verify the install

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

How do I install it?

For AAS Core, follow the preview guide and use only a package release whose notes explicitly state that it includes Core. Release 14.6.0 predates Core; Core-capable releases begin with the 15.x line.

For direct skill distribution, use a tool-specific flag such as --codex, --cursor, --gemini, or --claude to place skills in the directory your assistant watches. The default target is Antigravity; it requires --skills, a metadata filter, or explicit --all consent before cloning or writing because a full watched catalog can exhaust context or trigger a crash loop.

For Autohand Code, use the installer with a custom path:

npx agentic-awesome-skills --path ~/.autohand/skills
npx agentic-awesome-skills --path .autohand/skills

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

npx agentic-awesome-skills --agy

On `main`, the npm installer uses a shallow partial clone and checks out the canonical `skills/` tree after verifying the cloned commit against the immutable `gitHead` recorded for that exact npm package version. Git 2.25+ is required. Plugin mirrors, documentation and app assets are excluded from that temporary checkout; every canonical skill and support file remains available. See the [measured distribution comparison](docs/maintainers/distribution-efficiency.md). If the GitHub tag moved or npm identity metadata is unavailable, installation stops before copying content. Use `--tag main` only when you intentionally accept a mutable, explicitly unverified repository ref.

Antigravity watches `~/.agents/skills` and may load enough installed instructions
to exhaust its context, slow startup, trigger truncation errors, or enter a crash
loop. For that target, the installer stops before cloning or writing unless you
provide `--skills`, a metadata filter, or the explicit `--all` override. The bare
`npx agentic-awesome-skills` command uses the same protected Antigravity target.

The recommended flow is to ask Codex or Claude with the read-only AAS Core MCP
configured to inspect the project, search the complete catalog, and choose exact
skill IDs. The agent selects the IDs; AAS MCP validates them without installing; the agent
or user then previews the direct installation with the command above and repeats
it without `--dry-run` after review.

Other direct-install targets retain the legacy-compatible full-catalog behavior
when no selectors are supplied. The CLI prints the catalog's risk summary first:
a full install includes `critical` and authorized-use-only `offensive`
instructions. Installation copies files; it does not execute their commands,
but an agent may act on an installed skill later. Prefer an exact reviewed set:

If you deliberately accept the context and crash-loop risk, the complete
Antigravity catalog remains available through explicit consent:

The audit reads the selected skill directories without executing them and reports command, network, credential, filesystem, privileged, destructive, symlink, and binary signals. It is a review aid, not a safety certificate. See Security, trust, and antivirus alerts.

Prefer plugins for Claude Code, Codex, or another compatible client?

• Use a specialized plugin when you want a focused marketplace-style distribution.
• Use the full-library plugin only when you want the widest plugin-safe catalog.
• Read Plugins for compatible agent clients for host-specific installs, portable Agent Plugins bundles, and direct skills installs.

Recommended Specialized Plugins

Do not install everything first if you already know the work. Start with the focused plugin for your job, then add more only when the task expands.

All specialized plugins are generated as Claude Code and Codex plugin bundles. Bundles with flat, cross-host-safe skill layouts also receive a standard Agent Plugins 1.0 root manifest. For Antigravity, use the same SKILL.md content through the installer or supported skills paths.

Next-wave plugins cover marketing/SEO/growth, automation, observability/incident response, Python APIs, mobile apps, data engineering, privacy/compliance, and localization/international growth.

• Read the specialized plugin roadmap.
• Read the plugin guide for compatible agent clients.
• Compare the hosted specialized plugin landing page.
• Browse the generated plugin folders in plugins/.

What are AAS specialized plugins?

AAS specialized plugins are focused, domain-specific distributions of the skill library. They package the most relevant skills for web apps, security, data analytics, documents, DevOps, QA, OSS maintenance, and agent or MCP work so users can start with the right surface instead of activating the entire catalog.

Should I use the full library or a plugin?

Use the full library if you want the biggest catalog and direct filesystem control. Use a specialized plugin when you want a smaller, marketplace-style distribution for a specific workflow in Claude Code or Codex. For Antigravity, install the matching skills into the supported skills path. The complete explanation lives in Plugins for Claude Code and Codex.

How are plugins, bundles, and workflows different?

Plugins are installable packaging surfaces, bundles are curated skill recommendations, and workflows are ordered execution playbooks. Start with a plugin when the domain is clear, use bundles to compare adjacent skills, and use workflows when the important part is sequencing planning, coding, testing, auditing, or release work.

Where do I browse plugins, bundles, workflows, and the full catalog?

Start with Specialized Plugins when you want an installable domain pack. Use Bundles for role-based recommendations, Workflows for ordered execution playbooks, CATALOG.md for the full registry, and the hosted GitHub Pages catalog for searchable browsing.

Bundles & Workflows

Core, plugins, bundles, and workflows answer different questions. Codex or Claude selects skills; AAS Core records and validates that selection; plugins package a focused delivery surface; bundles are curated starting points; workflows are ordered playbooks for getting a result.

Use a specialized plugin first when your domain is clear. Use bundles to explore adjacent skills or assemble a custom install. Use workflows when the hard part is sequencing the work.

Use workflows for outcome-driven execution

• Read docs/users/workflows.md for human-readable playbooks.
• Use data/workflows.json for machine-readable workflow metadata.
• Initial workflows include shipping a SaaS MVP, security audits, AI agent systems, QA/browser automation, and DDD-oriented design work.

Compare alternatives

• Agentic Awesome Skills vs Awesome Claude Skills for breadth vs curated-list tradeoffs.
• Best Claude Code skills on GitHub for a high-intent shortlist.
• Best Cursor skills on GitHub for Cursor-compatible options and selection criteria.
• Vexilo · A field guide to Claude Code — different scope: a visual, searchable index of every Claude Code primitive (31 agents / 99 commands / 123 skills / 13 rules), organized around the 5-step workflow. Useful as a navigation layer over any skill library, not as a skill library itself. (companion repo)

Quick FAQ

Troubleshooting

Keep the root README short; use the dedicated docs for recovery and platform-specific guidance.

• If you are confused after installation, start with the Usage Guide.
• For Core setup, trust boundaries, stack manifests, and preview status, use the AAS Core guide.
• On native Windows, AAS_ADAPTER_WINDOWS_ACL_FAILED refers to the configuration path checked with PowerShell Get-Acl, not the cache and not icacls; do not approve until preview returns an approval digest.
• If you integrate agentic-awesome-skills into a host, read the discovery contract first: Stable Skills Manifest v1.
• For Windows truncation or context crash loops, use docs/users/windows-truncation-recovery.md.
• For Linux/macOS overload or selective activation, use docs/users/agent-overload-recovery.md.
• For OpenCode or other .agents/skills installs, prefer a reduced install such as npx agentic-awesome-skills --path .agents/skills --category development,backend --risk safe,none.
• For plugin install details, host compatibility, and marketplace-safe distribution, use docs/users/plugins.md.
• For contributor expectations and guardrails, use CONTRIBUTING.md, CODE_OF_CONDUCT.md, and SECURITY.md.