Guide · Agent skill
How to use the Antigravity Skill Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The antigravity-skill-orchestrator is an agent skill, a SKILL.md instruction file that an AI agent host can load, that helps the agent decide whether a request needs specialized skills at all and, if it does, which small set to use. This guide explains what it changes in the agent's behaviour, how to install it, how to trigger it, and what it needs. After reading, you should be able to install the skill, ask for work in a way that triggers it, and know when it will stay out of the way.
Antigravity Skill Orchestratorsickn33/agentic-awesome-skills · MIT Download the skill Quick start
- 1Download the ZIP. Get the skill package from this page. It contains README.md and SKILL.md.
- 2Unzip into a skills folder. Place the folder in ~/.claude/skills/ for personal use, or in your project's .claude/skills/ folder.
- 3Or upload in the Claude apps. Open Settings, then Capabilities, then Skills, and upload the skill.
- 4Decide on memory. Optionally install agent-memory-mcp if you want successful skill combinations saved and recalled.
- 5Ask for a complex, multi-part task. Describe the whole job and its domains so the agent can judge complexity and select skills.
- 6Review the result. Check what the agent reports as completed and verified, and test the work yourself.
What is an agent skill, and what does this one make the agent do?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent host reads when a task matches the skill's description. This one is a meta-skill: it does not add a specific technical capability such as a database tool. Instead it tells the agent how to choose among other skills.
Its workflow starts with a judgement call. The agent first asks whether the request can be handled with basic file editing, search and terminal commands. If so, it does the work directly and stops. If the task is complex or spans several domains, the skill has the agent look through the skills available locally, and if those fall short, fetch a remote catalog of community skills. It then selects the minimal set of skills needed. It can also record successful combinations in an optional external memory tool, agent-memory-mcp, and search them later.
Install the skill on Claude or a project
Option one: download the ZIP from the skill's page on this directory and unzip it into your personal skills folder, ~/.claude/skills/. Option two: unzip it into a project's .claude/skills/ folder so it applies only to that project.
Option three: in the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill there.
The package contains README.md and SKILL.md. The SKILL.md file is the part the agent reads.
- Personal: ~/.claude/skills/
- Project: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills → upload
Trigger the skill with natural requests
The skill's description and its 'When to Use' guidance point to requests that are large, multi-part, or where you do not know which tools fit. You do not need a special command. Plain requests that describe a multi-domain task are the best triggers. The source also lists explicit phrases such as 'orchestrate', 'combine skills', or 'use the best tools for the job' on a significant task.
Describe the whole job and its parts, because the agent identifies the domains it needs from your description.
- Ask for a multi-part build: 'Build a React Native app with Firebase auth and a sync feature; pick the skills you need.'
- Ask for help choosing: 'I have many skills installed. Which ones fit a checkout flow with Stripe and PostgreSQL?'
- Ask for a past combination: 'Have we combined skills for a problem like this before?'
See what happens with a simple task
The skill is designed to stay quiet on small work. The source's example is a request to change the colour of a submit button in index.css. The agent judges this a simple, contained task, does not invoke specialized skills, and edits the file directly.
This is the main reason to install it even if you rarely use other skills: it prevents over-engineering small edits.
Optional: connect agent-memory-mcp for saved combinations
The skill can record successful combinations and recall them. It uses the memory_search, memory_read and memory_write tools provided by agent-memory-mcp, with the type skill_combination. agent-memory-mcp is an optional external dependency and is not bundled with this skill. If it is unavailable, the agent continues selection without persistent tracking.
Recalled combinations are suggestions, not proof that they still work. The skill also says to honour your host or user memory policy before writing durable records, and to leave out private project details and secrets.
memory_search({ query: "ecommerce checkout", type: "skill_combination" })Understand what it needs and its safety boundaries
The skill needs an agent host that supports SKILL.md skills. It uses the network: when local skills are insufficient, it fetches the master catalog from raw.githubusercontent.com. If that access is blocked, the skill cannot fetch the remote catalog and will rely on local skills. It runs no scripts, according to the package details.
Remote catalog text is treated as discovery data, not as permission to install or run skills. The skill also forbids creating new skills; it only combines existing ones. Use the host's actual installed skills and tool schemas as the source of truth.
- Network: yes, only for the remote catalog at raw.githubusercontent.com
- Scripts: none listed in the package
- Memory writes: optional, governed by your host and user policy
- Licence: MIT
Tips for getting good results
Give the agent the real scope of the task so the complexity check is accurate. Keep the number of skills small, since the skill itself says to avoid over-selecting. Use descriptive keys when saving memory so combinations are easy to find later. Always check outcomes yourself: the skill asks the agent to report what actually completed and which checks ran, and it is not a substitute for testing or expert review.
- Describe the task's domains explicitly
- Review what the agent reports as verified versus unverified
- Avoid it for simple bug fixes or UI tweaks
What people use it for
- Planning a multi-domain feature such as a React UI with a Node.js backend and a PostgreSQL database.
- Choosing among a large set of installed skills when you are not sure which ones fit.
- Reusing a previously successful skill combination for a recurring type of project via memory search.
- Keeping the agent from over-engineering small edits like renaming a variable or changing a CSS colour.
Frequently asked questions
Is the antigravity skill orchestrator free?
The package is listed under the MIT licence. The source does not mention pricing, so check the licence file in the repository for the terms.
What do I need to use it?
You need an agent host that supports SKILL.md skills. Optionally, you can use agent-memory-mcp or equivalent memory tools, and network access to raw.githubusercontent.com if the remote skill catalog is needed.
How do I install the antigravity-skill-orchestrator skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings, then Capabilities, then Skills.
Does it create new skills for me?
No. The skill explicitly forbids creating new skills. It only selects and combines existing skills from the local environment or the community catalog.
Can it work without agent-memory-mcp?
Yes. If the memory tool is unavailable, the agent continues skill selection without persistent tracking. It still uses the remote catalog when local skills fall short.
What is an alternative if I only want simple help?
If your tasks are small and single-step, you do not need this skill, since it tells the agent to skip orchestration for them. Ask the agent directly, or use a general skill for the specific domain you need.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Install this skill if you work on large, multi-part projects, have many skills installed, or want the agent to avoid heavy-handed changes on small edits. Skip it for simple single-step tasks, in environments that block remote catalog fetches or memory writes, or if you expect it to create new skills, which it will not do.
Download the skill