Graviton Agent Skills

Portable Agent Skills that help developers migrate codebases to AWS Graviton (ARM64) processors.

What are Agent Skills?

Agent Skills are an open standard for packaging expert instructions in a format AI coding assistants can follow. Each skill is a folder containing instructions, references, and scripts that teach an agent how to perform a specific task — in this case, migrating applications to Graviton.

For the full specification, see agentskills.io.

Platform Compatibility

Each skill folder contains two entry point files to support the broadest range of AI tools:

Both files contain the same instructions — only the frontmatter differs. Your platform will pick up the file it recognizes and ignore the other.

Available Skills

How to Install

Each skill installs on its own. The commands below install the Java skill; set SKILL to another folder name from Available Skills (for example python-x86-to-graviton) to install that skill instead. Run the project-level commands (GitHub Copilot, Windsurf, Roo Code, Any platform) from your project’s root directory.

Kiro

Open the Agent Steering & Skills panel, click + > Import a skill > GitHub, and paste the URL of the skill you want:

https://github.com/aws/aws-graviton-getting-started/tree/main/tools/skills/languages/java-x86-to-graviton
https://github.com/aws/aws-graviton-getting-started/tree/main/tools/skills/languages/python-x86-to-graviton

Cursor

Open Settings (Cmd+Shift+J / Ctrl+Shift+J) > Rules > Add Rule > Remote Rule (GitHub) and paste the URL of the skill you want:

https://github.com/aws/aws-graviton-getting-started/tree/main/tools/skills/languages/java-x86-to-graviton
https://github.com/aws/aws-graviton-getting-started/tree/main/tools/skills/languages/python-x86-to-graviton

Claude Code

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p ~/.claude/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" ~/.claude/skills/ && \
  rm -rf /tmp/graviton-skill

GitHub Copilot / VS Code

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p .github/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" .github/skills/ && \
  rm -rf /tmp/graviton-skill

OpenAI Codex

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p ~/.codex/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" ~/.codex/skills/ && \
  rm -rf /tmp/graviton-skill

Windsurf

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p .windsurf/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" .windsurf/skills/ && \
  rm -rf /tmp/graviton-skill

Gemini CLI

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p ~/.gemini/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" ~/.gemini/skills/ && \
  rm -rf /tmp/graviton-skill

Roo Code

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p .roo/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" .roo/skills/ && \
  rm -rf /tmp/graviton-skill

Goose

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p ~/.config/goose/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" ~/.config/goose/skills/ && \
  rm -rf /tmp/graviton-skill

Any platform (project-level)

Copy the skill folder into the cross-platform standard path:

SKILL=java-x86-to-graviton   # or python-x86-to-graviton
git clone --filter=blob:none --sparse https://github.com/aws/aws-graviton-getting-started.git /tmp/graviton-skill && \
  git -C /tmp/graviton-skill sparse-checkout set "tools/skills/languages/$SKILL" && \
  mkdir -p .agents/skills && \
  cp -r "/tmp/graviton-skill/tools/skills/languages/$SKILL" .agents/skills/ && \
  rm -rf /tmp/graviton-skill

The .agents/skills/ path is supported by all Agent Skills-compatible platforms.

Skill Folder Structure

Each skill folder contains:

<skill-name>/
├── SKILL.md                    # Entry point (Agent Skills format)
├── POWER.md                    # Entry point (Kiro format)
├── summaries.md                # Canonical file index for agent discovery
├── phases/                     # Detailed phase instructions
├── document_references/        # Scope guardrails and output standards
└── README.md                   # Human-readable overview

Creating New Skills

See the skill template for the boilerplate structure. When creating a new skill:

1. Create a folder under languages/ named to match the skill’s name field
2. Create both SKILL.md (Agent Skills frontmatter) and POWER.md (Kiro frontmatter) with the same body content
3. Add a summaries.md listing every file in the skill, each with its purpose and when it is read. This is the canonical inventory — link it from SKILL.md/POWER.md, and do not duplicate the file list elsewhere in the skill
4. Add a README.md with a human-readable overview that points at summaries.md for the file map
5. Keep the main entry point under 500 lines — split detailed instructions into phases/, references/, or scripts/ subdirectories
6. Link each supporting file inline from the step that needs it, so an agent reading top to bottom reaches it just in time. summaries.md serves agents that enter mid-skill or need the whole map at once

Fill in language-specific details based on the existing Graviton documentation for each language.