- Shell 100%
|
|
||
|---|---|---|
| skills | ||
| .gitignore | ||
| AGENTS.md | ||
| IDEAS.md | ||
| LICENSE | ||
| README.md | ||
| SHARED.md | ||
agent-skills
This repository contains shared, portable skills and agent instructions for use across projects. This project is meant to be cloned directly into ~/.agents. All instructions assume this location. If you cloned this repo to another directory, please modify the discoverability steps accordingly.
- Skills are defined in
skills/and follow the Agent Skills specification. - Shared instructions are stored in SHARED.md.
~/.agents/ # This repository
├── README.md
├── IDEAS.md # Skill ideas and related workflow decisions
├── SHARED.md # Cross-project instructions
├── AGENTS.md # Instructions for maintaining this repo
└── skills/ # Home-level, shared skills
└── my-skill/
└── SKILL.md
See Skill compatibility and Shared instructions for the setup needed to make this content discoverable by the tools you use.
What is a skill?
A skill is a reusable package of instructions for a specific task, such as reviewing code or creating a commit. In a sense, it's a way to treat prompts as code. Skills can invoke other skills. This allows you to specify instructions for a repeatable task or workflow once and use them many times or from other related skills. Skills go beyond the capability of a normal prompt: they can bundle executable helpers and reference material alongside agent instructions.
Each skill is a directory with one required file:
my-skill/
├── SKILL.md # Markdown instructions and YAML metadata (required)
├── scripts/ # executable helpers (optional)
├── references/ # detailed documentation (optional)
└── assets/ # templates, images, schemas, etc. (optional)
Skills can be explicitly invoked or loaded automatically by an agent. For automatic selection, the agent first sees each skill's name and description, then loads the full instructions when a task matches. Supporting resources are read as needed. This progressive disclosure keeps unrelated material out of the conversation's context.
The recommended way to create new skills is with this repository's create-skill Agent Skill. Describe your goals and the workflow you have in mind. The agent will work with you through conversation to develop a shared understanding, then create the required files and metadata. Alternatively, you can create your own skills manually by following the Agent Skills specification.
Skill compatibility
The support column indicates whether skills in ~/.agents/skills/ are discovered directly or need a symlink. All listed tools can automatically select a relevant skill without explicit invocation. How to invoke lists the options for explicitly invoking a skill in each product.
| Tool | Support | How to invoke |
|---|---|---|
| Zed (native agent) | ✔ | /skill-name, @skill |
| Cursor | ✔ | /skill-name |
| Codex CLI / IDE extension | ✔ | $skill-name, select via /skills |
| Claude Code | * ✔ | /skill-name |
| Gemini CLI | ✔ | Ask by skill name |
| GitHub Copilot CLI / VS Code | ✔ | /skill-name |
| OpenCode | ✔ | Ask by skill name |
*Claude Code requires skill directories to be symlinked into
~/.claude/skills/.
Shared instructions
SHARED.md contains always-on instructions that apply across projects and tasks. Use it for persistent preferences, constraints, and requirements that agents should follow regardless of which skills they select. Tools do not universally discover SHARED.md. Symlink, copy, or merge contents into the destination below. Always check for existing content in the destination in case it needs preserved.
| Tool | Destination path or setting |
|---|---|
| Zed (native agent) | ~/.config/zed/AGENTS.md (macOS/Linux)%APPDATA%\Zed\AGENTS.md (Windows) |
| Cursor | Customize → Rules → User Rules |
| Codex CLI / IDE extension | ~/.codex/AGENTS.md |
| Claude Code | ~/.claude/CLAUDE.md |
| Gemini CLI | ~/.gemini/GEMINI.md |
| GitHub Copilot CLI | ~/.copilot/copilot-instructions.md |
| GitHub Copilot in VS Code | Copilot harness: ~/.copilot/copilot-instructions.mdLocal harness: profile instructions/settings |
| OpenCode | ~/.config/opencode/AGENTS.md |
Paths and loading behavior assume the default configuration of each tool. Custom settings and overrides may change the steps needed for your exact setup. After setup, start a fresh session and ask the agent to identify its loaded instructions and available skills.
Available skills
Focused skills handle independently useful tasks, while workflows coordinate those skills across stages. The skills listed here are complete and available in this repository. IDEAS.md tracks proposed and in-progress skills.
Focused skills
| Skill | Purpose |
|---|---|
| polish | Review and refine prose for clarity, organization, and concision while preserving meaning, depth, and voice. |
| converse | Explore one key idea at a time in an interactive conversation while preserving reasoning, nuance, and complexity. |
| commit | Create or amend Git commits using git jarvis commit for signing, authorship, and model attribution. |
Workflows
| Skill | Purpose |
|---|---|
| create-skill | Create, revise, or review Agent Skills to define repeatable workflows or reusable agent instructions. Collaborate with the user to define or refine the scenarios and behavior a skill should cover, then author, improve, or review the skill based on that shared understanding. |
Additional planned workflows are tracked in IDEAS.md.