Anomaly · install guide
OpenCode skills. Where they live and how to install them.
Where OpenCode loads skills
- In a project
.opencode/skills/.claude/skills/.agents/skills/- For all projects
~/.config/opencode/skills/~/.claude/skills/~/.agents/skills/- Docs
- opencode.ai
- Website
- opencode.ai
- OpenCode walks up from the working directory to the git worktree root, loading .opencode/skills, .claude/skills and .agents/skills along the way, then the global folders.
- Only name, description, license, compatibility and metadata (string-to-string map) are recognised; unknown frontmatter fields are ignored.
- name must be 1-64 characters, match ^[a-z0-9]+(-[a-z0-9]+)*$ and equal the containing folder name. description must be 1-1024 characters.
- Skills are loaded on demand through a native skill tool; names must be unique across all locations.
- Control access in opencode.json under permission.skill with allow, deny or ask, using wildcards such as internal-*. Set tools.skill to false on an agent to turn skills off for it.
How to install a skill
npx skills add {owner}/{repo} --skill {skill} -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
npx skills add anthropics/skills --skill mcp-builder -a opencodeExample with MCP Server Builder.
gh skill install {owner}/{repo} {skill} --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
gh skill install anthropics/skills mcp-builder --agent opencodeExample with MCP Server Builder.
git clone --depth 1 https://github.com/{owner}/{repo}.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/{path} .opencode/skills/{skill} && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
git clone --depth 1 https://github.com/anthropics/skills.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/skills/mcp-builder .opencode/skills/mcp-builder && rm -rf skills-srcExample with MCP Server Builder.
Guide
How to install agent skills in OpenCode
What skills are in OpenCode
OpenCode discovers reusable instructions from SKILL.md files in your repository or home directory. The agent sees a list of available skills, each with a name and description, inside the description of a native skill tool. When a task fits, the agent calls the tool, for example skill({ name: "git-release" }), and the full content loads then. Skills therefore stay out of the context until they are needed.
OpenCode recognises only these frontmatter fields: name and description (both required), plus optional license, compatibility and metadata (a string-to-string map). Unknown fields are ignored, so a skill written for another agent still loads, though extras such as allowed-tools have no effect here.
Where OpenCode looks for skills
| Scope | Folders |
|---|---|
| Project | .opencode/skills/<name>/SKILL.md, .claude/skills/<name>/SKILL.md, .agents/skills/<name>/SKILL.md |
| Global | ~/.config/opencode/skills/<name>/SKILL.md, ~/.claude/skills/<name>/SKILL.md, ~/.agents/skills/<name>/SKILL.md |
For project locations, OpenCode walks up from the current working directory until it reaches the git worktree root. It loads matching skill folders in .opencode/, .claude/ and .agents/ along the way. Skill names must be unique across all of these locations.
Install a skill
OpenCode's documentation describes placing files rather than any install command, so installation goes through a separate tool or a manual copy. The examples use frontend-design from anthropics/skills, located at skills/frontend-design.
Option 1: the skills CLI
npx skills add anthropics/skills --skill frontend-design -a opencode
This installs into .agents/skills/ in the current project. Add -g for a global install into ~/.config/opencode/skills/.
Option 2: GitHub CLI
gh skill install anthropics/skills frontend-design --agent opencode
gh skill install anthropics/skills frontend-design --agent opencode --scope user
Needs GitHub CLI 2.90.0 or later, where gh skill is a public preview.
Option 3: copy the folder
git clone --depth 1 https://github.com/anthropics/skills.git skills-src
mkdir -p .opencode/skills
cp -r skills-src/skills/frontend-design .opencode/skills/frontend-design
rm -rf skills-src
Use ~/.config/opencode/skills/ for a global install.
Check that it loaded
Ask OpenCode which skills it can use, or ask it to perform a task that matches the skill's description. OpenCode lists the available skills with their names and descriptions in the skill tool, so the agent can tell you what it sees. If the skill is missing, work through the checks below.
Update and remove
- Remove: delete the skill folder, or run
npx skills remove frontend-design. - Update: run
npx skills update, orgh skill updatefor skills installed withgh, or copy a newer version of the folder over the old one. - Hide a skill without deleting it: set it to
denyinopencode.json, as shown below.
Control which skills the agent can use
Permissions live in opencode.json under permission.skill. Each pattern maps to allow (loads immediately), deny (hidden from the agent) or ask (you are prompted first). Wildcards work:
{
"permission": {
"skill": {
"*": "allow",
"internal-*": "deny",
"experimental-*": "ask"
}
}
}
You can override the defaults per agent, and setting tools.skill to false for an agent turns skills off for it entirely.
Troubleshooting
- Check the filename is spelled
SKILL.mdin capitals. - Check the frontmatter has both
nameanddescription, and that the name matches the folder name exactly. Names must match^[a-z0-9]+(-[a-z0-9]+)*$and be at most 64 characters, with no leading, trailing or doubled hyphens. - The description must be 1 to 1024 characters.
- Make sure names are unique across every location. A duplicate in
.claude/skillsand.opencode/skillsis a conflict. - Check your
permission.skillrules: a skill set todenyis hidden from the agent.
Safety
A skill can instruct the agent to run commands and edit files, and the agent does that with your account's permissions. Read SKILL.md and any bundled scripts before installing. If you use third-party skills regularly, ask rules in permission.skill give you an approval step before a skill loads, which is a useful default for names you have not reviewed.
Top skills to try in OpenCode
| # | Skill | Repository | Stars | Used in | Tokens | Auto-check | Licence | Updated |
|---|---|---|---|---|---|---|---|---|
| 1 | Guides the design and implementation of Model Context Protocol servers in TypeScript or Python, from tool naming and error messages to evaluation. | anthropics/ | 180k | 62 repos | ~2.3k | Automated safety check: Pass | Apache-2.0 | yesterday |
| 2 | Handles everyday PDF jobs in Python and on the command line: extract text and tables, merge, split, rotate, watermark, fill forms, encrypt and OCR. | anthropics/ | 180k | 48 repos | ~2k | Automated safety check: Pass | Proprietary | yesterday |
| 3 | Builds multi-component claude.ai HTML artifacts as a small React, TypeScript and Tailwind project, then bundles it into one shareable HTML file. | anthropics/ | 180k | 40 repos | ~769 | Automated safety check: Pass | Apache-2.0 | yesterday |
| 4 | Makes the agent clarify intent and agree on a design with you before writing any code, scaling the process from a quick spike to a written spec. | obra/ | 296k | 11 repos | ~4.4k | Automated safety check: Pass | MIT | today |
| 5 | Walks your agent through drafting a new skill, testing it on sample prompts, reviewing the results with you and tuning its description so it triggers. | anthropics/ | 180k | 10 repos | ~8.2k | Automated safety check: Pass | Apache-2.0 | yesterday |
| 6 | Builds animated HTML slide decks that run in the browser with no dependencies, or converts PowerPoint files to the web, starting from visual style previews. | zarazhangrui/ | 30k | 16 repos | ~7k | Automated safety check: Pass | MIT | 3 mo ago |
| 7 | Runs an implementation plan by giving each task to a fresh subagent, with a review after every task and a broad review of the whole branch at the end. | obra/ | 296k | 10 repos | ~8.1k | Automated safety check: Pass | MIT | today |
| 8 | Design and build AI agents for any domain. An agent skill from shareAI-lab/learn-claude-code. | shareAI-lab/ | 78k | 6 repos | ~1.2k | Automated safety check: Pass | MIT | 8 days ago |
| 9 | Lets Claude Code ask questions of your Google NotebookLM notebooks through browser automation and return answers grounded in your uploaded sources. | PleasePrompto/ | 7.8k | 13 repos | ~2.4k | Automated safety check: Notes | MIT | 27 days ago |
| 10 | 10.React Doctor Scans React code for lint, accessibility, bundle size and architecture issues, reports a health score and checks that changes do not lower it. | makeplane/ | 60k | 12 repos | ~657 | Automated safety check: Pass | AGPL-3.0 | yesterday |
| 11 | Walks the last step of a branch: confirm tests pass, detect the git environment, ask how to integrate, carry out your choice and clean up the worktree. | obra/ | 296k | 5 repos | ~1.9k | Automated safety check: Pass | MIT | today |
| 12 | Explains how to write Claude Code plugin hooks, both prompt-based checks and bash commands, for events such as PreToolUse, Stop and SessionStart. | anthropics/ | 37k | 11 repos | ~4.1k | Automated safety check: Notes | Apache-2.0 | today |
| 13 | Produces math and technical explainer videos with Manim Community Edition: concept animations, equation derivations, algorithm walkthroughs and data stories. | browser-use/ | 28k | 6 repos | ~3k | Automated safety check: Pass | MIT | 4 days ago |
| 14 | Guides a conversation that takes a vague idea through divergent and convergent thinking and ends in a markdown one-pager covering scope and assumptions. | addyosmani/ | 102k | 6 repos | ~2k | Automated safety check: Pass | MIT | 3 days ago |
| 15 | Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking… | vercel-labs/ | 44k | 24 repos | ~864 | Automated safety check: Pass | Apache-2.0 | today |
| 16 | Requires the agent to find the root cause of a bug, test failure or build problem before proposing any fix, working through four ordered phases. | obra/ | 296k | 13 repos | ~2.4k | Automated safety check: Pass | MIT | today |
| 17 | Creates and edits JSON Canvas (.canvas) files with nodes, edges and groups, for Obsidian canvases, mind maps and flowcharts. | kepano/ | 49k | 6 repos | ~1.9k | Automated safety check: Pass | MIT | 21 days ago |
| 18 | Master TypeScript's advanced type system including generics, conditional types, mapped types, template literals, and utility types for building type-safe applications. | rolling-scopes/ | 10k | 24 repos | ~4.2k | Automated safety check: Pass | MPL-2.0 | 3 days ago |
| 19 | A skill your agent uses when starting any conversation - establishes how to find and use skills, requiring Skill tool invocation before ANY response including clarifying questions | farm-fe/ | 5.6k | 34 repos | ~1.4k | Automated safety check: Pass | MIT | 15 days ago |
| 20 | Has the agent carry out an implementation plan itself, task by task in the current session, keeping a ledger, proving each step with a test and ending with one whole-branch review. | obra/ | 296k | 2 repos | ~5.1k | Automated safety check: Pass | MIT | today |
| 21 | Explores a web app with the agent-browser CLI to find bugs and UX problems, then writes a report with screenshots, repro videos and step-by-step reproduction for each issue. | vercel-labs/ | 44k | 8 repos | ~2.7k | Automated safety check: Pass | Apache-2.0 | today |
| 22 | Creates original generative art in two steps: a written algorithmic philosophy, then a p5.js sketch with seeded randomness and an interactive viewer for exploring parameters. | anthropics/ | 180k | 37 repos | ~4.9k | Automated safety check: Pass | Apache-2.0 | yesterday |
| 23 | Perform comprehensive exploratory data analysis on scientific data files across 200+ file formats. | spacering-net/ | 3.8k | 15 repos | ~3.6k | Automated safety check: Pass | MIT | today |
| 24 | Turns a spec into a step-by-step plan of small tasks, written for an engineer who has not seen the codebase, before any code is touched. | obra/ | 296k | 12 repos | ~2.6k | Automated safety check: Pass | MIT | today |
Questions, answered from the data.
Where does OpenCode look for skills?
OpenCode loads skill folders (each with a SKILL.md file) inside a project from .opencode/skills/, .claude/skills/ and .agents/skills/ and for every project from ~/.config/opencode/skills/, ~/.claude/skills/ and ~/.agents/skills/.
How do I install a skill in OpenCode?
skills CLI: npx skills add {owner}/{repo} --skill {skill} -a opencode. GitHub CLI: gh skill install {owner}/{repo} {skill} --agent opencode. Manual copy: git clone --depth 1 https://github.com/{owner}/{repo}.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/{path} .opencode/skills/{skill} && rm -rf skills-src. Replace the placeholders with the skill's GitHub owner, repository and name; every skill page here shows the filled-in command.
What are the best skills for OpenCode?
Agent skills are portable SKILL.md folders, so the catalogue's top skills work in OpenCode too. The highest-ranked right now are MCP Server Builder (anthropics/skills), PDF Processing (anthropics/skills) and Web Artifacts Builder (anthropics/skills).
Sources
- https://opencode.ai/docs/skills/
- https://github.com/vercel-labs/skills
- https://cli.github.com/manual/gh_skill_install
Checked October 7, 2026