10 Best OpenCode Skills That Are Actually Useful in 2026

Skills are perhaps the simplest yet most effective layer when it comes to usefulness. We’ve had MCPs, plugins, and a dozen other markdown-file-driven approaches, but in terms of pure ROI, Skills are the best.

They work because they meet the model where it already is. Instead of forcing a new interface, a new protocol, or a heavy abstraction, Skills gives the model a compact, reusable set of instructions it can pull from at the right moment. That sounds almost too basic, but that is exactly why it works.

The underrated part is that Skills are also easy to reason about. A team can open one, read it, edit it, and immediately understand what behaviour it is supposed to produce. That makes them much easier to maintain than systems that hide logic behind APIs, chains, or orchestration layers.

I switch between coding harnesses, it’s Claude Code, and for personal stuff, it’s always OpenCode.

And over time, I’ve ended up using a bunch of Skills inside OpenCode for everything from routine coding tasks to oddly specific workflows. The Skills for ClaudeCode are the same as OpenCode; these are just markdown files anyway.

So I collated some of the best OpenCode Skills I’ve found. Some are genuinely useful, some are surprisingly clever, and a few are just plain esoteric (you can ignore them).

Also, to pair with some of their skills, I use Composio CLI. A very minimalist CLI that lets your agent authenticate with external apps like Figma, Linear, etc., for a seamless workflow.

Summary

Vercel React Best Practices: React and Next.js performance rules from Vercel Engineering.
Composio Skills: Provide OpenCode with a single interface for SaaS tools, auth, CLI actions, and MCP.
Anthropic Skill Creator: Create, test, benchmark, and improve your own SKILL.md packages.
Anthropic Frontend Design: Build a less generic frontend UI with a stronger visual direction.
Anthropic MCP Builder: Plan, implement, test, and evaluate MCP servers in TypeScript or Python.
Cloudflare Skills: Use Workers, D1, R2, Vectorise, Workers AI, Wrangler, and Agents SDK.
Replicate Skills: Find, compare, run, and prompt AI models on Replicate.
Hugging Face hf-cli: Use the modern Hugging Face Hub CLI for models, datasets, Spaces, and jobs.
Anthropic Webapp Testing: Test local web apps with Playwright scripts and server lifecycle helpers.

What is an OpenCode skill?

An Agent Skill is a folder with a SKILL.md file that OpenCode or any agent can discover and load through its native skill tool. The file uses YAML front matter for fields like name and description, and Markdown instructions that tell the agent how to perform a repeatable task.

OpenCode looks for skills in project and global locations, including .opencode/skills, ~/.config/opencode/skills, .claude/skills, and .agents/skills. The agent initially sees the skill names and descriptions, then loads full instructions only when a task matches or you ask for a skill directly.

Best OpenCode skills

1. Obra Superpowers: A complete agentic software development methodology with composable skills

Superpowers turn OpenCode from a fast code generator into a disciplined engineer. It depends on a bootstrap plugin that injects a "1% rule" into every session — if there's even a slim chance a skill applies, the agent must invoke it. It stands out because it ships a full methodology (brainstorm → plan → TDD → subagent execution → review) rather than a single narrow skill, and the skills auto-trigger from the moment you start a session.

Activates brainstorming before any creative work to extract a real spec
Enforces red/green/refactor TDD with the test-driven-development skill
Breaks features into 2–5 minute tasks via writing-plans
Runs subagent-driven-development with two-stage code review
Includes systematic-debugging, using-git-worktrees, and verification-before-completion
Skills auto-update on session start via a separate superpowers-skills repo

Install the plugin in OpenCode

Add to your opencode.json:

{ "plugin": ["superpowers@git+https://github.com/obra/superpowers.git"] }

Restart OpenCode and verify

opencode
# prompt: Tell me about your superpowers

invoke a skill directly

opencode
# prompt: Use the brainstorming skill to help me design a new auth flow.`

Resources: repository · OpenCode install guide · skills repository · marketplace

2. Vercel React Best Practices: React and Next.js performance rules from Vercel Engineering

This skill gives OpenCode a concrete rulebook for performance work in React and Next.js. It depends on the Agent Skills format and Vercel’s compiled rule files. It stands out because the rules are prioritised by impact, not grouped as a loose style guide.

Flags fetch waterfalls and await
Guides bundle-size refactors
Checks RSC and SSR serialisation patterns
Covers memoisation and re-render rules
Points to per-rule files with examples

Install the skill

npx skills add https://github.com/vercel-labs/agent-skills --skill react-best-practices

Restart OpenCode and invoke it

opencode
# prompt: Use the vercel-react-best-practices skill to review this Next.js page.

Resources: repository · skills directory

3. Composio Skills: Use SaaS tools, auth, CLI actions, and MCP from OpenCode

If I had to pick one OpenCode skill to install first, this is it. Not because I work adjacent to it, but because every other OpenCode skill on this list does one thing. Composio is the connective tissue between them.

OpenCode excels at editing code and following repo-specific instructions. What it doesn't do natively is talk to the 1000+ SaaS apps you actually use via MCP and CLI: GitHub issues, Linear tickets, Slack threads, Stripe customers, Salesforce records, Jira sprints, and Notion pages. Composio gives your agent a single interface to all of them. No separate auth flows, no token juggling, no rewriting integrations every time an API changes.

Installs a composio SKILL.md for OpenCode-compatible agents
Uses composio search, link, execute, and listen CLI commands
Supports SDK setup with composio init
Documents CLI and agent-building workflows
Covers MCP, Tool Router, triggers, and automation tags

Install the Composio skill

npx skills add composiohq/skills

Install and authenticate the Composio CLI

curl -fsSL https://composio.dev/install | bash
composio login
composio whoami

Initialise Composio in an app project

composio init

Resources: repository · SKILL.md · documentation

4. Vault Daydream: Mine your Obsidian vault for non-obvious connections between notes (esoteric pick)

This one's strange in the best way. Vault Daydream is a multi-agent skill that mimics the brain's default mode network, and a random connection between two unrelated memories suddenly surfaces. It's based on Gwern's "LLM Daydreaming" essay, which proposed that LLMs don't have breakthrough insights partly because they lack background processing. This skill is one of the few working implementations of that idea, where you can actually point at your own notes.

Sample random pairs of notes from your Obsidian vault (last 120 days by default)
Dispatches Sonnet subagents as the generator to find non-obvious links between pairs
Dispatches Haiku subagents as the critic, scoring on novelty, coherence, and usefulness
Filters insights below a 7.0/10 average score before surfacing them
Writes accepted daydreams back into your vault as new notes with backlinks
Costs roughly $0.40–0.50 per run (around 50 pairs) on Claude Code

Install the skill

git clone https://github.com/glebis/claude-skills.git
cp -r claude-skills/daydream ~/.claude/skills/
# Then edit instructions.md to point VAULT_ROOT at your Obsidian vault

Restart your agent and invoke it

claude
# prompt: /daydream

Resources: repository · Gwern's original essay

5. Anthropic Skill Creator: Create, test, benchmark, and improve SKILL.md files

Skill Creator is Anthropic’s own skill for making and improving other skills. It depends on test prompts, eval metadata, benchmark aggregation scripts, and a review viewer. It is most useful when you want OpenCode behaviour to become a reusable package instead of a one-off prompt.

Drafts SKILL.md frontmatter and instructions
Creates realistic trigger and task evaluations
Runs with skill and baseline comparisons
Aggregates pass rate, timing, and token data
Optimises skill descriptions for triggering

Install the skill

npx skills add https://github.com/anthropics/skills --skill skill-creator

Restart OpenCode and invoke it

opencode
# prompt: Use the skill-creator skill to turn our release workflow into a reusable skill.

Resources: repository · SKILL.md

6. Anthropic Frontend Design: Build production-grade UI with a stronger visual point of view

Frontend Design gives OpenCode a design brief before it writes UI code. Design skills ensures the AI agents adheres to specific guidelines and do not default to their purple gradient slop gen. It depends on the agent choosing an aesthetic direction and then implementing working HTML, CSS, JavaScript, React, or Vue. It is useful when your default agent output keeps producing generic dashboards and landing pages.

Forces a named aesthetic direction before coding
Covers typography, colour, motion, and layout
Discourages generic AI UI patterns
Supports React, Vue, HTML, CSS, and JS
Maps implementation complexity to design intent

Install the skill

npx skills add https://github.com/anthropics/skills --skill frontend-design

Restart OpenCode and invoke it

opencode
# prompt: Use frontend-design to redesign this onboarding page.

Resources: repository · SKILL.md

7. Anthropic MCP Builder: Build MCP servers with protocol, SDK, and evaluation guidance

MCP Builder guides OpenCode through the full lifecycle of a Model Context Protocol server. It depends on the MCP protocol docs, plus the TypeScript SDK or the Python FastMCP references. The differentiator is that it treats eval creation as part of the server build rather than an afterthought. MCP with OpenCode is a genuinely powerful combination, giving your agent access to gated external data sources.

The Plugin

Balances API coverage with workflow tools
Recommends TypeScript and streamable HTTP defaults
Uses Zod or Pydantic schemas for tools
Adds MCP Inspector testing steps
Defines 10 read-only evaluation questions

Install the skill

npx skills add https://github.com/anthropics/skills --skill mcp-builder

Restart OpenCode and invoke it

opencode
# prompt: Use mcp-builder to create a TypeScript MCP server for our internal API.

Resources: repository · SKILL.md

8. Cloudflare Skills: Build Workers, storage, AI, networking, security, and IaC on Cloudflare

Cloudflare’s platform skill gives OpenCode a product map for Cloudflare development. It depends on the Cloudflare docs, as limits, config, pricing, and APIs change. It is useful when a task spans Workers, Pages, KV, D1, R2, Vectorize, Workers AI, Wrangler, and deployment choices.

Routes tasks through Cloudflare product decision trees
Covers Workers, Pages, D1, R2, KV, and Durable Objects
Includes Workers AI, Vectorize, Agents SDK, and AI Gateway
Points agents to current docs and changelogs
Pairs with Cloudflare remote MCP servers

Install Cloudflare skills

npx skills add https://github.com/cloudflare/skills

Restart OpenCode and invoke it

opencode
# prompt: Use the cloudflare skill to design a Workers + D1 architecture for this API.

Resources: repository · SKILL.md

9. stop-slop: Strip AI tells out of docs, READMEs, and release notes (bonus pick)

This one's a bonus. It's not a coding skill at all. stop-slop gives OpenCode a rulebook for cleaning up the obvious AI-written tone in prose, depending on a SKILL.md plus three reference files (phrases, structures, examples) that load on demand. It's worth installing if you use OpenCode for docs and writing as much as code.

Removes filler phrases, throat-clearing openers, and em-dash reveals
Breaks formulaic structures like binary contrasts and dramatic fragmentation
Enforces active voice with a real human subject in every sentence
Includes a 1–10 scoring rubric across five dimensions with a 35/50 revision threshold
Pairs naturally with READMEs, blog posts, release notes, and commit messages

install the skill

bash

mkdir -p ~/.agents/skills
git clone https://github.com/hardikpandya/stop-slop.git ~/.agents/skills/stop-slop

Restart OpenCode and invoke it

opencode
# prompt: Use stop-slop to clean up the AI tone in this README draft.

Resources: repository · SKILL.md

10. Anthropic Webapp Testing: Test local web apps with Playwright and server helpers

Webapp Testing gives OpenCode a repeatable browser-testing workflow for local apps. It depends on Python Playwright scripts and a bundled server lifecycle helper. It stands out because it tells the agent to inspect the rendered page first, then act with selectors it discovered.

Runs native Python Playwright scripts
Starts one or more local servers with with_server.py
Waits for network idle before DOM inspection
Captures screenshots and browser logs
Uses a reconnaissance-then-action workflow

Install the skill

npx skills add https://github.com/anthropics/skills --skill webapp-testing

Restart OpenCode and invoke it

opencode
# prompt: Use webapp-testing to verify the checkout form at http://localhost:5173.

Rsources: repository · SKILL.md

End Note

Install the narrow skills that match your repo, then use Composio as the integration layer when OpenCode needs to leave the filesystem and act across GitHub, Slack, Linear, Jira, Notion, Salesforce, or other SaaS tools.

Try Composio →https://dashboard.composio.dev

End note: Treat skills like small, versioned playbooks. Install only what you will use, and keep them updated as your workflows evolve.