A curated collection of practical skills for OpenCode and agentic coding workflows.
OpenCode skills are reusable instruction bundles that teach an AI coding agent how to do a repeatable task: write tests first, package repo context, run browser QA, design frontends, automate SaaS tools, build Remotion videos, or operate Blender workflows.
Each skill lives in its own folder with a SKILL.md file containing metadata and execution guidance. OpenCode reads the metadata up front, then loads the full instructions only when the task matches the skill description.
The composio-cli skill lets OpenCode perform real actions: send emails, create issues, post to Slack, update Notion, triage Linear, and automate workflows across 1000+ apps through the Composio CLI.
git clone https://github.com/composio-community/opencode-skills.git
cd opencode-skills
mkdir -p .opencode/skills
cp -r skills/composio-cli .opencode/skills/For global install, copy it to ~/.config/opencode/skills/ instead:
mkdir -p ~/.config/opencode/skills
cp -r skills/composio-cli ~/.config/opencode/skills/curl -fsSL https://composio.dev/install | bash
composio login
composio whoamiGet a free account at dashboard.composio.dev if you do not already have one.
Restart OpenCode so it reloads skill metadata, then ask:
Use composio-cli to create a GitHub issue from this failing test output.
The skill will guide OpenCode through the safe Composio loop: discover the tool with composio search, inspect inputs with --get-schema, connect the account with composio link, preview writes with --dry-run, then execute.
For other workflows, copy any folder from skills/ into the same OpenCode skills directory.
- What Are OpenCode Skills?
- Skills
- Using Skills
- Creating Skills
- Related Awesome Lists
- Contributing
- License
Skills are modular workflow packages for AI agents. They are not MCP servers and they are not tools. MCP provides access to external systems, tools perform individual actions, and skills define the procedure, guardrails, quality bar, and decision logic the agent should follow.
OpenCode discovers skills from skill directories and loads them progressively:
- Metadata is always available through
nameanddescriptionfrontmatter. - Full instructions are loaded only when a skill is relevant.
- Optional scripts, references, or assets are used only when the skill asks for them.
The description is the trigger. Good descriptions say what the skill does and when the agent should use it.
- frontend-design-systems - Build production-quality React, Tailwind CSS, and shadcn/ui interfaces with strong spacing, alignment, readability, and non-generic visual design.
- playwright-mcp-workflows - Automate browsers, test applications, validate UI behavior, and run reliable end-to-end workflows through Playwright MCP.
- repo-context-packaging - Package a codebase into a compact, structured context bundle for another AI agent or long-context review.
- tdd-workflow - Drive implementation with the red-green-refactor loop when behavior can be specified as tests.
- vercel-react-best-practices - Build production React apps with Vercel-style readability, performance, maintainability, and stable UI behavior.
- claude-subconscious-workflows - Run background reasoning, reflective analysis, hidden planning layers, and self-improvement passes for complex tasks.
- context-engineering - Structure, compress, and orchestrate context for reliable Claude and AI agent workflows.
- idea-to-plan - Turn vague ideas or underspecified requests into written specs and implementation plans before coding.
- memory-management - Build lightweight memory systems for continuity across sessions, projects, and long-running work.
- multi-agent-orchestration - Split large tasks across parallel sub-agents with clear roles and a merge step.
- structured-project-execution - Execute multi-step projects with durable progress, decision, and plan tracking across handoffs.
- azure-messaging-workflows - Build scalable Azure messaging systems with queues, service buses, pub/sub, and event-driven patterns.
- azure-observability-workflows - Implement Azure observability with logging, monitoring, tracing, metrics, dashboards, and diagnostics.
- composio-cli - Operate external SaaS apps from the terminal via Composio CLI, including discovery, auth, tool schemas, execution, triggers, and cross-app automation.
- claude-seo-workflows - Run SEO research, content optimization, keyword strategy, technical SEO analysis, and search-focused content generation.
- marketing-skills-workflows - Perform marketing strategy, SEO audits, audience positioning, funnel analysis, and growth execution.
- powerpoint-generation - Generate professional PowerPoint presentations from rough notes, outlines, documents, or ideas.
- blender-geometry-nodes - Create procedural Blender systems with Geometry Nodes, procedural materials, and non-destructive generation.
- blender-hard-surface-modeling - Model precise hard-surface assets in Blender with clean topology, subdivision workflows, and bevel systems.
- blender-python-integration - Generate Blender Python scripts for procedural scenes, animations, assets, and workflow automation.
- blender-rigging-animation - Set up Blender, rig a character, and create a simple rendered animation.
- remotion-animation-rules - Build technically correct Remotion animations with frame-based rendering, spring physics, and native animation patterns.
- remotion-creative-storytelling - Develop concepts, hooks, story structure, and audience-focused narratives for Remotion videos.
- remotion-image-video-masking - Create layered Remotion masking effects with image/video masks, text occlusion, and frame-based animation.
- remotion-layout-debugging - Diagnose and fix layout instability, text jumping, resizing issues, and render inconsistency in Remotion.
- skill-doc-authoring - Write a well-formed
SKILL.mdfrom a workflow, SOP, or repeated process with reliable trigger metadata. - skill-improvement-opportunity-logger - Observe work sessions, identify repeated friction, log improvement opportunities, and refine workflows over time.
OpenCode can load skills from project or global locations.
Project-scoped skills:
.opencode/skills/<skill-name>/SKILL.md
.claude/skills/<skill-name>/SKILL.md
.agents/skills/<skill-name>/SKILL.md
Global skills:
~/.config/opencode/skills/<skill-name>/SKILL.md
~/.claude/skills/<skill-name>/SKILL.md
~/.agents/skills/<skill-name>/SKILL.md
After installing, restart OpenCode so it reloads skill metadata. To verify a skill, inspect its frontmatter and make sure name and description are present.
Skill layout:
skill-name/
|-- SKILL.md
|-- scripts/
|-- references/
`-- assets/
Basic SKILL.md template:
---
name: skill-name
description: What this skill does. Use when the user asks for specific trigger cases.
---
# Skill Name
## When to use
## Procedure
## Quality bar
## Anti-patternsBest practices:
- Keep the directory name and
name:lowercase with hyphen separators. - Make
descriptionspecific enough for the agent to trigger the skill correctly. - Keep the main body focused on execution steps and guardrails.
- Put long reference material in
references/and deterministic helpers inscripts/. - Use skill-doc-authoring when turning a workflow into a skill.
- ComposioHQ/awesome-claude-skills - Curated Claude Skills, plugins, and resources.
- ComposioHQ/awesome-codex-skills - Practical Codex skills for CLI and API workflows.
- anthropics/skills - Official Anthropic skills and examples.
- VoltAgent/awesome-agent-skills - Cross-agent skill collections for Claude Code, Codex, Gemini CLI, Cursor, and more.
- obra/superpowers - Agent skill methodology and composable software-development workflows.
PRs welcome. Add real, reusable skills with precise metadata, clear execution guidance, and no generic filler. If you add a skill, update the skill count badge and place it in the most specific category above.
MIT