| name | find-skills |
| description | Search the agent skills ecosystem to discover and install skills that extend AI coding agent capabilities. Use when user asks "how do I do X" (X being a common task), "find a skill for X", "is there a skill that can...", or expresses interest in extending agent capabilities. Triggers on "find a skill", "install a skill", "skill for [task]", "can you do X", "I need help with [domain]", "how do I [task]". Do NOT use when user has explicitly asked to proceed without a skill, or when the task is better handled by agent's built-in capabilities (file operations, git, basic coding). |
| triggers | ["find a skill","install a skill","skill for","is there a skill","how do I","search skills","discover skills","skill ecosystem","available skills","list skills"] |
| negatives | ["create a skill","write a skill","edit a skill","skill development"] |
| license | MIT |
| compatibility | opencode |
| metadata | {"workflow":"tooling","audience":"developers","version":"1.0.0"} |
Help users discover and install skills from the open agent skills ecosystem.
What is the Skills CLI?
The Skills CLI (npx skills) is the package manager for agent skills:
npx skills find [query] - Search for skills
npx skills add <package> - Install from GitHub
npx skills check - Check for updates
npx skills init [name] - Scaffold a new skill
Browse at: https://skills.sh/
Workflow
Step 1: Understand what they need
Identify: domain (React, testing, DevOps) → specific task (writing tests, creating animations) → likelihood a skill exists.
Step 2: Check the leaderboard first
Before running CLI: check https://skills.sh/ for well-known skills.
Top sources:
vercel-labs/agent-skills — React, Next.js, web design (100K+ installs)
anthropics/skills — Frontend design, document processing (100K+ installs)
zencoderai/skills — OSS security, git gate (50K+ installs)
Step 3: Search
npx skills find [query]
Examples:
- "how do I make my React app faster" →
npx skills find react performance
- "help with PR reviews" →
npx skills find pr review
- "create a changelog" →
npx skills find changelog
Step 4: Verify quality
- Install count: Prefer 1K+. Be cautious under 100.
- Source reputation: Official sources (
vercel-labs, anthropics, microsoft) over unknown authors.
- GitHub stars: Source repo < 100 stars → treat with skepticism.
Step 5: Present options
Include: skill name, what it does, install count, source, install command.
I found a skill that might help! "react-best-practices" provides
React/Next.js performance optimization guidelines from Vercel Engineering.
(185K installs)
To install:
npx skills add vercel-labs/agent-skills@react-best-practices
Learn more: https://skills.sh/vercel-labs/agent-skills/react-best-practices
Step 6: Install (if user agrees)
npx skills add <owner/repo@skill> -g -y
-g installs globally, -y skips confirmation.
Common Categories
| Category | Example Queries |
|---|
| Web Development | react, nextjs, typescript, css, tailwind |
| Testing | testing, jest, playwright, e2e |
| DevOps | deploy, docker, kubernetes, ci-cd |
| Documentation | docs, readme, changelog, api-docs |
| Code Quality | review, lint, refactor, best-practices |
| Design | ui, ux, design-system, accessibility |
| Productivity | workflow, automation, git |
Skill Quality Assessment
When presenting a skill, assess these dimensions:
| Dimension | What to check | Red flags |
|---|
| Maintenance | Last update < 6 months? | No updates in 1+ year |
| Depth | Has examples, not just description? | Single paragraph, no code |
| Compatibility | Supports your agent? | Only tested on one platform |
| Specificity | Solves your exact problem? | Generic content without actionable patterns |
Fallback: When No Skills Found
Acknowledge, offer direct help, suggest creating a custom skill:
I searched for skills related to "xyz" but didn't find matches.
I can still help with this task directly.
If this is something you do often, create your own skill:
npx skills init my-xyz-skill
Error Handling
| Cause | Fix |
|---|
npx skills find returns zero results for a valid query | Query may be too specific or use jargon not in skill descriptions. Broaden to the domain term (e.g., "testing" instead of "jest snapshot serializers"). Try 2-3 synonym variations. |
npx skills add fails with "package not found" | Verify the package spec format: <owner>/<repo>@<skill-name>. Check https://skills.sh/ for the exact package path. GitHub rate limits may apply — retry after 60s. |
| Installed skill does not activate (agent ignores it) | Skill description might not match the user's phrasing. Check the skill's description field in its SKILL.md frontmatter. It should use trigger phrases the user naturally says. |
npx skills check reports updates but add re-installs same version | The CLI caches resolution metadata. Run npx skills check --refresh to force re-resolution. If still stuck, uninstall with npx skills remove <package> and re-add. |
| Skills website (skills.sh) is unreachable | GitHub Pages may be down or DNS blocked. Fall back to direct GitHub search: gh search repos --topic agent-skill <query>. Present results from the CLI search as primary alternative. |
| User requests a skill for a niche domain with no ecosystem coverage | No skill exists yet. Offer to help directly. Propose creating a custom skill with npx skills init. Show the skill-creator workflow if the user wants to build and publish it. |
| Skill install conflicts with an existing skill of the same name | Two skills share a namespace. The CLI will refuse to overwrite. Uninstall the old one first: npx skills remove <old-package>. Verify the old skill isn't needed before removing. |
Skill Discovery Methodology
Finding skills in the ecosystem
npx skills list
npx skills find "docker" "kubernetes" "security"
npx skills check
Quality Evaluation Framework
Rate skills on 5 criteria before installing:
- Trigger clarity: Does the description clearly define when to activate?
- Workflow completeness: Numbered steps or procedural sections?
- Error handling: Does it document common failures and fixes?
- Sources cited: Verifiable references or just opinions?
- Anti-patterns: Explicit warnings about what NOT to do?
Minimum threshold for production: 4 of 5 criteria met.
Installation Troubleshooting
| Error | Cause | Fix |
|---|
SKILL.md not found | Directory missing from registry | Check if skill exists at source repository |
permission denied | Plugin directory not writable | chmod 755 ~/.config/opencode/skills/ |
skill already installed | Version conflict or duplicate | npx skills remove <name> then reinstall |
trigger not matching | Description too generic or too narrow | Edit frontmatter description: field and re-test |
npm ERR! during install | npm registry issue or network | Retry. If persistent, clone manually from GitHub |
Checklist
Sources
- Skills CLI documentation (skills.sh/docs) — package manager usage, publish flow, and repository structure
- Vercel Agent Skills repository (github.com/vercel-labs/agent-skills) — reference skill implementations with 100K+ installs
- Anthropic Skills specification (github.com/anthropics/skills) — skill format, metadata schema, and compatibility guidelines
- Zencoder Agent Skills repository (github.com/zencoderai/skills) — OSS security and workflow skills reference
- npm package registry documentation (docs.npmjs.com) — npx resolution, caching, and version management
- GitHub Search API documentation (docs.github.com/en/rest/search) — topic-based repository discovery as fallback
- "The Design of Everyday Things" by Don Norman (Basic Books, 2013) — discoverability principles applied to skill ecosystem UX
Anti-Patterns
| Pattern | Problem | Fix |
|---|
| Recommending a skill without checking install count or maintenance status | Abandoned skills with 0 installs waste the user's time and may contain broken or outdated instructions. | Check install count (>1K preferred), last update (<6 months), and source reputation before presenting. |
| Installing a skill the user didn't explicitly agree to install | Skills modify agent behavior. Installing without consent breaks trust. | Always present the skill first, wait for explicit confirmation, then install. Never auto-install. |
| Searching only once and giving up when zero results appear | First query may use the wrong terminology or be too narrow. | Try 2-3 synonym variations and broader domain terms before concluding no skill exists. |
| Suggesting a skill when the task is trivial and built-in capabilities suffice | Overhead of installing and loading a skill for a 30-second task wastes tokens and attention. | Only suggest skills for recurring patterns, complex domains, or tasks the user explicitly flagged as needing a skill. |
| Evaluating skill quality by description alone without reading the actual content | Descriptions are marketing. The SKILL.md content determines actual usefulness. | After identifying candidates, read the skill's SKILL.md on GitHub before presenting. Check for concrete patterns, code examples, and workflow steps. |
| Assuming a skill works on all platforms when it's only tested on one | A skill written for Claude Code may fail on OpenCode or Gemini CLI due to tool name differences. | Check the skill's compatibility metadata field. If missing, test the first step manually before recommending. |