A public, experimental collection of agent skills and role agents for AI coding tools — Claude Code, Cursor, Codex, Copilot, Aider, or any LLM you paste context into. General-purpose roles alongside specialized material in areas I work in daily: frontend (Angular, NgRx) and Web3 (wallet integrations, smart contract interaction, payments).
Work in progress. Use as starting material; verify what you adopt.
One command pulls everything into your tool's agents/skills folder.
Claude Code (user-wide):
curl -fsSL https://raw.githubusercontent.com/Solonnikov/agent-skills/main/install.sh | bashOther targets (run the installer with flags):
git clone https://github.com/Solonnikov/agent-skills.git
cd agent-skills
./install.sh --target cursor # .cursor/rules/ in the current dir
./install.sh --target copy --dest ./foo # plain copy to a path
./install.sh --agents-only
./install.sh --skills-only
./install.sh --helpDefault mode is symlink from a clone (so git pull updates your local copy) and copy when curl-piped.
Any other AI / LLM: the content is plain Markdown. Open any file, paste into your tool's system prompt, context, or rules section.
agent-skills/
├── install.sh
├── agents/
│ └── software-development/ # generic role agents (no framework lock-in)
└── skills/ # reusable how-to skills, one folder each
├── agent-skill-creator/
├── content-repurposing/
├── dapp-architect/
├── email-drafting/
├── evm-contract-scaffold/
├── git-workflow-recipes/
├── hardhat-etherscan-verification/
├── meeting-notes-structure/
├── ngrx-feature-scaffold/
├── prompt-engineering-patterns/
├── regex-cookbook/
├── research-brief-structure/
├── resume-tailor/
├── reown-appkit-web3/
├── solana-program-scaffold/
├── stripe-subscription-lifecycle/
├── technical-seo-checklist/
├── wagmi-contract-interaction/
└── writing-style-editor/
Skills — procedures, patterns, templates, and checklists an agent loads on demand. Each skill is a folder with a tight SKILL.md and long-form references/.
Agents — role-style Markdown definitions (who is responsible for what, how they decide, how they hand off). Generic by design; framework-specific knowledge lives in skills, not in agent descriptions.
| Skill | Category | Purpose |
|---|---|---|
| agent-skill-creator | Meta | Authoring new skills in this repo's format. |
| dapp-architect | Architect | End-to-end dApp design — brief, trust boundaries, contracts, client, indexing, off-chain, ops. |
| git-workflow-recipes | Everyday dev | Recipes for undoing, rebasing, resolving conflicts, rescuing lost work, PR cleanup. |
| regex-cookbook | Everyday dev | Copy-ready regex patterns, building blocks, language differences, debugging. |
| prompt-engineering-patterns | AI / LLM | Production prompt patterns — roles, few-shot, CoT, structured output, caching, evaluation. |
| ngrx-feature-scaffold | Angular | Scaffold a complete NgRx feature (actions, reducer, effects, selectors, facade, tests). |
| technical-seo-checklist | Web / SEO | Metadata, structured data, crawlability, Core Web Vitals, international targeting. |
| evm-contract-scaffold | Web3 (EVM) | Bootstrap a Solidity project — Foundry-first, OpenZeppelin patterns, testing, deployment. |
| hardhat-etherscan-verification | Web3 (EVM) | Verify deployed contracts — plugin path + manual V2 API fallback. |
| reown-appkit-web3 | Web3 | Integrate @reown/appkit multi-chain wallets (EVM, Solana, Bitcoin). |
| solana-program-scaffold | Web3 (Solana) | Anchor program scaffold — PDAs, SOL/SPL variants, constraint validation, testing. |
| wagmi-contract-interaction | Web3 (EVM) | Read, write, and watch EVM smart contracts with wagmi v2. |
| stripe-subscription-lifecycle | Backend / Payments | Stripe subscriptions end-to-end — webhook-driven state sync, cancellation, tier downgrades, testing. |
| writing-style-editor | Writing / Life | Tighten prose — cut fluff, vary sentence length, strengthen verbs, match a voice. |
| email-drafting | Writing / Life | Draft emails in the right tone — cold, follow-up, apology, decline, feedback. |
| meeting-notes-structure | Writing / Life | Notes that capture decisions and action items — templates for standups, 1-on-1s, retros, planning. |
| resume-tailor | Writing / Life | Tailor a résumé to a JD — parse, rewrite bullets, quantify, pass ATS. |
| content-repurposing | Writing / Life | One piece of content across multiple platforms — blog → thread → LinkedIn → newsletter. |
| research-brief-structure | Writing / Life | Briefs that answer a question — sources, synthesis, recommendation, gaps. |
agents/software-development/ — frontend-developer, backend-developer, test-engineer, code-reviewer, security-reviewer, ui-reviewer, web3-auditor, product-manager, devops-engineer, technical-writer, solution-architect.
| Tool | How the repo drops in |
|---|---|
| Claude Code | ./install.sh — files go to ~/.claude/agents/ and ~/.claude/skills/. Agents with YAML frontmatter are directly invocable. |
| Cursor | ./install.sh --target cursor — files go to ./.cursor/rules/. Rename to .mdc and adjust frontmatter per Cursor docs if needed. |
| Codex / Copilot / Aider / Continue | ./install.sh --target copy --dest <your-tool's-path>, or copy the Markdown content into your tool's context/rules/prompt field. |
| Any other LLM | Open the file in GitHub, copy the Markdown, paste into your system prompt or context. |
Content is written in plain Markdown so every tool that reads text can use it.
Pull requests and ideas welcome. See CONTRIBUTING.md for the full working flow — branch conventions, PR workflow, skill authoring steps, release rules, semver cadence, and commit/PR-title conventions.
Quick summary:
mainis protected — every change goes through a PR.- New skill → follow the format in
agent-skill-creator. - Patch for fixes, minor for a new skill/agent, major for breaking changes.
- Clean PR titles → clean auto-generated release notes.
Most of the content here is AI-generated or produced with significant AI assistance, then reviewed and edited. It is not a guarantee of accuracy, completeness, or fit for your use case — treat it like any other generated material and verify what you adopt.
These skills and agents are experimental. Their behavior and side effects are not fully characterized. Content may be wrong for your situation or unsafe without human review (running commands, changing files, applying security-related guidance). This is not legal, financial, or professional advice.
Do not use or apply this material on systems you do not own or are not authorised to change. Use at your own risk.
MIT. Individual skills or agents may state otherwise; check the file if in doubt.