My AI Agent Skills Stack (Open Source)
I maintain open-source skills that supercharge AI coding agents — softco, skill-builder, seo-master, frontend-design, and more. Here's what each one does.
AI coding agents are only as good as their domain knowledge. A raw agent knows how to code; a skilled agent knows your architecture, your constraints, and your product goals. I've built a library of open-source skills that pack that domain knowledge into reusable modules for Claude Code.
What a skill is
A skill is a SKILL.md file that loads into Claude Code via the /skill-name command. It gives the agent:
- Domain context — what this skill covers, key concepts, hard rules
- Reference files — detailed guides, examples, checklists
- Behavioral constraints — what to do, what never to do
Skills are plain Markdown. Any agent that can read text can use them.
The skills I've built
softco — Full product lifecycle agent
The flagship skill. softco acts as PM, architect, senior engineer, DevOps, QA, and tech writer in one loop. It produces specs, architecture diagrams, implementation plans, code, tests, and deployment scripts.
/softco build [feature description]
/softco audit [repo name]
/softco fix [bug description]
/softco deploy [service name]
Built on the insight that a solo founder needs the whole company in one agent, not just a code autocomplete.
skill-builder — Build skills from descriptions
Meta skill. Describe what you want a skill to do in plain English; skill-builder produces a complete SKILL.md, reference files, hooks, and marketplace scaffold.
/skill-builder "A skill that helps write Go gRPC services
with buf toolchain, proto conventions, and error handling patterns"
Outputs: SKILL.md, references/grpc-patterns.md, references/buf-config.md, hooks for pre-commit proto linting.
seo-master — SEO + GEO + AEO
Full 2026 SEO playbook covering traditional search, AI answer engines (ChatGPT, Claude, Gemini, Perplexity), structured data, titles/meta, content strategy, and Search Console setup. Framework-aware — knows Next.js App Router Metadata API, React SPA prerendering, Astro, Remix.
frontend-design — Production-grade UI
Strict guidelines for modern, production-grade UI/UX that avoids generic AI aesthetics (the "AI slop" look). References real design systems, component patterns, and accessibility rules.
openrouter-free — Free LLM routing
Routes to the best free LLM available via OpenRouter for a given task. Keeps API costs near zero for non-critical inference. Useful for batch processing, non-customer-facing tasks, and experimentation.
common-knowledge — Persistent agent memory
Git-backed memory store. Agents save and load knowledge across sessions — learned project context, gotchas, architectural decisions, lessons. Cross-platform, readable by any agent.
/ck save "learned that the auth middleware must run before CORS,
not after — middleware order matters in Chi"
/ck load context-heavy
/ck search "database schema"
clean-my-mac — macOS cleanup
Scans and removes build artifacts, node_modules, .next caches, Docker images, and temp files. Recovers gigabytes with one command.
Skills I use on every project
| Skill | When | |-------|------| | softco | Every feature, audit, and deploy | | seo-master | Every landing page and blog post | | frontend-design | Every UI component | | common-knowledge | Start and end of every session | | code-review | Before every PR |
How to install
Skills live in ~/.claude/skills/. Each skill is a directory with SKILL.md + references/:
# Via Claude Code plugin registry (if published)
# Or manually:
git clone https://github.com/shihabshahrier/skills ~/.claude/skills/mine
Then in Claude Code: /skill-name args — the agent loads the skill and runs with full domain context.
Building your own
The fastest way: describe the skill to skill-builder.
/skill-builder "A skill for writing PostgreSQL migrations using
golang-migrate. Should know: naming conventions (YYYYMMDD_description.up.sql),
reversible migrations, index creation strategies, backfill patterns for
large tables, and transaction safety for DDL."
skill-builder generates the full skill structure. Review and edit the output — it's yours.
FAQ
What are AI agent skills? Skills are knowledge modules for AI coding agents (like Claude Code) that provide domain-specific context, rules, and examples — turning a general-purpose agent into a specialist.
Are these skills open source? Yes — all skills I've built are open source on GitHub at shihabshahrier.
Do skills work with agents other than Claude Code? The SKILL.md format is plain Markdown. Any agent that can read context files can use skills. They're optimized for Claude Code but readable by any LLM.
What's the difference between a skill and a system prompt? System prompts are per-conversation. Skills persist across sessions, can include reference files, and are modular — you can stack multiple skills. They're also version-controlled and shareable.
Written by Shihab Shahriar Antor — AI Engineer & Founder of Shahriar Labs. See also: How I Use Claude Code to Ship Faster · The Tools I Built to Make AI Coding Agents 10x.