Skillkit
The package manager for AI agent skills
Open Source
Artificial Intelligence
GitHub

Featured onFebruary 7th, 2026

The universal skill platform for AI coding agents. Auto-generate instructions with Primer, persist learnings with Memory, and distribute across Mesh networks. One CLI for Claude, Cursor, Windsurf, Copilot, and 28 more.

Top comment

Upvotes245

▲ 245View on ProductHunt ⧉

Comments21

21 commentsSee comments on PH ⧉

Product of the Day3rd

Hey Product Hunt! 👋
I'm Rohit, the maker of SkillKit.
The problem I kept running into: My skills for Claude Code were useless in Cursor. My team was on Copilot. Each agent has its own format. Worse: every time I close a session, my AI forgets everything it learned.
So I built SkillKit - a universal platform for AI coding agents that goes way beyond skill management.
npx skillkit@latest
What makes it different from "skill installers":
🔄 Cross-Agent Translation (32 agents)
skillkit translate my-skill --from claude --to cursor,codex,copilot
One command. Write once, use in Claude Code, Cursor, Codex, Windsurf, Copilot, Gemini CLI, and 26 more.
🧠 Session Memory That Persists
Your AI agents learn patterns, but that knowledge dies every session. SkillKit captures learnings with semantic embeddings and makes them persistent:
skillkit memory compress    # Extract patterns from sessions
skillkit memory search "auth patterns"  # Recall past learnings
skillkit memory export      # Turn learnings into shareable skills
🤖 Multi-Agent Team Orchestration
Spawn teams of AI agents with leader/teammate hierarchies, task assignment, plan approval workflows, and code review stages:
skillkit team init
skillkit message send       # Inter-agent messaging
skillkit workflow run       # Orchestrate multi-step tasks
🌐 Mesh Network for Distributed Teams
Your agents can communicate across machines with E2E encrypted P2P:
- Ed25519 cryptography & XChaCha20-Poly1305 encryption
- UDP multicast LAN discovery
- Trust management with fingerprint verification
🎯 AI-Powered Recommendations
skillkit recommend --explain
# > 92% match: vercel-react-best-practices (Next.js detected)
# > 87% match: tailwind-v4-patterns (Tailwind 4 in package.json)
📚 Built-in Methodology Packs
TDD, Design-First, Root Cause Analysis, Structured Review - battle-tested development methodologies baked in.
More Features:
- 🧪 Skill Testing Framework with assertions
- 🔧 Auto-generate CI/CD configs (GitHub Actions, GitLab CI)
- 🌳 Hierarchical skill taxonomy with tree navigation
- 📡 Self-host your skills (RFC 8615 well-known URIs)
- 🔌 Plugin system for extensions
- 📊 Quality scoring and security audits
This is for you if:
- You switch between AI coding agents
- Your team uses different tools
- You want your AI to actually remember and learn
- You're building multi-agent workflows
- You want enterprise-grade skill management
Fully open source.
Website: agenstskills.com | Docs: agenstskills.com/docs

Comment highlights

Package manager for agent skills is clever! How do you handle version conflicts when multiple skills need different dependencies?
Also curious how you validate that skills are safe before installing - any sandboxing?

Every time I onboard a new team member, I end up exporting my Claude Code skills as a markdown dump and manually reformatting for whatever agent they picked. Cross-Agent Translation across 32 agents solves the distribution problem. Session Memory is the part I want to stress-test... semantic embeddings sound right, but the failure mode is usually recall quality degrading as the memory store grows. Compression strategy matters a lot there.

Wait, so this is basically npm but for AI agents? Pretty sick concept. I’m curious though—how does the 'Memory' feature actually handle persistence across different agents like Claude and Cursor? Is it just syncing a JSON file or something more complex?

This looks promising! How does Skillkit handle versioning and dependency management for agent skills? Would love to learn more about the workflow.

The cross-agent translation for 32 agents is impressive. I use Claude Code daily and switching context between tools is painful. Curious about the Primer feature — how does auto-generating instructions work in practice? Does it analyze your codebase patterns and suggest skill definitions, or is it more template-based? Also, for team scenarios where different members use different agents (Claude Code vs Cursor vs Copilot), how does skill synchronization work across the mesh network?

Same boat as kxbnb - been manually converting CLAUDE.md to .cursorrules whenever switching tools. The translate command alone is worth it. Also curious about the mesh network for multi-agent setups, sounds like it could be interesting for coordinating different specialized agents.

The concept of a package manager for AI skills is brilliant. As a dev, I’m curious—how does it handle dependency versioning for different LLM models? Great work on making it Open Source!

I maintain about 15 Claude Code skills for my daily workflow and the portability problem is real. Right now if I want the same behavior in Cursor I'm manually rewriting CLAUDE.md into .cursorrules. The translate command would save me a lot of time.
Curious about the memory feature -- how does it handle conflicting patterns across projects? Like if I have one repo that uses snake_case and another that uses camelCase, does it scope learnings per-project or is it global?

Really cool to see this direction. The portability issue between Claude Code, Cursor, Copilot, etc. is something I’ve run into as well — each agent having its own “skill dialect” makes it hard to build consistent workflows.

I’m working on a different layer of the stack (Iceberg Framework), where the focus is on deterministic execution and validation for LLM‑powered systems. What you’re doing with Skillkit around persistent skills and cross‑agent translation actually complements that nicely — one solves the portability problem, the other solves the reliability/consistency problem.
Love seeing more tooling appear in this space. The ecosystem really needs it.

Package manager for AI skills is brilliant! I'm working with Claude Code and the problem of 'teaching' it project-specific patterns is real. Does Skillkit let me create custom skills like 'how we structure Phoenix LiveView components' and have the agent remember across sessions? The cross-platform support (Claude, Cursor, etc.) is the killer feature - skills shouldn't be tool-locked. How does versioning work? Congrats on open sourcing this!