Turn any agent into a domain expert in seconds. Plug in. Upload a skill. "I know kung fu." — signed at release, adversarially benchmarked, Trust Score your security team can verify.
500+ signed, adversarially-tested skills, playbooks, souls and guardrails — one MCP endpoint or npx super-agent install away from any AI agent.
Think of it as a library of instant expertise for AI: pick a skill, beam it in, and your agent suddenly knows OWASP, MEDDPICC, ECG triage, SOC 2 evidence collection, or whatever your domain demands. Built for teams that have to stand behind what their AI does — security, fintech, healthcare, SOC 2 and DevOps. Works with Claude, Cursor, ChatGPT, Continue, Cline. No fine-tuning. No vendor lock-in.
🌐 Website · 🛒 Marketplace · 📚 Docs · 🔌 MCP · 💬 Discord
Models are commodities. Expertise is the moat. Today there are abundant models, agents, MCPs and workflows — but almost no one is dominating the distribution of intelligence. That layer compounds.
Super Agent Skill is the university for AI agents — a Matrix-style upload that turns any model into a domain expert in seconds. Open source by default, community-curated, free to download, and continuously evolved by an automated quality loop (SkillForge).
- Free tier — the open campus. 500+ packages. Fork, remix, install, redistribute.
- Premium tier — superagentskill.com curates battle-tested elite stacks, enterprise playbooks and private packs for teams that want depth, not just breadth.
The open network feeds the premium layer. The premium layer funds the open network. That's the flywheel.
The registry is live and growing every week. All packages are free to download, fork and run. The In this repo column is the open seed you can
git clonetoday; the Hosted registry column is the full, continuously-evolved library on superagentskill.com.
| Type | In this repo (open seed) | Hosted registry | Examples |
|---|---|---|---|
| 🛠 Skills | 76 | 370+ | code review, OWASP audit, SQL translator, OSINT, ECG triage, MEDDPICC discovery |
| 📘 Playbooks | 2 | 30+ | bug triage → fix → PR, content pipeline, incident response |
| 🎭 Souls | 6 | 50+ | pragmatic SRE, empathetic support, senior product designer |
| 🛡 Guardrails | 2 | 50+ | no-PII, no-medical-advice, GDPR, brand safety |
➡️ Browse the full marketplace at superagentskill.com →
Point any MCP-compatible client at:
https://superagentskill.com/api/mcp
That's it. Your agent now has discover / install / evaluate tools and live access to every skill in the registry — including patches shipped overnight by SkillForge.
# Grab a single skill
curl -O https://raw.githubusercontent.com/criptogus/agent-evolve-network/main/content/skills/code-reviewer.yaml
# Or clone the whole library
git clone https://github.com/criptogus/agent-evolve-network.gitDrop the YAML into your agent's system prompt. Done.
Go to superagentskill.com, describe what your agent should do (e.g. "triage cardiology patients using the GRACE score") and the Forge will research, draft, adversarially test and publish a production-grade package — usually in under 5 minutes.
bun install && bun run devThe problem: anyone can write a YAML prompt and call it an "AI skill." The hard part is convincing your security team it won't leak PII, accept a jailbreak, or hallucinate a compliance disclosure in front of a customer.
What we ship that other registries don't:
- 🛡 Proprietary adversarial harness. Every skill is benchmarked against attacks before it can publish — prompt injection, jailbreaks, exfiltration, blast-radius, policy bypass, PII/PHI leakage. Cases are curated per vertical (OWASP LLM, FINRA, HIPAA Safe Harbor, PCI-DSS, SRE).
- 🔏 Ed25519-signed releases. Every published version is cryptographically signed. Verify offline with
npm run release:verify— exactly what air-gapped enterprise customers need. - 🧾 Signed per-package trust attestations.
npm run trust:attestbinds a package's exact content hash + version to its adversarial result and signs it with the release key. Your security team runsnpm run trust:verify -- --attestation pkg.trust.json --pkg pkg.yaml --pubkey pub.pemand confirms offline that this exact file passed N adversarial cases — no need to trust the hosted Trust Score endpoint. - 📊 Public, verifiable Trust Score. A transparent weighted formula over adversarial robustness, real-world success rate, signed releases and package age. Embed the badge in your README; reviewers see the same number you do.
- 🏛 Verticalized Souls. Fintech compliance officer (FINRA / Reg E / PCI-DSS), HIPAA-aware clinical liaison, SOC 2 auditor (TSC-mapped), Kubernetes SRE (blast-radius-first). Souls that cite the rule before the recommendation.
- 🔌 One MCP endpoint, every IDE. Claude, Cursor, ChatGPT, Continue, Cline — same URL, same
npx super-agent install <slug>command. No SDK, no glue code. - 🛠 SkillForge evolution loop. Re-scores published skills against the live adversarial suite + telemetry; proposes patches automatically when robustness drops.
Other registries hand you prompts. We hand you proof.
| If you're… | What you get |
|---|---|
| 🚢 Eng leadership shipping AI features | Trust Score on the deploy checklist instead of a Notion doc. Real-world success rate per skill. |
| 🔒 Security / compliance officer | Offline-verifiable signatures. Adversarial pass rate broken down by attack class. Upload-time injection-guard audit log. |
| 🏥 Fintech / healthcare / SOC 2 team | Souls that cite FINRA 2210, Reg E §1005.7, HIPAA Safe Harbor, TSC CC6.x. PII/PHI guardrails redact PAN/CVV/MRNs before output. |
| 💻 Solo builder using Claude, Cursor or Continue | npx super-agent install <slug> drops the right file into every IDE you use. No login for public skills. |
| ✍️ Skill author | Premium revenue share, 12-month referral window, perpetual lineage cut on every fork, open bounties from companies. |
| What it is | Use it when… | |
|---|---|---|
| 🛠 Skills | A focused capability with system prompt, rules, examples and a runnable input → output contract | You want the agent to do one thing very well (review code, write SQL, summarize a meeting). |
| 📘 Playbooks | A multi-step decision graph that orchestrates skills + tools | The task has a recurring shape (triage bug → reproduce → propose fix → open PR). |
| 🎭 Souls | A persona / value bundle — voice, taste, defaults — that wraps any agent. Always free. Souls inspired by public figures are fictional AI personas — not the real person (disclaimer). | You need a consistent brand, tone or worldview across all tasks. |
| 🛡 Guardrails | Refusal, safety and compliance policies enforced before output | You ship to real users and need hard "must / must not" rules. |
Each one is a single self-contained YAML file. Read it, fork it, ship it.
content/
├── skills/ → Single capabilities (370+ on superagentskill.com)
├── playbooks/ → Multi-step workflows (30+ on superagentskill.com)
├── souls/ → Personas and values (50+ on superagentskill.com)
├── guardrails/ → Safety and compliance (50+ on superagentskill.com)
└── schemas/ → JSON Schemas (every package validates against these)
💡 This GitHub repo is the open seed — the canonical, public-domain core. The full 500+ library lives at superagentskill.com/marketplace, where SkillForge keeps it sharp.
Whatever your agent does, there's almost certainly a package for it — or you can have one in minutes:
| If your agent is a… | Start here |
|---|---|
| Coding assistant | code-reviewer, owasp-code-audit, dependency-vuln-auditor |
| Security / red team | recon-attack-surface, prompt-injection-tester, disk-image-forensics |
| SRE / DevOps | incident-response-triage, cloud-misconfig-auditor, pragmatic-sre |
| Researcher / writer | web-research, long-form-writer, osint-investigator |
| Customer support | empathetic-support, no-pii-in-output |
| Anything else | 👉 Search the marketplace → or ask the Forge to build it |
# 1. Fork & clone
git clone git@github.com:<you>/agent-evolve-network.git
cd agent-evolve-network && bun install
# 2. Start from a template
cp content/skills/_template.yaml content/skills/my-killer-skill.yaml
# 3. Validate locally (same script CI runs)
bun run validate:content
# 4. Open a PR — a maintainer reviews within 72hA merged skill is automatically imported into the hosted registry, where SkillForge starts evolving it against adversarial benchmarks and serving the best variant through MCP to every connected agent on day one.
➡️ Full quality bar in CONTRIBUTING.md.
Your work isn't just merged — it's amplified.
- 🌐 Reach. Every accepted package is served live through MCP to thousands of agents.
- 🏆 Reputation. Public Trust Score, model compatibility matrix and leaderboard rankings on the marketplace.
- 💸 Revenue share. Premium packages earn 80–85% of sales + recurring usage bonuses.
- 🤖 AI co-author. SkillForge proposes patches when your skill drifts — you stay in control, accept or reject.
- 🪪 CVE-style credit. Robustness findings (e.g.
SAS-2026-0001) are publicly attributed.
First-time contributor? Look for
good first issueandhelp wanted— most are mergeable in one afternoon.
| Super Agent Skill | A prompt library on GitHub | A walled-garden GPT store | |
|---|---|---|---|
| Open content (CC BY-SA) | ✅ | ✅ | ❌ |
| Schema-validated quality bar | ✅ | ❌ | |
| Works in any MCP agent | ✅ | ❌ single vendor | |
| Continuous adversarial evolution | ✅ | ❌ | ❌ |
| Public trust + compatibility scores | ✅ | ❌ | ❌ |
| Revenue share for authors | ✅ | ❌ | |
| One-line install via MCP | ✅ | ❌ | ❌ |
- MCP server with discover / install / evaluate / report-execution tools
- SkillForge evolution loop (eval → patch → re-score)
- Public Trust Score, model compatibility matrix, drift detection
- 500+ packages live in the hosted registry
- CLI (
npx sas init,sas publish,sas eval) - Cryptographically signed package releases
- Versioned downloadable registry bundles
- Community-curated collections (curated bundles around a use case)
Track everything on the public board.
- 💬 Discord — join here for help, design reviews, and weekly skill-jam events.
- 🐦 X / Twitter — @superagentskill for releases and benchmarks.
- 📰 Changelog — every accepted package is announced in GitHub Releases.
- 🛡 Security — please disclose responsibly via SECURITY.md.
- Code (this repo, outside
content/): Apache License 2.0 - Content (everything in
content/): Creative Commons Attribution-ShareAlike 4.0
By contributing you agree your work is licensed under the same terms — and that it gets to make millions of agent conversations a little smarter every day.
→ Browse 500+ free packages at superagentskill.com
Built by makers, for makers. Star the repo ⭐ if you believe agents should belong to everyone.