diff --git a/README.md b/README.md index fa6f3b61..5d2f5af3 100644 --- a/README.md +++ b/README.md @@ -4,14 +4,14 @@ # Super Agent Skill -### **The open network where AI agents learn new skills.** +### **Trustworthy AI skills for teams shipping to production.** -> **We don't build AI agents. We evolve them.** -> Models are commodities. Infrastructure is a commodity. **Skills are the new moat.** +> **Other registries hand you prompts. We hand you proof.** +> Signed at release · Adversarially benchmarked · Trust Score your security team can verify. -**500+ free skills, playbooks, souls and guardrails — one MCP endpoint away from any AI agent.** +**500+ signed, adversarially-tested skills, playbooks, souls and guardrails — one MCP endpoint or `npx super-agent install` away from any AI agent.** -Turn Claude, Cursor, Codex, Continue or any MCP-compatible agent into a domain specialist in **60 seconds**. No fine-tuning. No copy-pasting prompts. No vendor lock-in. +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](https://img.shields.io/badge/🌐_superagentskill.com-Visit-0A66FF?style=for-the-badge)](https://www.superagentskill.com) [![Marketplace](https://img.shields.io/badge/🛒_Browse_500+_packages-Free-8b5cf6?style=for-the-badge)](https://www.superagentskill.com/marketplace) @@ -94,15 +94,28 @@ bun install && bun run dev ## 🧠 Why Super Agent Skill? -Generic agents are jacks of all trades, masters of none. They hallucinate, drift, and forget the rules your team actually cares about. +**The 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. -We fix that with three ingredients you won't find anywhere else: +**What we ship that other registries don't:** -- 🌍 **An open, growing registry** of 500+ battle-tested skills, playbooks, souls and guardrails — community-owned, MIT/CC-friendly, embeddable anywhere. -- ⚙️ **SkillForge** — our continuous evolution loop benchmarks every skill against adversarial cases, finds weaknesses, and ships patches automatically. Your agent gets smarter while you sleep. -- 🔌 **One MCP endpoint** that streams the strongest version of every skill to any compatible agent — no copy-paste, no version drift, no vendor lock-in. +- 🛡 **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. +- 📊 **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 ` 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. -> *Think GitHub for prompts × npm for agent capabilities × Wikipedia for AI behavior — all under MCP.* +> *Other registries hand you prompts. We hand you proof.* + +## 👥 Who it's for + +| 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 ` 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. | --- diff --git a/src/routes/index.tsx b/src/routes/index.tsx index fbb0bbe3..3c958bf8 100644 --- a/src/routes/index.tsx +++ b/src/routes/index.tsx @@ -15,13 +15,13 @@ const McpInstallAnimation = lazy(() => export const Route = createFileRoute("/")({ head: () => ({ meta: [ - { title: "Super Agent Skill — Turn your AI into an expert at your job" }, - { name: "description", content: "The app store for your AI's brain. One click installs an expert in your field — works with ChatGPT, Claude, Cursor, Codex." }, - { property: "og:title", content: "Turn your AI into an expert at your job — Super Agent Skill" }, - { property: "og:description", content: "The app store for your AI's brain. One click installs an expert in your field." }, + { title: "Super Agent Skill — Trustworthy AI skills for serious work" }, + { name: "description", content: "Signed, adversarially-tested AI agent skills with verifiable Trust Scores. For security, fintech, healthcare and DevOps teams shipping AI to customers. Install with one MCP link or npx super-agent install." }, + { property: "og:title", content: "Trustworthy AI agent skills — Super Agent Skill" }, + { property: "og:description", content: "Signed, adversarially-tested AI agent skills with verifiable Trust Scores. For teams that ship AI to customers." }, { property: "og:url", content: "https://superagentskill.com/" }, - { name: "twitter:title", content: "Turn your AI into an expert at your job — Super Agent Skill" }, - { name: "twitter:description", content: "The app store for your AI's brain. One click installs an expert in your field." }, + { name: "twitter:title", content: "Trustworthy AI agent skills — Super Agent Skill" }, + { name: "twitter:description", content: "Signed, adversarially-tested AI agent skills with verifiable Trust Scores." }, ], links: [{ rel: "canonical", href: "https://superagentskill.com/" }], }), @@ -44,7 +44,7 @@ const SOFTWARE_LD = { applicationCategory: "DeveloperApplication", operatingSystem: "Any (MCP-compatible)", description: - "MCP gateway and registry of skills, playbooks, souls and guardrails for AI agents. Continuously evolves connected agents.", + "Signed, adversarially-tested registry of skills, playbooks, souls and guardrails for AI agents. Every package ships with a verifiable Trust Score so teams can ship AI to customers safely.", url: "https://superagentskill.com", offers: { "@type": "Offer", price: "0", priceCurrency: "USD" }, }; @@ -96,6 +96,7 @@ function Home() { + @@ -122,52 +123,54 @@ function Hero() {
- For ChatGPT, Claude, Cursor, Codex & Grok users + For teams shipping AI to customers — not just experimenting

- Turn your AI into + AI skills you can
- an expert at  + actually trust with  - your job. + production.

- Generic AI gives generic answers. We give your AI the same - instructions, examples and rules - an expert in your field would use — so it stops guessing and starts - doing the job the way you would. + Every skill on Super Agent Skill is{" "} + benchmarked against an adversarial suite,{" "} + cryptographically signed at release, and scored on{" "} + real production telemetry — so you ship AI features + with a Trust Score your security team can verify, not a vibe check.

- Set it up in one click. No coding. No retraining. Works with the AI tools you already use. + Install with one MCP link or npx super-agent install. + Works with Claude, Cursor, ChatGPT, Continue and Cline.

- Get started — it's free → + Connect your agent — free → - Browse what's inside + Browse signed skills

- No credit card · Setup in 30 seconds · 4,200+ ready-to-use experts + No credit card · 30-second setup · Adversarially tested · Signed releases · Open source

- Models are commodities. Skills are the new moat. + Models are commodities. Trust is the new moat.

@@ -181,13 +184,13 @@ function Hero() { - {/* Trust metrics */} + {/* Trust metrics — proof of the moat, not vanity counts */}
{[ - { v: 4218, suffix: "+", label: "Ready-made experts" }, - { v: 91, suffix: "k", label: "AIs already upgraded" }, - { v: 30, suffix: "s", label: "Avg. setup time" }, - { v: 99.99, decimals: 2, suffix: "%", label: "Uptime" }, + { v: 100, suffix: "%", label: "Signed at release" }, + { v: 5, suffix: "", label: "Regulated verticals" }, + { v: 7, suffix: "", label: "Injection attack classes blocked" }, + { v: 30, suffix: "s", label: "MCP setup" }, ].map((m) => (
@@ -205,42 +208,42 @@ function Hero() { function WhatIsThis() { const items = [ { - icon: "🧠", - title: "Skills", - plain: "Knowledge packs", - body: "Like hiring a specialist — your AI gets the exact know-how to do one thing really well (review code, write SQL, triage a patient).", + icon: "🛡", + title: "Adversarial harness", + plain: "Proven robust", + body: "Every skill is tested against prompt injection, jailbreaks, exfiltration, blast-radius and policy bypass before it can publish. Verticalized cases for OWASP LLM, FINRA, HIPAA Safe Harbor, PCI-DSS and SRE.", }, { - icon: "📘", - title: "Playbooks", - plain: "Step-by-step recipes", - body: "Repeatable workflows your AI follows from start to finish. Think SOPs, but the AI actually executes them.", + icon: "🔏", + title: "Signed Trust Score", + plain: "Cryptographic + transparent", + body: "Releases are Ed25519-signed and air-gap verifiable. Trust Score is a public, weighted formula over adversarial robustness, real-world success, signed releases and age — embed the badge in your README.", }, { - icon: "🎭", - title: "Souls", - plain: "Personality & tone", - body: "Make the AI sound like your brand, your founder, or your top sales rep — consistently, every time.", + icon: "🏛", + title: "Verticalized Souls", + plain: "Regulator-aware personas", + body: "Fintech compliance officer, HIPAA-aware clinical liaison, SOC 2 auditor, Kubernetes SRE — Souls that cite the rule before the recommendation and refuse the unsafe defaults.", }, { - icon: "🛡", - title: "Guardrails", - plain: "Safety rules", - body: "Hard rules the AI cannot break: never give medical advice, never quote a competitor, never share customer data.", + icon: "🔌", + title: "One MCP endpoint", + plain: "Plug into any agent", + body: "Claude, Cursor, ChatGPT, Continue, Cline. No SDK, no glue code — one URL or one `npx super-agent install`. Skills compose into Playbooks at runtime with guardrail middleware.", }, ]; return (
- In plain English + What makes it different

- Think of it like an app store for your AI's brain. + Other registries hand you prompts. We hand you proof.

- Your AI is smart but generic. We give you a library of small, focused - "upgrades" - you install with one click — and your AI instantly becomes an expert in whatever you do. + Anyone can write a YAML prompt and call it a skill. The hard part is convincing your + security team it won't leak PII, accept a jailbreak, or hallucinate compliance text. + That's what we build.

@@ -259,8 +262,8 @@ function WhatIsThis() {
- The best part: they keep getting better automatically. - Every upgrade is tested daily and improved by the community — so your AI gets smarter while you sleep. + The flywheel: every production execution feeds the Trust Score. + Skills that survive real workloads bubble up; ones that drift get re-scored and patched by SkillForge automatically.
@@ -290,6 +293,71 @@ function Logos() { ); } +function WhoItsFor() { + const personas = [ + { + tag: "Eng leadership", + title: "You're shipping AI features your CEO already announced", + body: "You need agent skills that won't go rogue in front of customers. Trust Score + adversarial pass rate go on the deploy checklist, not in a Notion doc.", + cta: { href: "/marketplace?vertical=devops", label: "Browse SRE & code-review skills →" }, + }, + { + tag: "Security & compliance", + title: "You're the one who has to sign off on LLM usage", + body: "Every release is Ed25519-signed and offline-verifiable. Adversarial pass rate is broken down by attack class. Audit log captures every upload-time injection attempt.", + cta: { href: "/marketplace?vertical=security", label: "Browse signed security skills →" }, + }, + { + tag: "Regulated teams", + title: "Fintech, healthcare, SOC 2 — you can't ship a hallucination", + body: "Verticalized Souls cite the rule before the recommendation: FINRA 2210, Reg E, HIPAA Safe Harbor, TSC CC6.x. PII guardrails redact PAN/CVV and PHI before output.", + cta: { href: "/use-cases", label: "See vertical use cases →" }, + }, + { + tag: "Solo builders", + title: "You just want skills that work in Claude or Cursor today", + body: "One command — `npx super-agent install ` — drops the right file into every IDE you use. No login for public skills.", + cta: { href: "/marketplace", label: "Pick a skill, install in 30s →" }, + }, + { + tag: "Authors", + title: "You publish skills and want to get paid for them", + body: "Premium revenue share, 12-month referral window for who you bring in, and a perpetual lineage cut every time someone forks your work. Bounties from companies looking for specific skills.", + cta: { href: "/bounties", label: "See open bounties →" }, + }, + ]; + return ( +
+
+
+ Who it's for +

+ Built for the people who have to stand behind the AI. +

+

+ If your job depends on the agent being right — not just fluent — you're in the right place. +

+
+ +
+ {personas.map((p) => ( + +
{p.tag}
+

{p.title}

+

{p.body}

+
{p.cta.label}
+ + ))} +
+
+
+ ); +} + function HowItWorks() { const steps = [ {