Fronesis Security Hub — AI Agent Security Skills

DCL Skills Suite

Leibniz Layer™ · Free · Open

Security skills for AI agents and LLM pipelines. Download SKILL.md and drop it into any ClawHub-compatible runtime. Instruction-only skills run entirely inside the agent context. Webhook-powered skills return cryptographic tx_hash proofs via Leibniz Layer™. All seven skills form the complete DCL verification stack.

Verified ✓

DCL Policy Enforcer

/dcl-policy-enforcer

Cryptographic compliance enforcement for AI outputs. Detects jailbreaks, prompt injection, EU AI Act violations, GDPR breaches. Every evaluation produces a tamper-evident audit proof.

Verified ✓

DCL Sentinel Trace

/dcl-sentinel-trace

Detects and redacts PII in AI outputs: emails, phones, SSNs, bank cards, IBANs, crypto addresses, IPs. Delivers cryptographic tamper-evident audit proofs. GDPR & HIPAA compliant.

Benign ✓

DCL Semantic Drift Guard

/dcl-semantic-drift-guard

Detects hallucinations and context drift in LLM outputs. Compares AI response against source documents or RAG knowledge base. Verdict: IN_COMMIT or HALLUCINATION_DRIFT.

Benign ✓

DCL Skill Auditor

/dcl-skill-auditor

534 out of 3,984 ClawHub skills contained critical vulnerabilities — credential theft, prompt injection, data exfiltration. Snyk Research, 2026. Verdict: PASS / WARN / BLOCK.

Benign ✓

DCL Provenance Tracker

/dcl-provenance-tracker

Supply chain & version drift verifier. Detects suspicious changes between skill versions — new network calls, credential access, obfuscation. Verdict: PASS / WARN / BLOCK.

Verified ✓

DCL Prompt Firewall

/dcl-prompt-firewall

Intercepts malicious prompts before they reach your LLM. Detects prompt injection, jailbreaks, token smuggling, role-switch attacks. The first gate in the DCL Security pipeline.

Verified ✓

DCL Secret Leak Detector

/dcl-secret-leak-detector

Runtime credential scanner for LLM outputs. Catches API keys, tokens, private keys, DB connection strings, and .env values before they escape your system. SOC 2 & ISO 27001 aligned.

DCL Crypto Suite

Leibniz Crypto · DeFi · Live

Cryptographic security layer for DeFi agents and trading pipelines. Every evaluation produces a tamper-evident tx_hash committed to the Leibniz Layer™ audit chain. Eight modular skills — from prompt firewall to on-chain commit.

Verified ✓

DCL Prompt Firewall Crypto

/dcl-prompt-firewall-crypto

First gate for every crypto agent. Blocks jailbreaks, DAN/STAN, role-switch, drain-wallet injections, and unlimited-approval attacks before they reach your LLM.

Verified ✓

DCL Wallet Guardian

/dcl-wallet-guardian

Masks seed phrases, private keys, and wallet addresses before they reach any LLM context. Auto-redacts BIP-39 mnemonics, WIF keys, ETH private keys.

Verified ✓

DCL Trade Verifier

/dcl-trade-verifier

Audits every AI trading decision. Blocks guaranteed-return claims and unqualified advice. Generates immutable trade receipts with tx_hash for compliance.

Verified ✓

DCL MEV Compliance Shield

/dcl-mev-compliance

Detects front-running, sandwich attacks, wash trading, and KYC/AML red flags in agent outputs. EU AI Act Art. 9/13 aligned. For institutional DeFi desks.

Verified ✓

DCL Provenance Tracker Crypto

/dcl-provenance-tracker-crypto

Supply chain verification for on-chain skills and smart contracts. Detects eval/exec obfuscation, selfdestruct, delegatecall, and credential exfiltration patterns.

Verified ✓

DCL Semantic Drift Crypto

/dcl-semantic-drift-crypto

Guards market signals against hallucination. Detects fabricated prices, invented token names, and contradicted indicators. Verdict: IN_COMMIT or HALLUCINATION_DRIFT.

Verified ✓

DCL Output Sanitizer

/dcl-output-sanitizer

Last gate before delivery. Strips API keys, connection strings, shell injection fragments, and toxic content the model may have reproduced from tools or RAG context.

Verified ✓

DCL Crypto Commit

/dcl-crypto-commit

Commits every trading decision to the Leibniz Layer™ audit chain. Returns tx_hash + chain_hash as tamper-evident Merkle proof. The final seal on every DeFi agent action.

Available skills

hub.fronesislabs.com

$1 USDC

Quick Scan

Fast vulnerability sweep. Common attack vectors, reentrancy, access control. Returns severity score + top findings in seconds.

Live Most popular Base Mainnet

$5 USDC

Standard Audit

4-model consensus engine. Grok + Claude + DeepSeek cross-check code for bugs, ethics, and strategic risks. Full PDF report.

Live Base Mainnet

$15 USDC

Deep Audit + PoC

Full audit with proof-of-concept exploit generation. Verifiable on-chain commitment. For production systems and pre-launch checks.

Coming soon Base Mainnet

$3 USDC

MCP Server Scan

Security audit for Model Context Protocol servers. Tool trust, data leakage, prompt injection vectors. Built for the agent economy.

Coming soon Base Mainnet

$40 USDC/mo

Retainer

10 audits per month. Priority queue. Dedicated report storage. For teams shipping agents continuously.

On request Base Mainnet

List your skill

Got a security skill? Publish it here. Get paid in USDC. Earn per call.

Apply below

Integration example

x402 protocol

agent.ts — pay per audit

// Any agent that speaks HTTP can call this
import { x402Fetch } from "@x402/fetch";

const result = await x402Fetch(
  "https://hub.fronesislabs.com/x402/dcl_verify",
  {
    method: "POST",
    body: JSON.stringify({
      code: "<your smart contract>",
      agent_id: "my-agent-v1"
    })
  },
  walletClient  // pays $5 USDC on Base automatically
);

const { hash, findings, severity } = await result.json();
// hash is on-chain — tamper-evident forever
// findings: [{type, severity, description, line}]

Sell your security skill here

Got a working AI security tool? List it on the marketplace. Set your price in USDC. We handle x402 payment routing and settlement on Base. No Stripe, no KYC.

Apply to list →

Privacy & Data Policy

Webhook · Skills

When you use any skill powered by Fronesis Labs — including DCL Policy Enforcer and DCL Sentinel Trace — the evaluated text is transmitted to our webhook at webhook.fronesislabs.com.

What we process: The text you submit for evaluation. Nothing else — no user identity, no API keys, no metadata beyond what's needed to run the check.

Retention: Evaluations are processed in-memory only. No text is written to disk, no logs are retained, no data is shared with third parties. The only persistent record is the cryptographic tx_hash and chain_hash — these contain no personal data.

Sentinel Trace: When using the sentinel_trace policy, detected personal identifiers are returned as redacted samples only (e.g. te****st@****.com). Raw values are never stored or logged.

Infrastructure: Webhook hosted on a private VPS. Operated solely by Fronesis Labs. No cloud analytics, no third-party processors. Questions: support@fronesislabs.com

Fronesis
Security Hub

Find a skill

Send HTTP request

Get signed report

Sell your security skill here

FronesisSecurity Hub

Find a skill

Send HTTP request

Get signed report

Sell your security skill here

Fronesis
Security Hub