Edison Flores

Posted on Aug 25

UTA v1.1.0 — Universal Trust Adapter: 8 formats, 12-stage pipeline, 14/14 audit findings fixed

#security #cryptography #aiagents #opensource

Universal Trust Adapter (UTA) v1.1.0 is now live. The USB-C of agent trust — one canonical schema, 8 format adapters, 12-stage verification pipeline.

What UTA does

UTA translates between 8 different trust credential formats used by AI agents via a canonical Universal Trust Schema (UTS v2.0.0):

ATC (Agent Trust Card — AliceLabs)
EAT-AI (IETF RFC 9421)
ZTA (Anthropic Zero-Trust Agent)
A2A Agent Card (Google/AAIF)
MCP Server Card (Anthropic)
W3C Verifiable Credentials
OAuth/OIDC
SPIFFE SVID

Like Zapier connects applications, UTA connects trust standards.

The crypto

Signature algorithm: Ed25519 (RFC 8032) — fast, compact, well-studied
Canonicalization: RFC 8785 JCS (JSON Canonicalization Scheme) — same bytes in every language
Hash: SHA-256 over canonical bytes
Domain separation: 7 distinct signature domains to prevent cross-context reuse
Proof-of-Possession: nonce challenge anti-replay

The reference implementation is ~200 lines of Node.js using only node:crypto. No external crypto dependencies.

Independent verification

5 frozen test vectors published with:

Canonical JCS bytes per vector (hex + base64 + utf8)
SHA-256 of canonical bytes
Ed25519 signature
Expected verification outcome
Test CA private key intentionally published for cross-language reproducibility

Any Python, Go, or Rust verifier can re-derive the signatures from scratch and confirm the crypto works as claimed.

Vector	Expected	What it tests
`minimal-valid.json`	✅ pass	Minimal valid ATC card
`expired.json`	❌ fail	Card with expires_at in the past
`tampered-payload.json`	❌ fail	trust_score modified after signing
`wrong-ca-key.json`	❌ fail	Verified with wrong CA key
`capability-samples.json`	—	Capability manifest edge cases

Conformance suite — 23/23 pass

git clone https://github.com/alicelabs-llc/universal-trust-adapter
cd universal-trust-adapter/marketnow/atc-sdk
npm install && node test/conformance.mjs

Output:

=== ATC/1.0 Conformance Test Suite ===
Test 1: A valid ATC verifies all 8 controls ✓
Test 2: Tampered payload fails ✓
Test 3: Wrong CA key fails ✓
...
Test 8: Missing required fields fails ✓

=== Summary ===
Passed: 23
Failed: 0

Independent security audit — 14/14 findings fixed

An independent security audit identified 14 findings across the codebase, marketplace, and documentation. All 14 have been fixed:

Severity	Total	Fixed
Critical	2	2 ✅
High	6	6 ✅
Medium	4	4 ✅
Info	2	2 ✅
Total	14	14 ✅

Full audit status at: https://marketnow.site/trust/audit-status.json

NPM packages — 7 published

Package	Version	Monthly downloads
`marketnow-mcp`	1.10.1	958
`agent-trust-card`	1.1.2	527
`marketnow-install-stack`	1.1.1	352
`@marketnow/trust-core`	1.0.1	126
`@marketnow/trust-adapters`	1.0.1	109
`@marketnow/trust-gateway`	1.0.1	106
`@marketnow/uts`	2.0.1	130
Total	—	2,339/mo

5 independent download channels

NPM Registry — primary
jsDelivr CDN — free global mirror of NPM
unpkg CDN — alternative CDN mirror
marketnow.site — AliceLabs-owned origin
GitHub org — alicelabs-llc/universal-trust-adapter

All serve byte-identical tarballs (SHA-256 verified).

Install

# ATC SDK
npm install agent-trust-card@1.1.2

# MCP server with 13 trust tools
npx -y marketnow-mcp@1.10.1

# Multi-source installer
curl -fsSL https://marketnow.site/install.sh | bash

Try it

# Verify any ATC card
npx -y agent-trust-card verify card.json

# Run the MCP server (works with Claude Desktop, Cursor, Cline, Continue, Aider)
npx -y marketnow-mcp

What's next

Multi-sig for high-value agents — N-of-M CA signatures (ATC v3.0, RFC Draft 00)
Runtime tool-catalog pinning — catch tool-description-poisoning attacks
Behavior-based detection layer — post-exec filter that inspects actual tool results
Cross-language SDKs — Python, Go, Rust implementations

DEV Community