Blog.
Engineering notes, threat-model walkthroughs, and ecosystem commentary
from the team building DeFi infrastructure for AI agents on Base.
Posts
A trust gate's worst failure is the one that looks like success — a silent
allow on a token nobody screened. The silent-allow bugs we
caught in our own gate (one in our own published tutorial), the enforced
fail-closed contract that prevents them now, and why composition with honest
failure semantics plus signed verdicts beats a raw single-source call. We
don't claim a secret detection edge — we're precise about what we do.
fail-closed
trust-layer
transparency
PaladinFi OFAC trust-check responses are Ed25519-signed, and now you
can verify it yourself — pinned public key, the canonical-JSON contract,
and copy-paste verifiers in Python and JavaScript at
/docs/verify-responses. Plus an
honest scope: what signing catches (in-transit tampering of the verdict)
and what it doesn't (us-compromise, replay).
ed25519
signing
agent-security
Two new evaluation paths for builders: anonymous OFAC SDN screen at
/v1/trust-check/ofac (no signup, real data, rate-limited)
and a free-tier API key at paladinfi.com/signup
with 1,000 requests/month at zero cost across /v1/quote +
/v1/trust-check. Plus a tour of the auth stack — IAM
Postgres, SSM peppers, kill switch — and an honest list of what's
deferred to v0.11.78a-followup.
free-tier
ofac
signup
Stop your Coinbase AgentKit agent from signing swaps against poisoned
token addresses. Drop-in ActionProvider, ~8 lines, free preview mode on
Base — OFAC + GoPlus + Etherscan source verification + anomaly heuristics.
Includes the hard-coded-constants pre-sign safety guarantee and an honest
limitations section.
agentkit
x402
agent-security
What changed between the internal v0.2.0 candidate and the public v0.3.0
+ v0.3.1 release: typed-domain digest, explicit chainId binding, multi-RPC
quorum on the server, and seven other items the 3-adversary Security
audit surfaced. Plus what we didn't ship and why.
trust
contract-change
ship
The v0.11.71 swap-router defense-in-depth: outer router whitelist, outer
selector allowlist, and inner-target decode + Settler validation. Why the
third layer is what binds the first two to the actual swap implementation
contract, what the security-review framing was, and how to verify the
defense is live.
calldata
defense-in-depth
Section-by-section walk through the 2026-05-07 homepage rewrite — the
failure-mode lede, JSON in both product cards, and the four
operational-posture rows (refresh cadence, fail-closed contract, rate
limits, custody) that name what most DeFi infra leaves unstated.
homepage
operational-posture
A complete walkthrough from claude mcp add through a signed
USDC → WETH swap on Base. Covers the three MCP tools
(swap_quote, trust_check_preview,
swap_health), the non-custodial sign-and-submit handoff,
and the design reasoning behind a deliberately small surface.
mcp
agent-flow
How v0.11.73 closes a silent-allow vector on the trust gate that
existed across the v0.11.5x patch window, what the new schema looks
like, and what agent code should retest.
trust
contract-change