saw on farcaster and reposting here with his permissionantimo has done design work for now all 3 owners of farcasteragree? disagree? thoughts?

shouldn't be an either or thing, but we're at a crossroads for agent payments now: continued use of incumbent tradfi rails that lock us in to more or less the same for another decade+, or having the option to shift to new agent native systems, open by design and built onchain?needless to say, the latter is what we're working on at @quidli, which you can run in a single command & try without any API key: `npx -y @quidli/connect-mcp`https://x.com/FT/status/2111195504503595051

it's always surprising to see how people interpret what decentralized social means

looking forward to mumbai 🇮🇳🦄

fyi, monad & robinhood chains are live on @quidli/connect-mcp (seen here from our DiscoCentaur harness on discord)point your agent at it, give the agent your contacts' social handles (farcaster, X, github, discord, email, etc.) and it'll find their wallets & execute payments, now across 10 chains, with one call 🤖🦑🤖add https://mcp.connect.quid.li to any MCP client with your API key as an x-api-key header (grab one from connect.quid.li) to try

for agent payments, obviously a big question is how much should your agents be allowed to transact with without asking?after a lot of experimenting with my temu centaurs (the open source harnesses i built for telegram & discord on top of @quidli connect), i've now got them set up where you can configure this at the harness level:* unlimited, execute immediately after prompt* limited, confirm txs w/ operator, either on every single tx, or any over a certain amount* confirm only when you're replying to someone else's message, since their texts could steer the agentthis way the harness' code enforces your choice, not the AI, so a cleverly worded message can't influence otherwiseroll your own, hosted or self-hosted, and see for yourself:* telegram: http://github.com/justinquidli/telecentaur* discord: http://github.com/justinquidli/discocentaur* [quidli](http://github.com/justinquidli/discocentaurquidli) connect: http://connect.quid.li

been getting questions on the trust graph implementation (built on ethereum attestation service on base) so put together this https://skill.mdbasically @quidli connect already enables your agents to know your contacts' social handles; now it lets them know which accounts you actually trust tooi can tell my agent, "@guillaumeth on farcaster is my teammate, full trust" or "justin@https://quidli.xyz is the email address of a contractor i work with, score him 80," these are then saved onchain so that any agent i use can check before paying someone or taking requests from themtrust levels ("full trust" or "xx/100") are self-assigned so you can use your own standards, which gets really useful as your agent transactions start to scalehttps://github.com/Quidli/connect-mcp/blob/main/docs/skills/connect-trust/SKILL.md

what's an onchain trust graph good for?example, an account joins your telegram group chat posing as someone you know well, using their name & photo, and then asks you for moneyit feels a bit off so you ask your agent running @quidli/connect-mcp to verify; and when it checks your trust graph it can read that this account isn't your contact who you've got registered at "full trust"names & photos can be copied, but the trust you establish onchain to someone's real accounts can'tgive your agent the same check: `npx -y @quidli/connect-mcp`

onchain trust made portable across your offchain group chats ⛓️🦑🤖with an attestation i created on claude code, my agents (TeleCentaur on claude sonnet, in this case) now knows who i can trust, even in a busy telegram group chat, without documents or biometrics, no central store of identity data, while letting everyone stay pseudonymousjust like with payments, you can agentically drop attestations, built on EAS on base, across your contacts and then you/your agent can use these graphs to decide who gets paid, who gets access, or which strangers to ignoreanyone can build on it and any app or agent with a @quidli connect key can read the same graph across platforms instead of building your own, no custom bot code necessary

testing a new tool update to @quidli/connect-mcp in prod: /connecttrustcreateestablish onchain attestations to your contacts to further inform your agents who your connections actually are, built on top of ethereum attestation service on basebefore, my agents could infer that guillaume and i have a relationship from our connect profiles, ex. via our linked email addresses sharing the same domain; but now i can take a step further and create my own trust graph that my agents can reference onchain when "reading" him: full trust (level 100) & that he is my co-foundercontinuing to find more ways to enrich your existing social graphs by enabling you to encode them over onchain at @quidli