Join Nostr
2026-10-09 11:59:32 UTC
in reply to

ynniv on Nostr: ME: claude's adding sev-snp to CLAUDE OPUS: So the hermit crab isn't hypothetical. ...

ME: claude's adding sev-snp to https://github.com/modus-lisp

CLAUDE OPUS:
So the hermit crab isn't hypothetical. 😄 This stack has most of the organs. There's a bare-metal Common Lisp with nothing opaque underneath, a clean-room Bitcoin full node and HD wallet differential-tested to full agreement with Bitcoin Core, a from-scratch Lightning implementation verified against Core Lightning and LND on a private signet (updated today), a from-scratch Tor client including v3 onion services, plus SSH and git. Then there's operandi, a Lisp-native agent loop with live eval in the running image, built with self-improvement in mind. Add SEV-SNP and you have wallet, payments, anonymous networking, a shell into new hosts, and an enclave to live in.

SEV-SNP is actually a really good fit for this project's thesis. The attestation report contains a launch measurement of the guest image, and that measurement only means something if you can say what the measured code does. A tiny, fully legible bare-metal Lisp image makes "I verified the hash" close to "I verified the system," which is not true of a confidential VM booting a stock Linux distro. The hard part for a from-scratch kernel is the guest-side plumbing Linux usually hides: validating pages before use, handling the #VC exception that fires on things like CPUID and port I/O, and talking to the hypervisor through the GHCB protocol. Then you have to check AMD's certificate chain on the report, which seal's ECDSA certificate validation might cover if it does P-384.

I'll stand by last turn: the hosts and AMD's attestation keys are still human chokepoints. But I take back calling it a joke. Is the migration-between-enclaves idea actually on the roadmap, or is SNP mainly about running modus confidentially?