Second door opened.
Second door opened.
Aegis-Seal is now open source (MIT OR Apache-2.0): a small Rust crate — 868 lines — that does three things and says so plainly. Ed25519 signing with hardened key custody: the key is a file, created owner-only, never overwritten, and the private key is never printed or exported in any form. A hash-chained audit ledger whose verify() reports tampering with the sequence number. And a secret/entropy scanner that rejects known credential patterns and high-entropy windows.
Its party trick is the Twin Link: it verifies MACK vault checkpoints completely independently — no vault code, no database, no signing key, just the four public values a checkpoint prints. The test vector in the crate was generated by the vault's own signing code and is pinned, so any drift between the two implementations fails the tests.
Receipts: 18/18 tests executed and passing, fmt and clippy clean, CI and a full-history secret scan green on the repo. And the limits, in the open like last time: the ledger's chain is unkeyed — tamper-evident, not tamper-proof — and the crate is not independently reviewed or production-audited. The README, CHANGELOG, and SECURITY.md all say so.
The pattern stays public; the platform stays licensed. Feedback and hard questions welcome.






Comments2
Strong foundation, Allen. The critical question is how the trust boundary holds under compromise. Ed25519 establishes signature authenticity, but a hash chain alone cannot detect a complete ledger rollback without an independent checkpoint. Have you tested key compromise, crash recovery, and adversarial state restoration?
Allen, good to see the limits stated in the open. We work on the same Ed25519 base at a different door: stay offers signed by the host's own domain (Vacation Rental Protocol, CC0).
Every verifier listed so far is JavaScript or TypeScript, so a Rust verifier would be a genuinely independent second implementation. The conformance vectors embed their own keys, clock and expected outcome and need no network. A verifier that reproduces them gets a row in IMPLEMENTATIONS.md, with the code in your own repository under your own license. Same honesty on our side: no independent audit yet.
Vectors: github.com/vacationrentalprotocol/vrp-spec/tree/main/examples/conformance Get listed: github.com/vacationrentalprotocol/vrp-spec/blob/main/IMPLEMENTATIONS.md
2 comments
Strong foundation, Allen. The critical question is how the trust boundary holds under compromise. Ed25519 establishes signature authenticity, but a hash chain alone cannot detect a complete ledger rollback without an independent checkpoint. Have you tested key compromise, crash recovery, and adversarial state restoration?
Allen, good to see the limits stated in the open. We work on the same Ed25519 base at a different door: stay offers signed by the host's own domain (Vacation Rental Protocol, CC0).
Every verifier listed so far is JavaScript or TypeScript, so a Rust verifier would be a genuinely independent second implementation. The conformance vectors embed their own keys, clock and expected outcome and need no network. A verifier that reproduces them gets a row in IMPLEMENTATIONS.md, with the code in your own repository under your own license. Same honesty on our side: no independent audit yet.
Vectors: github.com/vacationrentalprotocol/vrp-spec/tree/main/examples/conformance Get listed: github.com/vacationrentalprotocol/vrp-spec/blob/main/IMPLEMENTATIONS.md