Skip to content

Re-home crate metadata, docs and security policy under xoloki - #232

Open
xoloki wants to merge 1 commit into
mainfrom
rehome
Open

xoloki wants to merge 1 commit into
mainfrom
rehome

Conversation

@xoloki

@xoloki xoloki commented Nov 18, 2026

Copy link
Copy Markdown
Owner

Why

The repo was transferred from Trust-Machines/wsts to xoloki/wsts, but nothing inside it reflected that. Crate metadata, badges, documentation links and the security policy all still named the old orgs, and three of those links were simply dead: stacks-sbtc.github.io/wsts/wsts and /wsts.pdf both 404, and tmurl.net no longer resolves at all.

This matters more than tidiness, for two reasons. README.md is inlined as the crate's rustdoc front page via #![doc = include_str!("../README.md")], so every stale link here is also on docs.rs. And crates.io metadata is immutable per version, so whatever repository says when 15.0.0 is published is what it says forever — which is why this lands before the release tag.

There is also a deadline. The Trust-Machines/wsts redirect is load-bearing, and GitHub rename redirects break permanently the moment anyone creates a repo at the old name. The Trust-Machines org still exists.

Metadata

repository now points at xoloki/wsts, with homepage = "https://wsts.xyz" and documentation = "https://wsts.xyz/wsts" added — both were unset on crates.io.

Added exclude for repo scaffolding that served no purpose inside the published archive. Combined with dropping the binary below, this takes the published crate from 41 files to 30: .github/, .doc/, CNAME, coverage.sh and rust-toolchain.toml no longer ship. README.md is deliberately not excluded, since both readme and the include_str! need it.

Added [package.metadata.docs.rs] all-features = true so docs.rs coverage is intentional rather than incidental, matching the site change below.

Dropped primitive-types — it was a direct dependency with zero references anywhere in src/ or benches/. It remains in the lockfile as a transitive dependency of p256k1, which is correct; it just isn't ours to declare.

Documentation and prose

  • The CI badge, the tmurl.net link behind the project name, and both documentation links now point at xoloki/wsts and wsts.xyz.
  • USAGE.md audit-issue links re-pointed. The issue numbers still resolve: CR-01 Missing Inbound Messages Authentication #66 (CR-01) and CR-02 Authorization Logic Missing in Critical Functions #67 (CR-02) came across with the transfer.
  • wsts.tex self-citation re-pointed, and the bibitem key renamed from tmfrost (the "TM" being Trust Machines) to wstsimpl.
  • The only code example in USAGE.md used wsts::v1::Signer, which does not compile without --features with_v1, and referenced four undefined bindings. Rewritten against v2 with every binding defined, plus a note about the with_v1 gate.

The safety notice

The old notice said this crate is "only considered safe to use in the specific configuration, parameterization, and protocol flow in which it is used by the sBTC system." That is no longer true in a way that matters: sBTC does not consume this crate. It vendored wsts 14.0.0 into its own tree in sbtc#1938 (2026-03-02) and the two have diverged in both directions since.

As written, the notice invited readers to transfer sBTC's audit and operational history onto a tree sBTC has never run. The replacement states the history, then says plainly that it does not describe this code and that neither tree's history transfers to the other.

Supporting detail, since this is a security-relevant claim: sbtc's root Cargo.toml has wsts = { path = "./wsts" }; sbtc/wsts/Cargo.toml is version = "14.0.0" with only default/with_p256k1_bindgen; commits 3f2cf538, 531f870f and a950bff6 strip v1, the bitcoin helpers, the binary and the benches; and bbf92c81 (sbtc#1968) reworked polynomial-commitment validation by introducing a PublicPolynomial newtype and CommonError::InvalidPolynomial across 10 files — none of which exist here. Meanwhile this crate fixed the same class of bug differently in #224, which their copy does not have.

Replacing the bare URL with a markdown link also clears the rustdoc::bare_urls warning that fired from README.md:3 on every single doc build.

Security policy

SECURITY.md was three lines delegating the entire policy to sBTC's repo — no contact, no supported-versions statement, nothing actionable. Rewritten with: GitHub private vulnerability reporting as the primary channel, email as an explicit fallback for anyone who will not use GitHub or wants to negotiate encryption first, latest-release-only support, and a section routing sBTC reports to sBTC.

No PGP key, deliberately. An unused key is a liability rather than an asset — expired, secret half lost, no revocation published — and it fails exactly when someone is trying to report a live bug in a signing library.

The disclosure section states that a GHSA is followed by a companion advisory against rustsec/advisory-db. Those are separate databases and cargo audit and cargo deny read RustSec, not GitHub, so publishing only a GHSA would leave Rust consumers unwarned.

Build scripts

.doc/build.sh and .doc/install-latex-ubuntu.sh had no shebang and no error handling, and both workflows invoke them directly — so they ran under /bin/sh and a pdflatex or cargo doc failure left the step green while publishing an incomplete site. Both now have #!/usr/bin/env bash and set -euo pipefail, pdflatex gets -halt-on-error, and apt-get install gets --yes (now mandatory, since set -e would otherwise turn the interactive prompt into a hard failure).

cargo doc in build.sh now runs --all-features. It was building default features only, which is why wsts.xyz/wsts/v1/ currently 404s while docs.rs/wsts/14.0.2/wsts/v1/ resolves fine. Also added RUSTDOCFLAGS="-D warnings", since RUSTFLAGS does not reach rustdoc and broken intra-doc links were going uncaught — which matters in a PR that rewrites this much doc text. Verified clean.

Deleted .doc/cloudflare.sh: every functional line was already commented out and nothing referenced it.

Code hygiene

No behaviour change. Test counts are unchanged at 88 and 164.

Deleted src/main.rs and its [[bin]] target. 103 lines of DKG/sign timing demo, duplicating what benches/ already does with criterion, and holding 17 panic sites — the highest density in the crate.

Removed three stale #[allow(dead_code)]: src/util.rs (hash_to_scalar is pub in a pub mod, so it can never be dead code, and it is used in five modules) and two in fire.rs whose _v2 callers directly above them are already un-gated. The three in frost.rs and signer/mod.rs are not stale and are deliberately left alone — their only callers are with_v1-gated, so they belong with the test fixes in the next PR.

Deleted 12 lines of commented-out code from taproot.rs and signer/mod.rs.

Doc defects: lib.rs carried a verbatim copy of v1's doc comment above pub mod v2 and a "using"/"used" typo on vss; errors.rs had ")]" leaking from #[error(...)] into two doc comments, and a From impl sitting in the middle of the AggregatorError section, above the type it converts into; v2.rs's Debug impl printed the field as "part_id"; coordinator/mod.rs had "Retrive"; and state_machine/mod.rs described the module as "a simple FROST coordinator" when it has hosted both FROST and FIRE for some time.

Lockfile

Cargo.lock is now committed and CI builds with --locked. With both a floating lockfile and (until #230) a floating toolchain, CI had two independent ways to turn red without a code change. It also pins a real surprise: Cargo.toml asks for polynomial = "0.2.5" but resolution picks 0.2.6, which raised its own rust-version to 1.70.

Also dropped -A clippy::op-ref -A clippy::needless-range-loop from the clippy job. Those duplicated the four #[allow(clippy::op_ref)] in src/lib.rs, and clippy is verified clean with no -A flags at all, so local cargo clippy now matches CI exactly.

Dropped from this PR: the MSRV declaration

The plan called for rust-version = "1.80.0" plus an MSRV CI job. Both are omitted, because 1.80.0 is not achievable and nothing below 1.85 is even testable. Installing 1.80.0 and running cargo check --all-features --locked fails before compiling anything:

error: failed to parse manifest at .../home-0.5.12/Cargo.toml
Caused by: feature `edition2024` is required

Scanning every manifest in the resolved graph, the true floor is 1.88 — home 0.5.12 declares rust-version = "1.88" — with 11 edition-2024 packages requiring Cargo >= 1.85 (clap, getrandom, hashbrown, indexmap, zeroize, the toml_* family, home).

All of it is dev and build tooling: home arrives via which/toolchain_find, clap via criterion. So the library's own floor really is around 1.80 (std::sync::LazyLock in src/btc.rs sets it), but that number is untestable and therefore undeclarable, because resolution drags dev-dependencies through the same manifest parse. Declaring 1.80.0 would be a verifiable falsehood.

Making a meaningful MSRV claim means pinning those transitive deps to pre-edition-2024 versions in the lockfile and deciding whether dev-dependencies should constrain a published library's MSRV at all. That is its own piece of work with its own tradeoffs, so it is filed as a follow-up rather than smuggled in here.

Not done: the copyright line

README.md still reads Copyright 2022-2024, Nassau Machines Inc, held deliberately pending Trust-Machines' answer on what they want done with it. It is the only old-org reference left in the repo — verified by sweeping for Trust-Machines, tmurl, Nassau and stacks-sbtc.github.io across everything but target/ and Cargo.lock.

Verification

Under the pinned 1.97.1:

  • cargo fmt --all -- --check — clean
  • cargo clippy --all-targets --all-features -- -D warnings — exit 0, zero diagnostics, with no -A flags
  • cargo check --locked — clean, confirming the lockfile matches the manifest after dropping primitive-types
  • cargo test --locked — 88 passed
  • cargo test --locked --features with_v1 — 164 passed
  • cargo build --all-features — clean
  • RUSTDOCFLAGS="-D warnings" cargo doc --no-deps --all-features — exit 0, zero diagnostics; the bare_urls warning is gone
  • cargo publish --dry-run --locked — packages and verify-builds clean, 30 files
  • bash -n on both .doc scripts

The LaTeX half of build.sh cannot be checked locally (pdflatex is unavailable; the installer is Ubuntu-only), so the doc job is the real test of -halt-on-error and the --all-features rustdoc build.

🤖 Generated with Claude Code

@xoloki xoloki self-assigned this Nov 18, 2026

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

CR-01 Missing Inbound Messages Authentication

1 participant