20 November 2026 · source under review 71d0c41d2d7726a1dae3bf2fdba3b2a4b8cac6cf. The Windows client and desktop are 0.2.2. Edge and the shared protocol crate are 0.2.0. The assessment below covers this checkout and the local Windows release files. The live observations describe the named OpenBYL endpoints at review time. They do not establish which binary revision is installed on the VPS. Workspace manifests[1] · Beacon version[2] · Desktop version[3]
Release decision
Account pilot verification
On 20 November 2026, two independent Windows Beacon 0.3.0 processes on one laptop signed in through separate browser sessions using the same test account. Each received a distinct device identity and published a separate loopback app through the production VPS and the proxied Cloudflare wildcard. Both public HTTPS addresses returned fresh values from their respective apps.
Signed-out requests to private routes received HTTP 401. Each route accepted its own device's browser or API credential, and the local app received no gateway Authorization header. Switching one hostname to public and back left the other private. Revoking one device disconnected its tunnel while the second still served requests. The test resources were retired afterward. The signed-in installer download matched the local release by SHA-256. The Windows upgrade on the existing studio profile preserved the Burro route and its public health check passed. Two physical laptops, a clean Windows install under application control, and a disaster-recovery rehearsal remain outside this run. Windows account setup
Original release assessment
Continue the trusted-device pilot. Hold a public Windows RC. Real processes passed the forwarding, enrollment isolation, reconnect, and intentional-stop tests. The reviewed installer is unsigned. Clean-machine distribution, full staging recovery, and unrelated publishers sharing a browser domain each need different evidence. The following are proposed decisions for the release owner.
| Scope and owner | Evidence needed to close the decision |
|---|---|
| Public Windows package — release maintainer | Verify trusted timestamped signatures on the installer, Beacon, desktop app, and uninstaller. On a clean Windows 11 machine with its normal application-control policy, demonstrate install, import, first connection, update, and uninstall. Record the signed package's new hash and the delivery location. Signing and packaging[4] |
| Managed website publishing — Edge operator | In a staging zone, enroll two distinct PCs. Publish a fresh hostname through real DNS, Nginx and ACME. Verify a private action both signed out and signed in. Back up and restore its route, then revoke one device. Confirm its traffic stops and the other owner recovers after the relay restart. Edge runbook |
| Unrelated customers publishing sites — domain owner | Decide where customer content may live. Demonstrate that tenant sites cannot receive first-party credentials and that sensitive first-party actions enforce their origin and authorization policies. Review all static targets for control-plane exposure. Keep enrollment limited to trusted publishers until that boundary is established. Security findings S1–S4 |
| Dependencies — build maintainer | Triage the advisories below for the exact target and lockfiles. Update the affected field-guide dependency chain or record a time-bounded build-time exception with an owner. Repeat both audits for the candidate revision. |
| Capacity and operational claims — service operator | Validate the expected mix of streaming and short requests on the intended VPS, recording errors, throughput and provider transfer. If the independent nonce probe is part of release acceptance, restore its local origin and capture a fresh matching public response. Publish capacity or recovery targets only after measurement. Workload limitations |
Tests run against this revision
| Check | Result and coverage |
|---|---|
| Rust formatting, Clippy, locked build, workspace tests | Passed on Ubuntu under WSL with Rust 1.100.0. Clippy ran with warnings denied. All 18 unit tests passed, covering configuration, dashboard guards, telemetry, session replacement, publishing state and framing. Windows-only GUI behavior is outside the Linux test binary. Workspace[1] |
| Transport process smoke | Passed all seven reported checks. Real Edge and Beacon processes carried HTTP, concurrent binary and half-closed streams. Two enrolled devices kept separate routes. Identity failures were rejected. Reconnection and Stop/Start restored fresh requests, including after a saved stop and a canceled retry. Transport smoke test[5] |
| Managed publishing | Passed. Real Edge and Beacon processes registered and forwarded routes, isolated two devices, replayed offline requests, and recovered after restart. The Linux publishing helper was a test double in this suite. Five separate helper tests passed with DNS, Nginx and certificate interactions isolated. Publishing smoke[6] · Helper tests[7] |
| Windows x64 setup | Passed on this PC using an isolated 0.2.2 installer built from the release binaries. Exercised first install with deferred enrollment, profile import, shortcut choices, publishing upgrade, reinstall, and identity-preserving uninstall. The isolated harness does not launch the connector. Installer test[8] |
| Nuxt guide | Generated the static homepage in a fresh directory from the tracked source and lockfile, using Node 26.13.1, npm ci, and npm run generate. npm reported an install-script policy warning for esbuild. Nitro warned about an unresolved cache-driver import. The output included .output/public/index.html. Server-rendered operation was outside this check. Guide build instructions[9] |
Public endpoint observations
These ordinary read-only requests used the existing PC and deployment. The review did not restart the relay, change DNS, or sign in with a Burro key.
| Observation | Result | Interpretation |
|---|---|---|
| Local Beacon dashboard | Online, connection enabled | The connector had an accepted Edge session. |
https://edge.openbyl.com/ |
200 | The VPS-hosted guide was reachable. |
https://aburro.openbyl.com/api/health |
200, name=burro, ok=true |
The public Burro route returned the expected app response. |
https://aburro.openbyl.com/api/status without cookies or a key |
401 | Burro denied the signed-out status request. Signed-in operations need a separate acceptance test. |
Fresh /_tunnel/probe/<nonce> on the guide hostname |
502 publicly. The connection to 127.0.0.1:24781 was refused on Windows. |
The optional proof origin was unavailable. This run could not compare a public nonce and proof to the local copy. Burro's health check covered a different route. |
Use the fresh-nonce procedure for a future run. It checks both the random challenge and the proof value, so a cached homepage cannot satisfy it. OpenBYL verification procedure[10]
Dependency and artifact review
Rust. cargo-audit 0.23.0 checked 386 locked dependencies against RustSec database revision f291af9b1cc0d21dfcea1709d6b037963cef262b. It reported zero vulnerability entries and two warnings. The warnings were for unmaintained proc-macro-error 1.0.4 and an unsoundness advisory for glib 0.18.5. Neither package appeared in the resolved dependency trees checked for Windows GNU x64 Beacon Desktop or Linux GNU x64 Edge. Recheck for any additional release target. RustSec advisories · proc-macro-error advisory · glib advisory
Nuxt field guide. The public npm registry reported eight moderate package entries, all traced to the same minimatch advisory. It reported zero high or critical entries. Affected dependencies participate in the Nuxt toolchain. The documented VPS deployment serves generated static files with Nginx. Browser requests to that deployment do not invoke the Nuxt build tools. Evaluate the build environment separately. minimatch advisory · Guide deployment[9]
Windows. Authenticode returned NotSigned for the 0.2.2 installer and the installed beacon.exe and beacon-desktop.exe. The following SHA-256 values identify the reviewed local artifacts. A checksum alone cannot establish their publisher. Recompute them after signing or rebuilding. Windows signing guidance
Artifact in dist/release/ |
SHA-256 |
|---|---|
Beacon-Setup-0.2.2.exe |
B927AD15429135D4C9576171A1C105989E802EDBCC3F53321DE88DD8DC90931C |
Beacon-Portable-0.2.2-x64.zip |
727CBF6A5F28D9D41AF51F36771FBAABED4BD94C2FC5A3CB74C0C59E7FCCF76A |
Repeat the checks
From the candidate checkout on Linux, use Python 3.11+, OpenSSL, and the Rust toolchain with rustfmt and Clippy. Setting the build output explicitly makes both smoke scripts find the same binaries. These processes use disposable identities and loopback ports.
set -e
cargo fmt --all -- --check
cargo clippy --workspace --locked --all-targets -- -D warnings
cargo test --workspace --locked
cargo build --locked --target-dir target
python3 scripts/demo.py --verify --dashboard-port 0 --bin-dir target/debug
python3 scripts/publish_smoke.py --bin-dir target/debug
python3 scripts/test_publisher.py
Run cargo audit --file Cargo.lock with RustSec's cargo-audit installed. Run npm audit --package-lock-only --ignore-scripts in sites/tunnel-n-here-we-are/. Record the audit date, advisory identifiers, target platform and disposition. For Windows, use the isolated installer procedure before the clean-PC acceptance test. cargo-audit · Isolated Windows installer procedure[11]
Repository references
Cargo.tomlcrates/beacon/Cargo.tomlcrates/beacon-desktop/Cargo.tomlpackaging/windows/README.mdscripts/demo.pyscripts/publish_smoke.pyscripts/test_publisher.pypackaging/windows/test_installer.pysites/tunnel-n-here-we-are/README.mdREADME.md#operate-and-verifypackaging/windows/README.md#exercise-setup-without-affecting-a-live-connector