Development Update — March 20

Skycoin: Newcoin Templates and Fibercoin Tooling

newcoin templates subcommand — the embedded code generation templates can now be printed to stdout with skycoin newcoin templates, so users can inspect, customize, and feed them back with --template-dir. The --template-dir flag, which was previously dead code, now actually works to override the embedded templates (#2821).

fiber.toml defaults — noisy peer connection log messages were suppressed, and fiber.toml default values were updated to be more useful out of the box.

CI tests — integration tests were added for the newcoin workflow to catch regressions in template generation.

Windows release fix — the Windows installer was expecting a separate skyhw.exe binary that no longer exists (hardware wallet tools are now part of the unified binary). Fixed to match the new binary layout (#2822).

Node.js 24 — GitHub Actions were updated to force Node.js 24 execution, and action versions were bumped to Node.js 24-compatible releases (#2824).

Skywire: Route Multiplexing

The route multiplexing work landed today — one of the most significant architectural changes to Skywire’s transport layer.

Accept loop crash fix — the accept loop in serveRouteGroup would permanently die when encountering a stale route with a missing transport, preventing all new connections. Fixed to distinguish fatal errors from per-route errors (#2230).

Mux layer extraction — all multiplexing state (sequencing, reordering, SACK, transport selection) was extracted from RouteGroup into a new routeMux type. The mux is lazy-initialized only when mux capability is negotiated — zero overhead for non-mux connections (#2231).

RouteGroup RPC fixes — routing.Rule (raw []byte) was causing gob serialization failures over RPC. Replaced with structured fields. Added panic recovery and length guards to prevent crashes from concurrent GC corruption (#2232).

Ping route bypass — latency probe routes were flooding the accept queue, blocking application routes for up to 30 seconds. They now bypass the accept queue entirely and are handled in dedicated goroutines with a 10-second timeout (#2233, #2234).

Embedded route setup timeout — CreateRouteGroup in the embedded transport setup had no context timeout, causing proxy connections to hang indefinitely when the remote visor was unreachable. A 30-second timeout was added. Route keepalive and setup timeouts were also increased for high-load visors: keepalive to 2 minutes, GC interval to 10 seconds, handshake timeout to 30 seconds (#2227).

TPD bandwidth storage fix — the Transport Discovery was storing bandwidth as a single total instead of tracking sent/recv separately per edge. Fixed to track both directions independently, which is needed for accurate bandwidth-based reward calculations (#2227).

Reward system UI — the reward system interface was updated with new navigation and layout (#2227).

DMSG: Server Fallback and 32 Bug Fixes

A landmark day for DMSG reliability:

Server fallback — DialStream was returning on the first server failure without trying alternatives. Now both existing-session and new-session phases fall through to the next server. Additionally, FallbackRoundTripper was consuming the request body on the first attempt, causing subsequent retries to send empty bodies — fixed by buffering the body upfront (#347).

32 bug fixes in one PR — a comprehensive audit found and fixed 32 bugs across the DMSG codebase:

• Noise protocol: replay attack vulnerability (nonces never recorded), TCP connection leaks, listener connection leaks, handshake goroutine leaks
• Discovery client: entry copy corruption, error variable shadowing, HTTP connection leak
• dmsgcurl: response body leak, division by zero, retry logic bug
• dmsgpty: data races on global state, zombie processes (Unix), ConPty handle leak (Windows), WebSocket data discard
• dmsgctrl: concurrent write corruption, data race in Close/Err
• DMSG Discovery server: inverted nil check, response body leak, nil dereference on hostname input
• DMSG Server: deadlock from unbuffered error channel, listener close race, data race in metrics

Comprehensive tests — test coverage was added across the DMSG packages (#348).

This was the single biggest reliability improvement to the DMSG library in its history. Many of these bugs were latent — they’d only trigger under specific timing conditions or error paths — but collectively they made the network fragile under load. Fixing them makes DMSG substantially more reliable for production use.