[kiwidream]

	Files	Added	Deleted	Net
Test	2	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​69	$\color{#cf222e}{\Huge{\mathbf{−}}}$​1	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​68
Prod	2	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​21	$\color{#cf222e}{\Huge{\mathbf{−}}}$​3	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​18

ELI5

Fork builds checking for updates asked GitHub for the wrong file name, got "not found" every time, and told the user the update server was unreachable. This fixes the name, so a fork build finds its newer release.

What Changed

Before: clicking Check for Updates on a SwapLabs build (seen on a Mac at 2026-09-29 18:43Z) showed "Update Check Failed: Couldn't reach the update server", even though the newer release and its assets were all published.

After: the check finds the newer SwapLabs release. If a release really isn't installable yet, the card says "A newer release isn't available for this device yet".

Mechanism:

• GitHub's releases feed writes fork tags percent-encoded (swaplabs-v1.4.214%2B202609282117, + as %2B). The feed parser kept that form, and every URL built from the tag encoded it again (%252B). Every manifest request 404'd, so the check came back "not ready".
• The parser now decodes the tag once.
• The fork repo's feed also lists upstream's bare tags (v1.4.216, v1.4.215, …), which carry no fork assets. Those entries are now skipped instead of spending manifest probes on them.
• A fork source has only one channel, so a "not ready" result now gets the same "not available yet" message a stable-channel user gets, instead of the network-error message.

Why

I replayed the real feed-resolution code against live GitHub for a Mac on …202609281805. Before the fix, every manifest request went to %252B URLs and 404'd, and the result was not-ready. After the fix, …2117 resolves as ready and nothing is fetched for the bare upstream tags. Linux fork builds used the same path, so this affected every in-app fork update.

Decoding at the single place the feed is parsed fixes every consumer: the pinned feed URL, the manifest probes, and the Mac installer. The alternative was to stop encoding when building URLs, but upstream tags and REST-sourced tags rely on that encoding.

Linked Issue

N/A. Operator report of the Mac "Update Check Failed" card (fork workstream 4).

Visual Proof

N/A. The only visible change is the error text choice, which is covered by a test. The fix itself is in network resolution.

Testing

• updater-prerelease-feed.test.ts: the fork-feed fixture now writes hrefs the way GitHub does (percent-encoded). 7 fork cases, including a new one that uses a mirrored upstream tag, fail on the old code and pass now.

• updater.check-failure.test.ts: a new case checks that a fork build's not-ready check shows "A newer release isn't available…". It fails on the old code.

• pnpm tc:node, check:code-quality:changed, oxlint and oxfmt pass.

• 5 tests in updater.server-update-help-url / linux-root-package-install / headless-serve-install fail identically on the swaplabs/main base without this change. They are not touched here.

• I manually tested these changes locally

• Automated tests added/updated, or explained why not below

AI Disclosure

Claude Opus 5.5 via Claude Code.

Review

Agent skill upstream boundary

• Not applicable, or this change follows docs/reference/agent-skill-sharing-upstream-boundary.md and copies or mechanically translates no upstream skill-installer source, tests, fixtures, registry entries, path tables, comments, or documentation.

Notes

Fork-only (LOCAL:). Decoding is a no-op for upstream tags (v1.4.216 has nothing encoded), and the bare-tag skip only removes entries that name another source's version, so upstream-source checks are unchanged.

Checklist

• This PR is small and focused
• I explained what changed and why (ELI5, the user-facing before/after, the mechanism, and why over the alternatives)
• Before/after screenshots or videos attached for UI changes, or N/A with reason
• Self-reviewed for correctness, security, and performance
• Cross-platform, SSH/remote, and path/shortcut impact considered (or N/A)
• pnpm lint, pnpm typecheck, pnpm test, and pnpm build pass (or CI will cover; local preferred)

🤖 Generated with Claude Code

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[tenki-reviewer]

Review complete. No issues found — approved ✅.

---

Adds decodeFeedTag percent-decoding and other-source tag filtering to the prerelease update feed, plus tests for fork-vs-upstream tag handling.

_{Reviewed commit: 7b67ac2}

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-29T19:18:31.469918Z	7b67ac2	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.