[cu1ch3n]

What

• Version 0.10.0 in package.json and package-lock.json.
• CHANGELOG.md: the Unreleased section becomes [0.10.0] - 2026-11-12, with entries and links for every user-visible change since 0.9.4 (Remove fixed deadlines from frontend API requests #106–Offer Refresh when Codex loses its sign-in #137), a new Removed section (Retire the standalone review skill #127, Drop tasks that accept straight into the manuscript #128, Remove the empty vendor/latex-review package #132), and compare links for 0.9.1–0.10.0.
• docs/releases/0.10.0/release-notes.md, which the Release workflow uses as the GitHub release body: jumping from the PDF to its LaTeX (Edit the manuscript's LaTeX from its PDF #119), reviewing without waiting for TeX (Never make the author wait for compilation #112), the phone layout (Redesign the mobile UI with Apple-native hierarchy and touch ergonomics #118), the rest of the release, and what upgrading from 0.9.4 means (Codex installed on the computer for the desktop app, paper settings moving into .git/papilot, old write-through tasks, reinstalling from 0.8.2 or earlier).

How it was checked

• Below this commit, main at e909eff was identical to a merge of Run every browser check in CI #136, Give Codex a working sandbox for workspace tasks on Windows #135, Remove the latex-review names and keep paper settings in .git/papilot #134 and Offer Refresh when Codex loses its sign-in #137 that was checked before landing them; main has since gained only Fit the phone marking tools on narrower iPhones #142 (CSS for the phone marking bar). That merge passed: typecheck, both i18n checks, root unit tests (922: 908 passed, 14 skipped) and engine tests (641 passed) on macOS with Node 22.
• On a Windows 11 machine (Chinese locale, Git with core.autocrlf=true, TinyTeX): npm ci, typecheck and build pass. The remaining unit and engine test failures come from that machine (no symlink privilege, no POSIX sh on PATH, TinyTeX without beamer/KOMA/xeCJK, UTF-8 .cmd fixtures under code page 936); they pass on Windows CI. One is a real, older issue, and 0.9.4 has it too: staging a single change from the PDF comparison fails when Git converts line endings (autocrlf=true).
• Upgrade on that Windows machine with the packed 0.10.0 tarball: a paper opened and built by papilot@0.9.4, then 0.10.0 started on the same data folder. 13/13 checks: state moved to .git/papilot with a junction left at .git/latex-review, the earlier PDF reachable at its recorded path, refs/latex-review gone, .latex-review.json settings copied into .git/papilot/settings.json and the file left unchanged, no new files in the working tree, the manuscript PDF shown, no 0.9.4 process left.
• The release workflow's own checks: the tag matches package.json, and CHANGELOG.md has a [0.10.0] section.

Release

Push v0.10.0 on this commit; once the release and its installers exist, fast-forward main to it.