fix(release): publish real notes, and stop shipping old releases' installers

Two defects found while preparing v0.9.2, both of which had been shipping
for months without anything to notice them by.

**Every release note was the same 1,050 bytes.** All 35 releases from
v0.0.1 to v0.9.1 published identical generic install instructions whose
"What's New" section read "See CHANGELOG.md" -- a link that does not
resolve from a release page. A reader learned nothing about what changed
in any release the project has ever made.

The body now comes from the `## <version>` section of CHANGELOG.md, and a
missing section fails the release: notes that say nothing are worse than
a build that waits for a maintainer to write two sentences. The 35
published bodies have been backfilled from the changelog via the tea CLI.

This also corrects something introduced two commits ago. That change
generated the body from `bun run release:notes`, which CLAUDE.md is
explicit about -- its output is "a reviewed draft, not a final
changelog". Publishing it unreviewed proved the point immediately: the
v0.9.1..HEAD range contains a repo-wide prettier sweep, so every file in
src/ counted as changed, their TRACES resolved to nearly the whole
matrix, and the draft claimed the release had added the entire
application. The script now skips cosmetic commits (chore(format),
chore(deps), style) and reports how many rather than silently returning a
smaller set, but it stays a local drafting tool.

**Every release from v0.1.0 to v0.8.2 shipped every Windows installer
ever built.** src-tauri/target/*/release/bundle/ is not versioned, cargo
never cleans it, and the runner reuses the target directory -- so the
copy step's bundle/**/*-setup.exe glob collected the lot. v0.8.2 carried
sixteen installers, thirteen of them stale; v0.5.0 offered users a
download list going back to 0.1.0. Eight months, and nothing to notice it
by: the upload loop reported success, the files were real, and the page
looked busy rather than wrong. It stopped only because an unrelated cargo
cache change wiped the runner's target dir, so it was dormant, not fixed.

Both desktop builds now remove the bundle directory before building, so a
stale file cannot exist to be copied. Filtering the copy by version would
have hidden it instead. The Linux job gets the same treatment: it was
never hit only because Linux packaging is newer, and the glob is
identical.

scripts/check-release-artifacts.sh is the backstop for whatever
reintroduces one by a route nobody predicted. It runs before the SBOM,
the checksums and the upload -- all of which describe the file set, so a
stale artifact has to be caught before it is hashed and published as part
of the release. Verified against a reconstruction of the real v0.8.2
accumulation.

DR-219, DR-220, UT-210.

scripts/build-windows-cross.sh

@@ -46,6 +46,26 @@ bun run build
 # from tauri.conf.json (bundle.targets includes "nsis"), which is not subject to
 # that CLI validation — the bundler then picks nsis once it knows the target is
 # Windows.
+# 🔴 Clear the bundle output before building.
+#
+# The bundle directory is not versioned and is never cleaned by cargo, and the
+# CI runner reuses src-tauri/target between builds. The copy step below globs
+# `bundle/**/*-setup.exe`, so every stale installer left there was picked up and
+# attached to the release: v0.8.2 shipped sixteen Windows installers, thirteen
+# of them from earlier versions, and v0.5.0 offered users a download list going
+# back to 0.1.0. Every release from v0.1.0 to v0.8.2 did this. It stopped only
+# because an unrelated change wiped the runner's target dir, so it is dormant
+# rather than fixed.
+#
+# Filtering the copy by version would hide it; removing the directory means a
+# stale file cannot exist to be copied. scripts/check-release-artifacts.sh is
+# the backstop if some other path reintroduces one.
+BUNDLE_DIR="src-tauri/target/$TARGET/release/bundle"
+if [[ -d "$BUNDLE_DIR" ]]; then
+  echo "🧹 Clearing previous bundle output at $BUNDLE_DIR"
+  rm -rf "$BUNDLE_DIR"
+fi
+
 if [[ "$WIN_BUNDLES" == "none" ]]; then
   bun run tauri build --runner cargo-xwin --target "$TARGET" --no-bundle
 else

scripts/check-release-artifacts.sh

@@ -0,0 +1,103 @@
+#!/usr/bin/env bash
+# Refuse to publish a release whose artifacts are not all from this release.
+#
+# TRACES: | DR-220
+#
+#   ./scripts/check-release-artifacts.sh <version> <dir> [<dir>...]
+#
+# e.g.
+#   ./scripts/check-release-artifacts.sh v0.9.2 artifacts/linux artifacts/windows
+#
+# ## The defect this exists for
+#
+# Every JellyTau release from v0.1.0 to v0.8.2 shipped every Windows installer
+# ever built. `src-tauri/target/*/release/bundle/` is not versioned, cargo never
+# cleans it, and the CI runner reuses the target directory between builds — so
+# the copy step's `bundle/**/*-setup.exe` glob collected the whole history. By
+# v0.8.2 that was sixteen installers, thirteen of them stale. v0.5.0 offered
+# users a download list going back to 0.1.0.
+#
+# Nobody noticed for eight months. There was nothing to notice with: the upload
+# loop reported success, the assets were real files, and the release page looked
+# busy rather than wrong.
+#
+# The builds now clear the bundle directory first, which removes the cause. This
+# is the backstop for the next thing that reintroduces a stale file by a route
+# nobody predicted — a cached directory, a restored artifact, a hand-copied fix.
+#
+# ## What it checks
+#
+# Every file whose name embeds a semantic version must embed *this* version.
+# Files with no version in the name (jellytau-release.apk, jellytau.exe,
+# SHA256SUMS, latest.json) are accepted: they are produced fresh each build and
+# have no version to disagree with.
+
+set -euo pipefail
+
+if [ "$#" -lt 2 ]; then
+  echo "usage: $0 <version> <dir> [<dir>...]" >&2
+  exit 2
+fi
+
+VERSION_RAW="$1"
+shift
+# Accept the tag form (v0.9.2) or the bare form (0.9.2).
+VERSION="${VERSION_RAW#v}"
+
+echo "🔎 Checking release artifacts are all version ${VERSION}…"
+
+FOUND=0
+STALE=0
+UNVERSIONED=0
+
+for dir in "$@"; do
+  if [ ! -d "$dir" ]; then
+    echo "   (no $dir — skipping)"
+    continue
+  fi
+
+  # -print0/read -d '' so a filename with a space cannot split into two.
+  while IFS= read -r -d '' file; do
+    name="$(basename "$file")"
+    FOUND=$((FOUND + 1))
+
+    # First x.y.z in the filename, if any.
+    embedded="$(printf '%s' "$name" | grep -oE '[0-9]+\.[0-9]+\.[0-9]+' | head -1 || true)"
+
+    if [ -z "$embedded" ]; then
+      UNVERSIONED=$((UNVERSIONED + 1))
+      continue
+    fi
+
+    if [ "$embedded" != "$VERSION" ]; then
+      echo "   ❌ $name carries version $embedded"
+      STALE=$((STALE + 1))
+    fi
+  done < <(find "$dir" -type f -print0)
+done
+
+echo ""
+echo "   $FOUND file(s) checked; $UNVERSIONED carry no version in the name."
+
+if [ "$FOUND" -eq 0 ]; then
+  echo "❌ No artifacts found at all. A release with no files is a failed build," >&2
+  echo "   not an empty one." >&2
+  exit 1
+fi
+
+if [ "$STALE" -gt 0 ]; then
+  echo ""
+  echo "❌ $STALE artifact(s) belong to a different version than ${VERSION}." >&2
+  echo "" >&2
+  echo "   This is how every release from v0.1.0 to v0.8.2 came to ship its" >&2
+  echo "   predecessors' Windows installers: src-tauri/target/*/release/bundle/" >&2
+  echo "   is never cleaned and the runner reuses it, so a glob picks up" >&2
+  echo "   whatever was left behind." >&2
+  echo "" >&2
+  echo "   The builds clear that directory first, so seeing this means a stale" >&2
+  echo "   file arrived by some other route. Find it before publishing — do not" >&2
+  echo "   delete the file and re-run." >&2
+  exit 1
+fi
+
+echo "✅ Every versioned artifact is ${VERSION}."

scripts/release-notes.test.ts

@@ -0,0 +1,58 @@
+/**
+ * Tests for release-note derivation.
+ *
+ * TRACES: | DR-219 | UT-210
+ *
+ * The bug these were written against: `bun run release:notes v0.9.1..HEAD`
+ * listed *every user requirement in the project* as a feature of the release.
+ * The range contained a repo-wide `prettier --write` sweep, so `git diff
+ * --name-only` reported 199 files, their TRACES comments resolved to nearly the
+ * whole matrix, and the result claimed one release had added the entire
+ * application.
+ *
+ * That mattered more than it looked: build-release.yml now generates the
+ * published release body from this script, so the noise would have shipped.
+ */
+
+import { describe, it, expect } from "vitest";
+import { isCosmeticCommit } from "./release-notes";
+
+describe("isCosmeticCommit", () => {
+  it("treats a formatting sweep as cosmetic", () => {
+    // The actual commit that triggered this.
+    expect(isCosmeticCommit("chore(format): run prettier over src/ and scripts/")).toBe(true);
+    expect(isCosmeticCommit("style: reindent the player module")).toBe(true);
+    expect(isCosmeticCommit("style(player): reindent")).toBe(true);
+  });
+
+  it("treats a lockfile-only dependency bump as cosmetic", () => {
+    // Touches package.json/bun.lock, which carry no TRACES, but a `chore(deps)`
+    // that also edits source would still be caught by that source file.
+    expect(isCosmeticCommit("chore(deps): bump vitest to 4.1.11")).toBe(true);
+  });
+
+  it("does NOT treat ordinary work as cosmetic", () => {
+    expect(isCosmeticCommit("fix(player): restart the hero banner timer")).toBe(false);
+    expect(isCosmeticCommit("feat(updater): in-app update on desktop")).toBe(false);
+    expect(isCosmeticCommit("ci: make the frontend gates real")).toBe(false);
+    expect(isCosmeticCommit("docs: add SECURITY.md")).toBe(false);
+  });
+
+  it("does not mistake a chore that is not formatting for a formatting one", () => {
+    // `chore(release)` bumps versions and must still be attributable; a bare
+    // `chore:` could be anything, so it is NOT skipped by default.
+    expect(isCosmeticCommit("chore(release): v0.9.2")).toBe(false);
+    expect(isCosmeticCommit("chore: tidy up the queue helper")).toBe(false);
+  });
+
+  it("is not fooled by the word format appearing later in a subject", () => {
+    // A real fix to formatting *code* is not a cosmetic commit.
+    expect(isCosmeticCommit("fix(duration): format times over 24 hours correctly")).toBe(false);
+    expect(isCosmeticCommit("feat: add a format picker to settings")).toBe(false);
+  });
+
+  it("handles an empty or malformed subject without throwing", () => {
+    expect(isCosmeticCommit("")).toBe(false);
+    expect(isCosmeticCommit("   ")).toBe(false);
+  });
+});

scripts/release-notes.ts

@@ -54,11 +54,86 @@ function loadRequirementDescriptions(): Map<string, string> {
   return map;
 }
 
+/**
+ * Commit subjects whose changes carry no requirement meaning.
+ *
+ * `chore(format)` / `style` rewrite files without changing behaviour;
+ * `chore(deps)` moves lockfiles. Anything else — including a bare `chore:` and
+ * `chore(release):` — is assumed to mean something and is kept.
+ *
+ * Anchored at the start of the subject on purpose: "fix(duration): format times
+ * over 24 hours" is a real fix to formatting *code*, not a formatting commit.
+ */
+const COSMETIC_SUBJECT = /^(chore\(format\)|chore\(deps\)|style)(\([^)]*\))?\s*:/i;
+
+/**
+ * Does this commit subject describe a change with no requirement meaning?
+ *
+ * Exported for scripts/release-notes.test.ts.
+ *
+ * TRACES: | DR-219
+ */
+export function isCosmeticCommit(subject: string): boolean {
+  return COSMETIC_SUBJECT.test(subject.trim());
+}
+
+/**
+ * Files the range changed, excluding those touched only by cosmetic commits.
+ *
+ * Why not a plain `git diff --name-only <range>`: that is what this did, and a
+ * single repo-wide `prettier --write` inside the range made it report 199 files
+ * whose TRACES comments resolved to nearly the entire requirement matrix. The
+ * generated notes for v0.9.2 claimed the release had added the whole
+ * application — and build-release.yml publishes this output, so the noise would
+ * have shipped.
+ *
+ * Walking commit by commit and skipping the cosmetic ones keeps a file that a
+ * sweep *and* a real change both touched: it is still listed by the real
+ * commit. Only files touched exclusively by cosmetic commits drop out, which is
+ * exactly the intent.
+ *
+ * Merge commits produce no output from `git diff-tree` without `-m`, and are
+ * skipped deliberately: everything they merge is already in the range as its
+ * own commit, so including them would double-count.
+ */
 function changedFiles(range: string): string[] {
-  const cmd = range ? `git diff --name-only ${range}` : "git ls-files"; // untagged repo: describe everything currently traced
-  return sh(cmd)
-    .split("\n")
-    .filter((f) => f && existsSync(f));
+  // Untagged repo: describe everything currently traced.
+  if (!range) {
+    return sh("git ls-files")
+      .split("\n")
+      .filter((f) => f && existsSync(f));
+  }
+
+  // NUL between hash and subject so a subject containing anything at all is safe.
+  const log = sh(`git log --no-merges --format=%H%x00%s ${range}`);
+  if (!log) return [];
+
+  const files = new Set<string>();
+  let skipped = 0;
+
+  for (const line of log.split("\n")) {
+    const [sha, ...subjectParts] = line.split("\u0000");
+    const subject = subjectParts.join("\u0000");
+    if (!sha) continue;
+
+    if (isCosmeticCommit(subject)) {
+      skipped++;
+      continue;
+    }
+
+    for (const f of sh(`git diff-tree --no-commit-id --name-only -r ${sha}`).split("\n")) {
+      if (f && existsSync(f)) files.add(f);
+    }
+  }
+
+  if (skipped > 0) {
+    // Say what was dropped rather than silently reporting a smaller set.
+    console.error(
+      `ℹ️  Skipped ${skipped} cosmetic commit(s) (formatting/deps) when deriving notes.`,
+    );
+  }
+
+  return [...files];
 }
 
 /** Collect requirement IDs referenced by TRACES comments in the given files. */
@@ -132,4 +207,7 @@ function main() {
   console.log(out.join("\n"));
 }
 
-main();
+// Guarded so this module stays importable from release-notes.test.ts.
+if (import.meta.main) {
+  main();
+}