Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

packages/desktop-electron/scripts/finalize-latest-yml.ts (1)

115-119: ⚠️ Potential issue | 🟡 Minor

Add no assets match to the isMissingAssetError regex to match the workflow.

The workflow (.github/workflows/build.yml line 341) includes no assets match as a benign error pattern, but the script's regex on line 118 omits it. This discrepancy could cause the workflow step to succeed while the script subsequently fails on the same error condition.

Update the regex to: /no assets to download|no matches found|no assets match|could not find any assets/i

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@packages/desktop-electron/scripts/finalize-latest-yml.ts` around lines 115 -
119, The isMissingAssetError function's regex is missing the "no assets match"
pattern used by the workflow; update the regular expression inside
isMissingAssetError to include the "no assets match" alternative (so it tests
for "no assets to download|no matches found|no assets match|could not find any
assets" case-insensitively) to ensure the script treats that message as a benign
missing-asset error.

.github/workflows/build.yml (1)

324-347: ⚠️ Potential issue | 🔴 Critical

Handle the first publish before trying to download prior release assets.

This step still runs before the first --publish always path in the job, at Line 669 for macOS and Line 776 for Windows. On the first finalize/full run for a version, the GitHub release may not exist yet, so gh release download "$tag" fails and the workflow exits before publication ever happens. Treat a missing release as benign, or probe for the release before downloading.

🔧 Suggested guard

           existing_dir="$RUNNER_TEMP/existing-latest-yml"
           mkdir -p "$existing_dir"
           tag="v${{ steps.package_version.outputs.version }}"
+
+          if ! gh release view "$tag" --repo "$GITHUB_REPOSITORY" >/dev/null 2>&1; then
+            echo "Release $tag does not exist yet; skipping existing metadata download."
+            exit 0
+          fi
 
           download_or_warn() {
             local pattern="$1"
             local err="$RUNNER_TEMP/${pattern}.err"
             if gh release download "$tag" --pattern "$pattern" --dir "$existing_dir" --repo "$GITHUB_REPOSITORY" 2>"$err"; then

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In @.github/workflows/build.yml around lines 324 - 347, The download step can
fail when the GitHub release doesn't yet exist; update the download_or_warn
logic to first probe for the release (e.g., with gh release view "$tag") and
treat a non-existent release as benign, or catch the specific 404/no-release
error from gh before attempting gh release download; update the function
download_or_warn (and its callers that use tag/existing_dir) to skip downloads
if the release is missing and only error for other failures so the workflow can
proceed to publish on first-run.

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

packages/desktop-electron/scripts/finalize-latest-yml.ts (1)

115-118: ⚠️ Potential issue | 🟠 Major

Handle the gh “no assets match” variant here too.

gh release download can emit no assets match for the benign missing-pattern case. This matcher will currently treat that as fatal, so finalize can fail instead of falling back to cached metadata.

🔧 Proposed fix

-  return /no assets to download|no matches found|could not find any assets/i.test(message)
+  return /no assets to download|no matches found|no assets match|could not find any assets/i.test(message)

Run this read-only check to confirm the mismatch against the rest of the release flow:

#!/bin/bash
set -euo pipefail

printf 'Current matcher in packages/desktop-electron/scripts/finalize-latest-yml.ts:\n'
sed -n '115,119p' packages/desktop-electron/scripts/finalize-latest-yml.ts

printf '\nRepo references to benign gh download messages:\n'
rg -n 'no assets match|no assets to download|no matches found|could not find any assets' \
  .github/workflows/build.yml \
  packages/desktop-electron/scripts/release-metadata-contract.test.ts

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@packages/desktop-electron/scripts/finalize-latest-yml.ts` around lines 115 -
118, The isMissingAssetError function's regex omits the "no assets match" gh
message so benign missing-asset cases are treated as fatal; update the regex in
isMissingAssetError to also match "no assets match" (in addition to the existing
"no assets to download", "no matches found", and "could not find any assets") so
finalize-latest-yml will treat that variant as a non-fatal missing-asset case.

packages/desktop-electron/src/main/problem-report.ts (1)

21-24: ⚠️ Potential issue | 🟠 Major

Make sessionExport JSON-safe before stringifying it.

info and messages are still arbitrary unknown values, but Lines 55 and 65 serialize them directly. A BigInt or circular object will throw here and break the “Report a Problem” flow before truncation runs.

Also applies to: 54-68

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@packages/desktop-electron/src/main/problem-report.ts` around lines 21 - 24,
The sessionExport object (type SessionExport) can contain non-JSON-safe values
in its info and messages fields; before calling JSON.stringify on sessionExport
(where sessionExport is constructed/serialized), sanitize info and messages with
a JSON-safe serializer or replacer that handles BigInt (convert to string) and
circular references (e.g., track seen objects and replace cycles with a
placeholder) so stringify cannot throw; update the code paths that serialize
sessionExport (the serialization around sessionExport) to use that safe
serializer or a try/catch fallback that converts problematic values to string
before truncation runs.

packages/desktop-electron/src/main/ipc.ts (1)

142-145: ⚠️ Potential issue | 🟠 Major

Reject missing sender windows so desktop-context sync can retry.

Returning undefined here resolves ipcRenderer.invoke() successfully. In packages/app/src/pages/session.tsx, retries only happen on rejected promises, so a missing BrowserWindow is treated as a successful sync and the stale context is never retried.

♻️ Proposed fix

 ipcMain.handle("set-desktop-context", (event: IpcMainInvokeEvent, context: DesktopContext) => {
   const win = BrowserWindow.fromWebContents(event.sender)
-  if (!win) return
+  if (!win) {
+    throw new Error("set-desktop-context: sender window not found")
+  }
   return deps.setDesktopContext(context, win)
 })

Verification: inspect the handler and the renderer retry paths together. The current code should show that retries are only in .catch(...), while the main handler has a success-return path for !win.

#!/bin/bash
set -e

echo "=== main IPC handler ==="
sed -n '139,146p' packages/desktop-electron/src/main/ipc.ts

echo
echo "=== session retry path ==="
sed -n '349,381p' packages/app/src/pages/session.tsx

echo
echo "=== non-session route bridge ==="
sed -n '140,150p' packages/app/src/app.tsx

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@packages/desktop-electron/src/main/ipc.ts` around lines 142 - 145, The IPC
handler "set-desktop-context" currently returns undefined when
BrowserWindow.fromWebContents(event.sender) is missing, which resolves
ipcRenderer.invoke() and prevents renderer-side retries; change the handler to
reject the invocation instead when no sender window is found (e.g., throw a
descriptive Error or return a rejected Promise) so that
deps.setDesktopContext(context, win) is only called for valid windows and the
renderer's retry logic (which retries only on rejected promises) will run on
missing-window cases.

packages/desktop-electron/src/preload/types.ts (1)

83-85: ⚠️ Potential issue | 🟡 Minor

Keep installUpdate typed in sync with the IPC contract.

packages/desktop-electron/src/main/ipc.ts exposes installUpdate as a boolean result, but ElectronAPI.installUpdate still erases that to Promise<void>. That makes the preload bridge inconsistent and prevents renderer code from detecting “install started” vs “nothing to install”.

♻️ Proposed fix

-  installUpdate: () => Promise<void>
+  installUpdate: () => Promise<boolean>

Verification: compare the preload signature with the main-process contract and current call sites.

#!/bin/bash
set -e

echo "=== main IPC contract ==="
sed -n '54,66p' packages/desktop-electron/src/main/ipc.ts

echo
echo "=== preload API contract ==="
sed -n '79,87p' packages/desktop-electron/src/preload/types.ts

echo
echo "=== installUpdate usages ==="
rg -n --type=ts --type=tsx '\binstallUpdate\b|\bupdate\?\(\): Promise<void>' packages/desktop-electron/src packages/app/src

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@packages/desktop-electron/src/preload/types.ts` around lines 83 - 85, The
preload API's installUpdate signature is out of sync with the main IPC contract:
change installUpdate: () => Promise<void> in
packages/desktop-electron/src/preload/types.ts to installUpdate: () =>
Promise<boolean> so the preload bridge returns the boolean "install started"
result; then update any renderer call sites that assume void to handle the
boolean result (e.g., code calling ElectronAPI.installUpdate or awaiting
installUpdate) and adjust logic to detect true/false accordingly; use the
existing symbols runUpdater and checkUpdate to locate the surrounding API
surface when updating the type.

.github/workflows/build.yml (1)

324-346: ⚠️ Potential issue | 🔴 Critical

Skip the pre-download when the versioned release does not exist yet.

This step still runs before the later --publish always packaging step creates v${{ steps.package_version.outputs.version }}. On a first publish for that version, gh release download fails with a missing-release error, and the current asset-only matcher treats it as fatal, so the job can stop before packaging starts. Guard the step with a release-exists check (or explicitly handle the missing-release case) before calling download_or_warn().

🔧 Suggested fix

       - name: Download existing updater metadata
         # macOS finalize can run after notarization as a separate phase; Windows finalizes during full packaging.
         if: ${{ (runner.os == 'macOS' && (inputs.phase == 'finalize' || inputs.phase == 'full') && inputs.arch == matrix.arch_label) || (runner.os == 'Windows' && inputs.phase == 'full') }}
         shell: bash
         run: |
           set -euo pipefail

           existing_dir="$RUNNER_TEMP/existing-latest-yml"
           mkdir -p "$existing_dir"
           tag="v${{ steps.package_version.outputs.version }}"
+
+          release_err="$RUNNER_TEMP/release-view.err"
+          if ! gh release view "$tag" --repo "$GITHUB_REPOSITORY" > /dev/null 2>"$release_err"; then
+            if grep -qiE 'release not found|HTTP 404: Not Found' "$release_err"; then
+              echo "Release $tag does not exist yet; skipping existing metadata download."
+              exit 0
+            fi
+            cat "$release_err" >&2
+            exit 1
+          fi

           download_or_warn() {
             local pattern="$1"
             local err="$RUNNER_TEMP/${pattern}.err"
             if gh release download "$tag" --pattern "$pattern" --dir "$existing_dir" --repo "$GITHUB_REPOSITORY" 2>"$err"; then

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In @.github/workflows/build.yml around lines 324 - 346, The step uses
download_or_warn() with tag="v${{ steps.package_version.outputs.version }}" but
doesn’t guard against the release itself being absent; add a pre-check that the
GitHub release exists for tag before calling download_or_warn (or explicitly
detect the "release not found" message and treat it like the asset-not-found
case). Implement the check right after tag is set (e.g., call gh release view
"$tag" and skip downloads if it returns a missing-release error) so
download_or_warn() is only invoked when the release exists and the job won’t
fail on first-time publishes.

packages/desktop-electron/src/main/updater.ts (1)

73-79: ⚠️ Potential issue | 🟠 Major

Reintroduced concurrency behavior: concurrent callers lose the final update result.

At Line 74, returning { status: "busy" } for later callers means they never receive the eventual ready/failed resolution of the in-flight check.

Suggested fix

     check() {
-      if (inflight) return Promise.resolve({ status: "busy" as const })
+      if (inflight) return inflight
       inflight = run().finally(() => {
         inflight = undefined
       })
       return inflight
     },

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@packages/desktop-electron/src/main/updater.ts` around lines 73 - 79, The
current check() implementation returns a static { status: "busy" } for
concurrent callers so they never get the eventual result; change the
early-return to return the existing inflight promise instead of
Promise.resolve({ status: "busy" as const }) so all callers await the same run()
result. Ensure inflight remains the Promise produced by run() (and is cleared in
finally as shown) so subsequent callers receive the final resolved value from
run() via the inflight variable used in check().

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes