fix(checkpoints): surface skipped_oversize to users and stop misreporting failed deletes as restored

Follow-up to the salvaged #95207 fix, completing the misreport bug class:

- restore() now also drops delete_targets whose unlink failed (OSError
  swallowed) from restored_files — the sibling of the kept-oversize
  misreport the salvaged fix closed.
- /rollback output in the CLI (cli_commands_mixin) and gateway
  (slash_commands + gateway.rollback.kept_oversize locale key in all 17
  catalogs) now tells the user which files were kept because the size
  cap excluded them from every checkpoint; previously the file was
  correctly preserved but the user got no notice it was not reverted.
- Regression test for the failed-unlink misreport.

gateway/slash_commands.py

@@ -3479,6 +3479,14 @@ class GatewaySlashCommandsMixin:
                     "gateway.rollback.kept_user_edits",
                     files=shown + more,
                 )
+            oversize = result.get("skipped_oversize") or []
+            if oversize:
+                shown = ", ".join(oversize[:5])
+                more = f" (+{len(oversize) - 5})" if len(oversize) > 5 else ""
+                msg += "\n" + t(
+                    "gateway.rollback.kept_oversize",
+                    files=shown + more,
+                )
             return msg
         return t("gateway.rollback.restore_failed", error=result["error"])
 

hermes_cli/cli_commands_mixin.py

@@ -165,6 +165,11 @@ class CLICommandsMixin:
                 more = f" (+{len(skipped) - 5} more)" if len(skipped) > 5 else ""
                 print(f"  ↷ Kept your hand-edits: {shown}{more}")
                 print("  Use /rollback <N> --all to restore those too.")
+            oversize = result.get("skipped_oversize") or []
+            if oversize:
+                shown = ", ".join(oversize[:5])
+                more = f" (+{len(oversize) - 5} more)" if len(oversize) > 5 else ""
+                print(f"  ↷ Kept (too large for checkpoints, no stored copy to revert to): {shown}{more}")
             print("  A pre-rollback snapshot was saved automatically.")
 
             # Also undo the last conversation turn so the agent's context

locales/af.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Ongeldige kontrolepunt-nommer. Gebruik 1-{max}."
     restored:              "✅ Herstel na kontrolepunt {hash}: {reason}\n'n Voor-terugrol-momentopname is outomaties gestoor."
     kept_user_edits:       "↷ Jou handwysigings is behou: {files}\nGebruik /rollback <N> --all om dié ook te herstel."
+    kept_oversize:         "↷ Behou (te groot vir kontrolepunte, geen gestoorde kopie om na terug te keer nie): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/ar.yaml

@@ -300,6 +300,7 @@ gateway:
     invalid_number:        "رقم نقطة تحقّق غير صالح. استخدم 1-{max}."
     restored:              "✅ استُعيد إلى نقطة التحقّق {hash}: {reason}\nحُفظت لقطة ما قبل التراجع تلقائيًا."
     kept_user_edits:       "↷ احتُفظ بتعديلاتك اليدوية: {files}\nاستخدم ‎/rollback <N> --all لاستعادتها أيضًا."
+    kept_oversize:         "↷ احتُفظ به (أكبر من أن يُحفظ في نقاط التفتيش، لا توجد نسخة مخزنة للرجوع إليها): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/de.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Ungültige Checkpoint-Nummer. Verwenden Sie 1-{max}."
     restored:              "✅ Auf Checkpoint {hash} wiederhergestellt: {reason}\nEin Pre-Rollback-Snapshot wurde automatisch gespeichert."
     kept_user_edits:       "↷ Deine manuellen Änderungen wurden behalten: {files}\nNutze /rollback <N> --all, um auch diese wiederherzustellen."
+    kept_oversize:         "↷ Behalten (zu groß für Checkpoints, keine gespeicherte Kopie zum Zurücksetzen): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/en.yaml

@@ -292,6 +292,7 @@ gateway:
     invalid_number:        "Invalid checkpoint number. Use 1-{max}."
     restored:              "✅ Restored to checkpoint {hash}: {reason}\nA pre-rollback snapshot was saved automatically."
     kept_user_edits:       "↷ Kept your hand-edits: {files}\nUse /rollback <N> --all to restore those too."
+    kept_oversize:         "↷ Kept (too large for checkpoints, no stored copy to revert to): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/es.yaml

@@ -277,6 +277,7 @@ gateway:
     invalid_number:        "Número de checkpoint inválido. Usa 1-{max}."
     restored:              "✅ Restaurado al checkpoint {hash}: {reason}\nSe guardó automáticamente un snapshot previo al rollback."
     kept_user_edits:       "↷ Se conservaron tus ediciones manuales: {files}\nUsa /rollback <N> --all para restaurarlas también."
+    kept_oversize:         "↷ Conservado (demasiado grande para los checkpoints, no hay copia guardada a la que revertir): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/fr.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Numéro de point de contrôle invalide. Utilisez 1-{max}."
     restored:              "✅ Restauré au point de contrôle {hash} : {reason}\nUn instantané pré-rollback a été enregistré automatiquement."
     kept_user_edits:       "↷ Vos modifications manuelles ont été conservées : {files}\nUtilisez /rollback <N> --all pour les restaurer aussi."
+    kept_oversize:         "↷ Conservé (trop volumineux pour les checkpoints, aucune copie enregistrée vers laquelle revenir) : {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/ga.yaml

@@ -284,6 +284,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Uimhir seicphointe neamhbhailí. Úsáid 1-{max}."
     restored:              "✅ Aischurtha go seicphointe {hash}: {reason}\nSábháladh roghchóip réamh-rollback go huathoibríoch."
     kept_user_edits:       "↷ Coinníodh do chuid athruithe láimhe: {files}\nÚsáid /rollback <N> --all chun iad sin a aischur freisin."
+    kept_oversize:         "↷ Coinnithe (ró-mhór do sheicphointí, níl aon chóip stóráilte le filleadh uirthi): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/hu.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Érvénytelen ellenőrzőpont-szám. Használj 1-{max} közötti értéket."
     restored:              "✅ Visszaállítva a(z) {hash} ellenőrzőpontra: {reason}\nA visszaállítás előtti pillanatkép automatikusan elmentve."
     kept_user_edits:       "↷ A kézi szerkesztéseid megmaradtak: {files}\nHasználd a /rollback <N> --all parancsot, hogy azokat is visszaállítsd."
+    kept_oversize:         "↷ Megtartva (túl nagy az ellenőrzőpontokhoz, nincs tárolt másolat, amire vissza lehetne állni): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/it.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Numero di checkpoint non valido. Usa 1-{max}."
     restored:              "✅ Ripristinato al checkpoint {hash}: {reason}\nUno snapshot pre-rollback è stato salvato automaticamente."
     kept_user_edits:       "↷ Le tue modifiche manuali sono state conservate: {files}\nUsa /rollback <N> --all per ripristinare anche quelle."
+    kept_oversize:         "↷ Conservato (troppo grande per i checkpoint, nessuna copia salvata a cui tornare): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/ja.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "無効なチェックポイント番号です。1-{max} を使用してください。"
     restored:              "✅ チェックポイント {hash} に復元しました: {reason}\nロールバック前のスナップショットが自動的に保存されました。"
     kept_user_edits:       "↷ 手動編集は保持されました: {files}\nそれらも復元するには /rollback <N> --all を使用してください。"
+    kept_oversize:         "↷ 保持しました（チェックポイントには大きすぎるため、戻せる保存コピーがありません）: {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/ko.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "잘못된 체크포인트 번호입니다. 1-{max}을 사용하세요."
     restored:              "✅ 체크포인트 {hash}(으)로 복원됨: {reason}\n롤백 전 스냅샷이 자동으로 저장되었습니다."
     kept_user_edits:       "↷ 직접 수정한 내용은 유지되었습니다: {files}\n해당 파일도 복원하려면 /rollback <N> --all 을 사용하세요."
+    kept_oversize:         "↷ 유지됨 (체크포인트에 저장하기엔 너무 커서 되돌릴 저장본이 없습니다): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/pt.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Número de checkpoint inválido. Usa 1-{max}."
     restored:              "✅ Restaurado para o checkpoint {hash}: {reason}\nFoi guardado automaticamente um snapshot anterior ao rollback."
     kept_user_edits:       "↷ As suas edições manuais foram mantidas: {files}\nUse /rollback <N> --all para restaurar essas também."
+    kept_oversize:         "↷ Mantido (grande demais para os checkpoints, sem cópia guardada para reverter): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/ru.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Недействительный номер контрольной точки. Используйте 1-{max}."
     restored:              "✅ Восстановлено до контрольной точки {hash}: {reason}\nСнимок перед откатом сохранён автоматически."
     kept_user_edits:       "↷ Ваши ручные правки сохранены: {files}\nИспользуйте /rollback <N> --all, чтобы восстановить и их."
+    kept_oversize:         "↷ Сохранено (слишком большой для контрольных точек, нет сохранённой копии для отката): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/tr.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Geçersiz kontrol noktası numarası. 1-{max} aralığını kullanın."
     restored:              "✅ {hash} kontrol noktasına geri yüklendi: {reason}\nGeri alma öncesi anlık görüntü otomatik olarak kaydedildi."
     kept_user_edits:       "↷ Elle yaptığınız düzenlemeler korundu: {files}\nOnları da geri yüklemek için /rollback <N> --all kullanın."
+    kept_oversize:         "↷ Korundu (denetim noktaları için çok büyük, geri dönülecek kayıtlı kopya yok): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/uk.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "Недійсний номер контрольної точки. Використовуйте 1-{max}."
     restored:              "✅ Відновлено до контрольної точки {hash}: {reason}\nЗнімок перед відкатом збережено автоматично."
     kept_user_edits:       "↷ Ваші ручні правки збережено: {files}\nВикористайте /rollback <N> --all, щоб відновити і їх."
+    kept_oversize:         "↷ Збережено (завеликий для контрольних точок, немає збереженої копії для відкату): {files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/zh-hant.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "無效的檢查點編號。請使用 1-{max}。"
     restored:              "✅ 已還原至檢查點 {hash}：{reason}\n已自動儲存回復前的快照。"
     kept_user_edits:       "↷ 已保留您的手動編輯：{files}\n若也要還原這些檔案，請使用 /rollback <N> --all。"
+    kept_oversize:         "↷ 已保留（檔案過大無法納入檢查點，沒有可還原的儲存副本）：{files}"
     restore_failed:        "❌ {error}"
 
   diff:

locales/zh.yaml

@@ -280,6 +280,7 @@ Future messages in this room will use that transcript until `/reset` or another
     invalid_number:        "无效的检查点编号。请使用 1-{max}。"
     restored:              "✅ 已恢复到检查点 {hash}：{reason}\n已自动保存回滚前的快照。"
     kept_user_edits:       "↷ 已保留您的手动编辑：{files}\n如需一并恢复这些文件，请使用 /rollback <N> --all。"
+    kept_oversize:         "↷ 已保留（文件过大无法纳入检查点，没有可恢复的存储副本）：{files}"
     restore_failed:        "❌ {error}"
 
   diff:

tests/tools/test_checkpoint_manager.py

@@ -528,6 +528,33 @@ class TestSafeRestore:
         assert "scratch.txt" in result["restored_files"]
         assert "skipped_oversize" not in result
 
+    def test_safe_restore_does_not_report_a_failed_delete_as_restored(
+        self, mgr, work_dir, monkeypatch,
+    ):
+        """A delete_target whose unlink fails stays on disk — reporting it in
+        restored_files is the same silent misreport the oversize fix closed."""
+        base = self._checkpoint(mgr, work_dir)
+
+        stubborn = work_dir / "stubborn.txt"
+        stubborn.write_text("agent scratch\n")
+        mgr.record_agent_write(str(stubborn))
+
+        import pathlib
+
+        real_unlink = pathlib.Path.unlink
+
+        def failing_unlink(self, *args, **kwargs):
+            if self.name == "stubborn.txt":
+                raise OSError(13, "Permission denied")
+            return real_unlink(self, *args, **kwargs)
+
+        monkeypatch.setattr(pathlib.Path, "unlink", failing_unlink)
+        result = mgr.restore(str(work_dir), base, safe=True)
+
+        assert result["success"] is True
+        assert stubborn.exists()
+        assert "stubborn.txt" not in result["restored_files"]
+
     def test_unsafe_restore_overwrites_everything(self, mgr, work_dir):
         base = self._checkpoint(mgr, work_dir)
         (work_dir / "main.py").write_text("agent version\n")

tools/checkpoint_manager.py

@@ -1124,6 +1124,8 @@ class CheckpointManager:
                     "debug": err or None}
 
         skipped_user_edits: List[str] = []
+        kept_oversize: List[str] = []
+        failed_deletes: List[str] = []
         restore_paths: Optional[List[str]] = None
         if safe and not file_path:
             plan = self.safe_restore_plan(abs_dir, commit_hash)
@@ -1157,7 +1159,6 @@ class CheckpointManager:
             # Hermes-created files absent from it (delete to restore state).
             checkout_targets: List[str] = []
             delete_targets: List[str] = []
-            kept_oversize: List[str] = []
             for rel in restore_paths:
                 ok_in_commit, _, _ = _run_git(
                     ["cat-file", "-e", f"{commit_hash}:{rel}"],
@@ -1184,6 +1185,7 @@ class CheckpointManager:
                         target.unlink()
                 except OSError as exc:
                     logger.debug("Safe restore: could not remove %s: %s", rel, exc)
+                    failed_deletes.append(rel)
             if not checkout_targets:
                 ok, stdout, err = True, "", ""
             else:
@@ -1218,11 +1220,13 @@ class CheckpointManager:
             result["file"] = file_path
         if restore_paths is not None:
             # Only what was actually acted on. A kept oversize path was not
-            # restored, and reporting it as such is how the data loss above
-            # stayed silent: the user was told "Restored" for a file that had
-            # just been unlinked.
+            # restored (and a failed unlink left the file in place), and
+            # reporting either as restored is how the data loss above stayed
+            # silent: the user was told "Restored" for a file that had just
+            # been unlinked.
+            not_restored = set(kept_oversize) | set(failed_deletes)
             result["restored_files"] = [
-                rel for rel in restore_paths if rel not in kept_oversize
+                rel for rel in restore_paths if rel not in not_restored
             ]
             result["skipped_user_edits"] = skipped_user_edits
             if kept_oversize: