diff --git a/src/iac_code/a2a/executor.py b/src/iac_code/a2a/executor.py
index 226a2d44..8c642210 100644
--- a/src/iac_code/a2a/executor.py
+++ b/src/iac_code/a2a/executor.py
@@ -1731,9 +1731,7 @@ def _resolve_cwd(self, metadata: Any | None) -> str:
             raise ValueError("Invalid A2A workspace metadata.")
         logical_cwd = os.path.normpath(cwd)
         resolved_cwd = resolve_workspace_path(Path(logical_cwd))
-        if not trust_request_cwd() and not any(
-            _is_relative_to(resolved_cwd, root) for root in _allowed_cwd_roots()
-        ):
+        if not trust_request_cwd() and not any(_is_relative_to(resolved_cwd, root) for root in _allowed_cwd_roots()):
             raise ValueError("Invalid A2A workspace metadata.")
         if resolved_cwd.exists():
             if not resolved_cwd.is_dir():
diff --git a/src/iac_code/a2a/input_required.py b/src/iac_code/a2a/input_required.py
index 4418de08..c0ba54a3 100644
--- a/src/iac_code/a2a/input_required.py
+++ b/src/iac_code/a2a/input_required.py
@@ -178,35 +178,29 @@ def permission_display_fields(request: PermissionRequestEvent, *, language: str
             command = safe_input.get("command") or safe_input.get("cmd")
         if isinstance(command, str) and command.strip():
             command_fallback = translate_message("shell command", language=language)
-            target = translate_message(
-                "the current local workspace; command: {command}", language=language
-            ).format(command=_display_text(command, fallback=command_fallback, maximum=240))
+            target = translate_message("the current local workspace; command: {command}", language=language).format(
+                command=_display_text(command, fallback=command_fallback, maximum=240)
+            )
         else:
             target = translate_message("the current local workspace", language=language)
         effect = "read" if is_read_only else ("local_execution" if read_only_known else "unknown")
     elif tool_name in {"write_file", "edit_file"}:
         title = translate_message("Change a workspace file", language=language)
-        purpose = translate_message(
-            "Write a file needed for the requested infrastructure task.", language=language
-        )
+        purpose = translate_message("Write a file needed for the requested infrastructure task.", language=language)
         target = _safe_input_target(safe_input, language=language) or translate_message(
             "a file in the current workspace", language=language
         )
         effect = "file_change"
     elif tool_name in {"read_file", "glob", "grep"} or is_read_only:
         title = translate_message("Read workspace data with {tool}", language=language).format(tool=public_tool)
-        purpose = translate_message(
-            "Read local data needed for the requested infrastructure task.", language=language
-        )
+        purpose = translate_message("Read local data needed for the requested infrastructure task.", language=language)
         target = _safe_input_target(safe_input, language=language) or translate_message(
             "the current local workspace", language=language
         )
         effect = "read"
     else:
         title = translate_message("Run {tool}", language=language).format(tool=public_tool)
-        purpose = translate_message(
-            "Run this operation for the requested infrastructure task.", language=language
-        )
+        purpose = translate_message("Run this operation for the requested infrastructure task.", language=language)
         target = _safe_input_target(safe_input, language=language) or translate_message(
             "the current task workspace or cloud account", language=language
         )
@@ -319,9 +313,7 @@ def _cloud_operation_title(product: str, action: str, *, is_read_only: bool, lan
     if action == "CreateStack":
         return translate_message("Create {product} stack", language=language).format(product=product_label)
     if action == "ContinueCreateStack":
-        return translate_message("Continue creating {product} stack", language=language).format(
-            product=product_label
-        )
+        return translate_message("Continue creating {product} stack", language=language).format(product=product_label)
     if action == "UpdateStack":
         return translate_message("Update {product} stack", language=language).format(product=product_label)
     if action == "DeleteStack":
@@ -344,9 +336,7 @@ def _safe_input_target(value: Any, *, language: str) -> str:
     for key in ("file_path", "filePath", "path", "region_id", "regionId", "resource_id", "resourceId"):
         candidate = value.get(key)
         if isinstance(candidate, str) and candidate.strip():
-            return _display_text(
-                candidate, fallback=translate_message("the current task scope", language=language)
-            )
+            return _display_text(candidate, fallback=translate_message("the current task scope", language=language))
     return ""
 
 
diff --git a/src/iac_code/a2a/pipeline_executor.py b/src/iac_code/a2a/pipeline_executor.py
index cf073fec..127cb079 100644
--- a/src/iac_code/a2a/pipeline_executor.py
+++ b/src/iac_code/a2a/pipeline_executor.py
@@ -1235,11 +1235,7 @@ def _create_pipeline(
             def permission_context_getter() -> Any:
                 return getattr(agent_loop, "_permission_context", None)
 
-        surface = (
-            A2A_RICH_CANDIDATE_SURFACE
-            if self._candidate_presentation == RICH_CANDIDATE_PRESENTATION
-            else "a2a"
-        )
+        surface = A2A_RICH_CANDIDATE_SURFACE if self._candidate_presentation == RICH_CANDIDATE_PRESENTATION else "a2a"
         return create_pipeline(
             pipeline_name,
             provider_manager=runtime.provider_manager,
diff --git a/src/iac_code/a2a/task_store.py b/src/iac_code/a2a/task_store.py
index 0f1ae69d..a0906315 100644
--- a/src/iac_code/a2a/task_store.py
+++ b/src/iac_code/a2a/task_store.py
@@ -864,9 +864,7 @@ async def has_active_work(self) -> bool:
                 any(record.active_task is not None and not record.active_task.done() for record in self._tasks.values())
                 or any(not task.done() for task in self._context_runtime_tasks.values())
                 or any(
-                    not task.done()
-                    for starts in self._context_execution_starts.values()
-                    for task in starts.values()
+                    not task.done() for starts in self._context_execution_starts.values() for task in starts.values()
                 )
                 or any(self._context_reconciliation_waiters.values())
                 or any(lock.locked() for lock in self._reconciliation_locks.values())
diff --git a/src/iac_code/i18n/locales/de/LC_MESSAGES/messages.po b/src/iac_code/i18n/locales/de/LC_MESSAGES/messages.po
index e00a5449..39668e99 100644
--- a/src/iac_code/i18n/locales/de/LC_MESSAGES/messages.po
+++ b/src/iac_code/i18n/locales/de/LC_MESSAGES/messages.po
@@ -4086,6 +4086,16 @@ msgstr ""
 "eine erfolgreiche Prüfung mit übereinstimmenden Parametern und Nachweisen"
 " abgedeckt sein."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"The final instance spec must match the architecture plan, budget "
+"deviation must be disclosed, and a contract discount must only be "
+"reported when it actually reduces the price."
+msgstr ""
+"Die endgültige Instanzspezifikation muss dem Architekturplan entsprechen,"
+" Budgetabweichungen müssen offengelegt werden, und ein Vertragsrabatt "
+"darf nur ausgewiesen werden, wenn er den Preis tatsächlich senkt."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Complete the current step by calling this tool to submit the conclusion. "
@@ -4194,15 +4204,15 @@ msgstr ""
 "{message} complete_step.conclusion muss eines dieser Felder enthalten: "
 "{fields}."
 
-#: src/iac_code/pipeline/engine/complete_step_tool.py
-msgid "A completion guard is misconfigured."
-msgstr "Eine Abschlussbedingung ist falsch konfiguriert."
-
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 #, python-brace-format
 msgid "{message} Validation issue: {code} ({detail})."
 msgstr "{message} Validierungsproblem: {code} ({detail})."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid "A completion guard is misconfigured."
+msgstr "Eine Abschlussbedingung ist falsch konfiguriert."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid "A conclusion content hash is required before completing the current step."
 msgstr ""
diff --git a/src/iac_code/i18n/locales/de/LC_MESSAGES/webui.po b/src/iac_code/i18n/locales/de/LC_MESSAGES/webui.po
index 8bdf883e..c382a8b8 100644
--- a/src/iac_code/i18n/locales/de/LC_MESSAGES/webui.po
+++ b/src/iac_code/i18n/locales/de/LC_MESSAGES/webui.po
@@ -545,6 +545,14 @@ msgstr "{n}Wo"
 msgid "{n}y"
 msgstr "{n}J"
 
+#: src/iac_code/web/static/js/app.js
+msgid "Operation failed"
+msgstr "Vorgang fehlgeschlagen"
+
+#: src/iac_code/web/static/js/app.js
+msgid "Archive failed"
+msgstr "Archivierung fehlgeschlagen"
+
 #: src/iac_code/web/static/js/app.js
 msgid "Read-only"
 msgstr "Schreibgeschützt"
@@ -561,14 +569,6 @@ msgstr "Bitte geben Sie einen Inhalt ein"
 msgid "Please enter a name"
 msgstr "Bitte geben Sie einen Namen ein"
 
-#: src/iac_code/web/static/js/app.js
-msgid "Operation failed"
-msgstr "Vorgang fehlgeschlagen"
-
-#: src/iac_code/web/static/js/app.js
-msgid "Archive failed"
-msgstr "Archivierung fehlgeschlagen"
-
 #: src/iac_code/web/static/js/app.js
 #, python-brace-format
 msgid "Remove {label}?"
diff --git a/src/iac_code/i18n/locales/es/LC_MESSAGES/messages.po b/src/iac_code/i18n/locales/es/LC_MESSAGES/messages.po
index be131234..3241cb84 100644
--- a/src/iac_code/i18n/locales/es/LC_MESSAGES/messages.po
+++ b/src/iac_code/i18n/locales/es/LC_MESSAGES/messages.po
@@ -4063,6 +4063,17 @@ msgstr ""
 "Cada restricción estricta explícita del usuario debe estar cubierta por "
 "una comprobación satisfactoria con parámetros y evidencias coincidentes."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"The final instance spec must match the architecture plan, budget "
+"deviation must be disclosed, and a contract discount must only be "
+"reported when it actually reduces the price."
+msgstr ""
+"La especificación final de la instancia debe coincidir con el plan de "
+"arquitectura, se debe divulgar la desviación del presupuesto y un "
+"descuento contractual solo se puede informar cuando realmente reduce el "
+"precio."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Complete the current step by calling this tool to submit the conclusion. "
@@ -4168,15 +4179,15 @@ msgstr ""
 "{message} complete_step.conclusion debe incluir uno de estos campos: "
 "{fields}."
 
-#: src/iac_code/pipeline/engine/complete_step_tool.py
-msgid "A completion guard is misconfigured."
-msgstr "La guarda de finalización está mal configurada."
-
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 #, python-brace-format
 msgid "{message} Validation issue: {code} ({detail})."
 msgstr "{message} Problema de validación: {code} ({detail})."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid "A completion guard is misconfigured."
+msgstr "La guarda de finalización está mal configurada."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid "A conclusion content hash is required before completing the current step."
 msgstr ""
diff --git a/src/iac_code/i18n/locales/es/LC_MESSAGES/webui.po b/src/iac_code/i18n/locales/es/LC_MESSAGES/webui.po
index 567f2626..945258ef 100644
--- a/src/iac_code/i18n/locales/es/LC_MESSAGES/webui.po
+++ b/src/iac_code/i18n/locales/es/LC_MESSAGES/webui.po
@@ -547,6 +547,14 @@ msgstr "{n}sem"
 msgid "{n}y"
 msgstr "{n}a"
 
+#: src/iac_code/web/static/js/app.js
+msgid "Operation failed"
+msgstr "La operación falló"
+
+#: src/iac_code/web/static/js/app.js
+msgid "Archive failed"
+msgstr "Error al archivar"
+
 #: src/iac_code/web/static/js/app.js
 msgid "Read-only"
 msgstr "Solo lectura"
@@ -563,14 +571,6 @@ msgstr "Introduce contenido"
 msgid "Please enter a name"
 msgstr "Introduce un nombre"
 
-#: src/iac_code/web/static/js/app.js
-msgid "Operation failed"
-msgstr "La operación falló"
-
-#: src/iac_code/web/static/js/app.js
-msgid "Archive failed"
-msgstr "Error al archivar"
-
 #: src/iac_code/web/static/js/app.js
 #, python-brace-format
 msgid "Remove {label}?"
diff --git a/src/iac_code/i18n/locales/fr/LC_MESSAGES/messages.po b/src/iac_code/i18n/locales/fr/LC_MESSAGES/messages.po
index 67ef1a4c..3258d9a4 100644
--- a/src/iac_code/i18n/locales/fr/LC_MESSAGES/messages.po
+++ b/src/iac_code/i18n/locales/fr/LC_MESSAGES/messages.po
@@ -4072,6 +4072,17 @@ msgstr ""
 "être couverte par une vérification réussie avec des paramètres et des "
 "preuves concordants."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"The final instance spec must match the architecture plan, budget "
+"deviation must be disclosed, and a contract discount must only be "
+"reported when it actually reduces the price."
+msgstr ""
+"La spécification finale de l’instance doit correspondre au plan "
+"d’architecture, tout écart budgétaire doit être signalé, et une remise "
+"contractuelle ne peut être mentionnée que si elle réduit réellement le "
+"prix."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Complete the current step by calling this tool to submit the conclusion. "
@@ -4175,15 +4186,15 @@ msgstr ""
 "{message} complete_step.conclusion doit inclure l’un de ces champs : "
 "{fields}."
 
-#: src/iac_code/pipeline/engine/complete_step_tool.py
-msgid "A completion guard is misconfigured."
-msgstr "Une garde de finalisation est mal configurée."
-
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 #, python-brace-format
 msgid "{message} Validation issue: {code} ({detail})."
 msgstr "{message} Problème de validation : {code} ({detail})."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid "A completion guard is misconfigured."
+msgstr "Une garde de finalisation est mal configurée."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid "A conclusion content hash is required before completing the current step."
 msgstr ""
diff --git a/src/iac_code/i18n/locales/fr/LC_MESSAGES/webui.po b/src/iac_code/i18n/locales/fr/LC_MESSAGES/webui.po
index 3e48caa8..0fd779ca 100644
--- a/src/iac_code/i18n/locales/fr/LC_MESSAGES/webui.po
+++ b/src/iac_code/i18n/locales/fr/LC_MESSAGES/webui.po
@@ -545,6 +545,14 @@ msgstr "{n} sem"
 msgid "{n}y"
 msgstr "{n} an"
 
+#: src/iac_code/web/static/js/app.js
+msgid "Operation failed"
+msgstr "Échec de l’opération"
+
+#: src/iac_code/web/static/js/app.js
+msgid "Archive failed"
+msgstr "Échec de l’archivage"
+
 #: src/iac_code/web/static/js/app.js
 msgid "Read-only"
 msgstr "Lecture seule"
@@ -561,14 +569,6 @@ msgstr "Veuillez saisir du contenu"
 msgid "Please enter a name"
 msgstr "Veuillez saisir un nom"
 
-#: src/iac_code/web/static/js/app.js
-msgid "Operation failed"
-msgstr "Échec de l’opération"
-
-#: src/iac_code/web/static/js/app.js
-msgid "Archive failed"
-msgstr "Échec de l’archivage"
-
 #: src/iac_code/web/static/js/app.js
 #, python-brace-format
 msgid "Remove {label}?"
diff --git a/src/iac_code/i18n/locales/ja/LC_MESSAGES/messages.po b/src/iac_code/i18n/locales/ja/LC_MESSAGES/messages.po
index 83869fb5..f4e2ae35 100644
--- a/src/iac_code/i18n/locales/ja/LC_MESSAGES/messages.po
+++ b/src/iac_code/i18n/locales/ja/LC_MESSAGES/messages.po
@@ -3874,6 +3874,13 @@ msgid ""
 "with matching parameters and evidence."
 msgstr "ユーザーが明示した各ハード制約は、パラメーターと根拠が一致する「満たしている」チェックで網羅する必要があります。"
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"The final instance spec must match the architecture plan, budget "
+"deviation must be disclosed, and a contract discount must only be "
+"reported when it actually reduces the price."
+msgstr "最終的なインスタンス仕様はアーキテクチャ計画と一致していなければならず、予算の乖離は明示的に開示し、契約割引は実際に価格が下がる場合にのみ記載してください。"
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Complete the current step by calling this tool to submit the conclusion. "
@@ -3967,15 +3974,15 @@ msgid ""
 "{fields}."
 msgstr "{message} complete_step.conclusion には次のいずれかのフィールドが必要です: {fields}。"
 
-#: src/iac_code/pipeline/engine/complete_step_tool.py
-msgid "A completion guard is misconfigured."
-msgstr "完了ガードの設定に誤りがあります。"
-
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 #, python-brace-format
 msgid "{message} Validation issue: {code} ({detail})."
 msgstr "{message} 検証上の問題: {code} ({detail})。"
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid "A completion guard is misconfigured."
+msgstr "完了ガードの設定に誤りがあります。"
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid "A conclusion content hash is required before completing the current step."
 msgstr "現在のステップを完了する前に、結論内容のハッシュが必要です。"
diff --git a/src/iac_code/i18n/locales/ja/LC_MESSAGES/webui.po b/src/iac_code/i18n/locales/ja/LC_MESSAGES/webui.po
index 3a9abeb3..292d76cc 100644
--- a/src/iac_code/i18n/locales/ja/LC_MESSAGES/webui.po
+++ b/src/iac_code/i18n/locales/ja/LC_MESSAGES/webui.po
@@ -539,6 +539,14 @@ msgstr "{n}週間"
 msgid "{n}y"
 msgstr "{n}年"
 
+#: src/iac_code/web/static/js/app.js
+msgid "Operation failed"
+msgstr "操作に失敗しました"
+
+#: src/iac_code/web/static/js/app.js
+msgid "Archive failed"
+msgstr "アーカイブに失敗しました"
+
 #: src/iac_code/web/static/js/app.js
 msgid "Read-only"
 msgstr "読み取り専用"
@@ -555,14 +563,6 @@ msgstr "内容を入力してください"
 msgid "Please enter a name"
 msgstr "名前を入力してください"
 
-#: src/iac_code/web/static/js/app.js
-msgid "Operation failed"
-msgstr "操作に失敗しました"
-
-#: src/iac_code/web/static/js/app.js
-msgid "Archive failed"
-msgstr "アーカイブに失敗しました"
-
 #: src/iac_code/web/static/js/app.js
 #, python-brace-format
 msgid "Remove {label}?"
diff --git a/src/iac_code/i18n/locales/pt/LC_MESSAGES/messages.po b/src/iac_code/i18n/locales/pt/LC_MESSAGES/messages.po
index 953c6a8e..a677b0c5 100644
--- a/src/iac_code/i18n/locales/pt/LC_MESSAGES/messages.po
+++ b/src/iac_code/i18n/locales/pt/LC_MESSAGES/messages.po
@@ -4044,6 +4044,16 @@ msgstr ""
 "Cada restrição rígida explícita do usuário deve ser coberta por uma "
 "verificação satisfatória com parâmetros e evidências correspondentes."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"The final instance spec must match the architecture plan, budget "
+"deviation must be disclosed, and a contract discount must only be "
+"reported when it actually reduces the price."
+msgstr ""
+"A especificação final da instância deve corresponder ao plano de "
+"arquitetura, o desvio de orçamento deve ser divulgado e um desconto "
+"contratual só pode ser informado quando realmente reduz o preço."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Complete the current step by calling this tool to submit the conclusion. "
@@ -4144,15 +4154,15 @@ msgstr ""
 "{message} complete_step.conclusion deve incluir um destes campos: "
 "{fields}."
 
-#: src/iac_code/pipeline/engine/complete_step_tool.py
-msgid "A completion guard is misconfigured."
-msgstr "Uma proteção de conclusão está configurada incorretamente."
-
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 #, python-brace-format
 msgid "{message} Validation issue: {code} ({detail})."
 msgstr "{message} Problema de validação: {code} ({detail})."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid "A completion guard is misconfigured."
+msgstr "Uma proteção de conclusão está configurada incorretamente."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid "A conclusion content hash is required before completing the current step."
 msgstr ""
diff --git a/src/iac_code/i18n/locales/pt/LC_MESSAGES/webui.po b/src/iac_code/i18n/locales/pt/LC_MESSAGES/webui.po
index d3e33fd4..4fa71155 100644
--- a/src/iac_code/i18n/locales/pt/LC_MESSAGES/webui.po
+++ b/src/iac_code/i18n/locales/pt/LC_MESSAGES/webui.po
@@ -541,6 +541,14 @@ msgstr "{n}sem"
 msgid "{n}y"
 msgstr "{n}a"
 
+#: src/iac_code/web/static/js/app.js
+msgid "Operation failed"
+msgstr "Falha na operação"
+
+#: src/iac_code/web/static/js/app.js
+msgid "Archive failed"
+msgstr "Falha ao arquivar"
+
 #: src/iac_code/web/static/js/app.js
 msgid "Read-only"
 msgstr "Somente leitura"
@@ -557,14 +565,6 @@ msgstr "Insira o conteúdo"
 msgid "Please enter a name"
 msgstr "Insira um nome"
 
-#: src/iac_code/web/static/js/app.js
-msgid "Operation failed"
-msgstr "Falha na operação"
-
-#: src/iac_code/web/static/js/app.js
-msgid "Archive failed"
-msgstr "Falha ao arquivar"
-
 #: src/iac_code/web/static/js/app.js
 #, python-brace-format
 msgid "Remove {label}?"
diff --git a/src/iac_code/i18n/locales/zh/LC_MESSAGES/messages.po b/src/iac_code/i18n/locales/zh/LC_MESSAGES/messages.po
index a1f9264f..75ed84da 100644
--- a/src/iac_code/i18n/locales/zh/LC_MESSAGES/messages.po
+++ b/src/iac_code/i18n/locales/zh/LC_MESSAGES/messages.po
@@ -3838,6 +3838,13 @@ msgid ""
 "with matching parameters and evidence."
 msgstr "每个用户明确提出的硬约束都必须由一条状态为满足的检查覆盖，且参数和证据一致。"
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"The final instance spec must match the architecture plan, budget "
+"deviation must be disclosed, and a contract discount must only be "
+"reported when it actually reduces the price."
+msgstr "最终实例规格必须与架构规划一致，预算偏离必须显式披露，且只有在合同优惠确实带来减免时才可标注优惠价格。"
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Complete the current step by calling this tool to submit the conclusion. "
@@ -3929,15 +3936,15 @@ msgid ""
 "{fields}."
 msgstr "{message} complete_step.conclusion 必须包含以下字段之一: {fields}。"
 
-#: src/iac_code/pipeline/engine/complete_step_tool.py
-msgid "A completion guard is misconfigured."
-msgstr "完成守卫配置错误。"
-
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 #, python-brace-format
 msgid "{message} Validation issue: {code} ({detail})."
 msgstr "{message} 校验问题：{code}（{detail}）。"
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid "A completion guard is misconfigured."
+msgstr "完成守卫配置错误。"
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid "A conclusion content hash is required before completing the current step."
 msgstr "完成当前步骤前需要提供结论内容哈希。"
diff --git a/src/iac_code/i18n/locales/zh/LC_MESSAGES/webui.po b/src/iac_code/i18n/locales/zh/LC_MESSAGES/webui.po
index b2136c3b..0f2fc8c9 100644
--- a/src/iac_code/i18n/locales/zh/LC_MESSAGES/webui.po
+++ b/src/iac_code/i18n/locales/zh/LC_MESSAGES/webui.po
@@ -537,6 +537,14 @@ msgstr "{n}周"
 msgid "{n}y"
 msgstr "{n}年"
 
+#: src/iac_code/web/static/js/app.js
+msgid "Operation failed"
+msgstr "操作失败"
+
+#: src/iac_code/web/static/js/app.js
+msgid "Archive failed"
+msgstr "归档失败"
+
 #: src/iac_code/web/static/js/app.js
 msgid "Read-only"
 msgstr "只读"
@@ -553,14 +561,6 @@ msgstr "请输入内容"
 msgid "Please enter a name"
 msgstr "请输入名称"
 
-#: src/iac_code/web/static/js/app.js
-msgid "Operation failed"
-msgstr "操作失败"
-
-#: src/iac_code/web/static/js/app.js
-msgid "Archive failed"
-msgstr "归档失败"
-
 #: src/iac_code/web/static/js/app.js
 #, python-brace-format
 msgid "Remove {label}?"
diff --git a/src/iac_code/pipeline/engine/complete_step_tool.py b/src/iac_code/pipeline/engine/complete_step_tool.py
index 7cabdb76..22f18a87 100644
--- a/src/iac_code/pipeline/engine/complete_step_tool.py
+++ b/src/iac_code/pipeline/engine/complete_step_tool.py
@@ -13,6 +13,7 @@
 
 from iac_code.i18n import _
 from iac_code.pipeline.display_names import display_step_name
+from iac_code.pipeline.engine.cost_consistency import validate_cost_consistency
 from iac_code.pipeline.engine.hard_constraints import collect_hard_constraints, validate_hard_constraint_checks
 from iac_code.pipeline.engine.types import StepResult, StepStatus
 from iac_code.tools.base import Tool, ToolContext, ToolResult
@@ -60,6 +61,10 @@
         "Every explicit user hard constraint must be covered by a satisfied check with matching parameters "
         "and evidence."
     ),
+    "cost_plan_consistency_required": (
+        "The final instance spec must match the architecture plan, budget deviation must be disclosed, "
+        "and a contract discount must only be reported when it actually reduces the price."
+    ),
 }
 _COMPLETION_GUARD_MESSAGE_KEY_BY_TEXT = {text: key for key, text in _COMPLETION_GUARD_MESSAGE_TEXT_BY_KEY.items()}
 
@@ -100,6 +105,10 @@ def _completion_guard_message_i18n_markers() -> tuple[str, ...]:
             "Every explicit user hard constraint must be covered by a satisfied check with matching parameters "
             "and evidence."
         ),
+        _(
+            "The final instance spec must match the architecture plan, budget deviation must be disclosed, "
+            "and a contract discount must only be reported when it actually reduces the price."
+        ),
     )
 
 
@@ -324,6 +333,7 @@ def _validate_completion_guards(self, conclusion: dict) -> str | None:
             required_tool_result = guard.get("require_tool_result")
             required_conclusion_sha256 = guard.get("require_conclusion_sha256")
             required_constraint_coverage = guard.get("require_context_constraint_coverage")
+            required_cost_consistency = guard.get("require_cost_consistency")
             required_field = guard.get("required_conclusion_field")
             required_any_of = guard.get("required_conclusion_any_of") or []
             successful_tools = self._completion_guard_state.get("successful_tools", set())
@@ -383,8 +393,54 @@ def _validate_completion_guards(self, conclusion: dict) -> str | None:
                 )
                 if validation_error is not None:
                     return validation_error
+            if isinstance(required_cost_consistency, dict):
+                validation_error = self._validate_cost_consistency(
+                    required_cost_consistency,
+                    conclusion,
+                    self._completion_guard_message(guard, None),
+                )
+                if validation_error is not None:
+                    return validation_error
         return None
 
+    def _validate_cost_consistency(
+        self,
+        requirement: dict[str, Any],
+        conclusion: dict[str, Any],
+        message: str | None,
+    ) -> str | None:
+        context_snapshot = self._completion_guard_state.get("context_snapshot")
+        if not isinstance(context_snapshot, dict):
+            context_snapshot = {}
+        planned_compute = self._resolve_dotted(
+            context_snapshot,
+            str(requirement.get("planned_compute_field") or "candidate.planned_compute"),
+        )
+        planned_budget = self._resolve_dotted(
+            context_snapshot,
+            str(requirement.get("planned_budget_field") or "candidate.planned_budget"),
+        )
+        issues = validate_cost_consistency(
+            planned_compute,
+            planned_budget,
+            conclusion,
+            spec_reconciliation_field=str(requirement.get("spec_reconciliation_field") or "spec_reconciliation"),
+            budget_deviation_field=str(requirement.get("budget_deviation_field") or "budget_deviation"),
+        )
+        if not issues:
+            return None
+        base_message = message or _(
+            "The final instance spec must match the architecture plan, budget deviation must be disclosed, "
+            "and a contract discount must only be reported when it actually reduces the price."
+        )
+        issue_summaries = [f"{issue.code}[{issue.detail}]" if issue.detail else issue.code for issue in issues]
+        code = issues[0].code if len(issues) == 1 else "multiple_cost_consistency_issues"
+        return _("{message} Validation issue: {code} ({detail}).").format(
+            message=base_message,
+            code=code,
+            detail="; ".join(issue_summaries),
+        )
+
     def _validate_context_constraint_coverage(
         self,
         requirement: dict[str, Any],
diff --git a/src/iac_code/pipeline/engine/cost_consistency.py b/src/iac_code/pipeline/engine/cost_consistency.py
new file mode 100644
index 00000000..61e09699
--- /dev/null
+++ b/src/iac_code/pipeline/engine/cost_consistency.py
@@ -0,0 +1,183 @@
+"""Consistency checks between architecture planning baselines and cost estimation results."""
+
+from __future__ import annotations
+
+import re
+from dataclasses import asdict, dataclass
+from decimal import Decimal, InvalidOperation
+from typing import Any
+
+_AMOUNT_PATTERN = re.compile(r"\d+(?:[.,]\d+)*(?:\.\d+)?")
+
+BUDGET_WITHIN = "within"
+BUDGET_ABOVE = "above"
+BUDGET_BELOW = "below"
+BUDGET_UNKNOWN = "unknown"
+
+_SPEC_PARAMETER_NAMES = {
+    "instance_type": ("InstanceType", "EcsInstanceType"),
+    "image_id": ("ImageId",),
+}
+
+
+@dataclass(frozen=True)
+class CostConsistencyIssue:
+    code: str
+    detail: str = ""
+
+    def to_dict(self) -> dict[str, str]:
+        return {key: value for key, value in asdict(self).items() if value}
+
+
+def parse_monthly_amounts(monthly_estimate: Any) -> tuple[Decimal | None, Decimal | None]:
+    """Extract (list_price, discounted_price) from a rendered monthly estimate string.
+
+    A single amount means only the list price is disclosed; the discounted price is None.
+    """
+
+    if not isinstance(monthly_estimate, str):
+        return None, None
+    amounts = [amount for amount in (_decimal_or_none(raw) for raw in _AMOUNT_PATTERN.findall(monthly_estimate))]
+    amounts = [amount for amount in amounts if amount is not None]
+    if not amounts:
+        return None, None
+    if len(amounts) == 1:
+        return amounts[0], None
+    return amounts[0], amounts[1]
+
+
+def evaluate_budget_deviation(planned_budget: Any, monthly_estimate: Any) -> tuple[str, Decimal | None]:
+    """Compare the actual monthly cost against the planned budget range."""
+
+    actual, _discounted = parse_monthly_amounts(monthly_estimate)
+    if actual is None:
+        return BUDGET_UNKNOWN, None
+    if not isinstance(planned_budget, dict):
+        return BUDGET_UNKNOWN, actual
+    minimum = _decimal_or_none(planned_budget.get("monthly_min"))
+    maximum = _decimal_or_none(planned_budget.get("monthly_max"))
+    if minimum is None or maximum is None:
+        return BUDGET_UNKNOWN, actual
+    if minimum > maximum:
+        minimum, maximum = maximum, minimum
+    if actual > maximum:
+        return BUDGET_ABOVE, actual
+    if actual < minimum:
+        return BUDGET_BELOW, actual
+    return BUDGET_WITHIN, actual
+
+
+def reconcile_instance_spec(
+    planned_compute: Any,
+    deployment_parameters: Any,
+    preview_parameters: Any,
+) -> list[CostConsistencyIssue]:
+    """Verify the planned compute spec is carried unchanged into pricing and preview parameters."""
+
+    if not isinstance(planned_compute, dict):
+        return []
+    issues: list[CostConsistencyIssue] = []
+    deployment = deployment_parameters if isinstance(deployment_parameters, dict) else {}
+    preview = preview_parameters if isinstance(preview_parameters, dict) else {}
+    for field, parameter_names in _SPEC_PARAMETER_NAMES.items():
+        planned = planned_compute.get(field)
+        if not isinstance(planned, str) or not planned.strip():
+            continue
+        pricing_value = _first_parameter(deployment, parameter_names)
+        if pricing_value is None:
+            issues.append(CostConsistencyIssue("spec_missing_in_deployment_parameters", field))
+        elif not _spec_equal(pricing_value, planned):
+            issues.append(CostConsistencyIssue("spec_deviates_from_plan", f"{field}: {planned} -> {pricing_value}"))
+        preview_value = _first_parameter(preview, parameter_names)
+        if preview_value is not None and pricing_value is not None and not _spec_equal(preview_value, pricing_value):
+            issues.append(CostConsistencyIssue("spec_preview_mismatch", f"{field}: {pricing_value} -> {preview_value}"))
+    return issues
+
+
+def evaluate_discount_disclosure(monthly_estimate: Any) -> list[CostConsistencyIssue]:
+    """Reject a contract-discount claim when the discounted price equals the list price."""
+
+    if not isinstance(monthly_estimate, str) or "合同优惠" not in monthly_estimate:
+        return []
+    list_price, discounted = parse_monthly_amounts(monthly_estimate)
+    if list_price is None or discounted is None:
+        return []
+    if list_price == discounted:
+        return [CostConsistencyIssue("discount_without_reduction", f"{list_price}")]
+    return []
+
+
+def validate_cost_consistency(
+    planned_compute: Any,
+    planned_budget: Any,
+    conclusion: dict[str, Any],
+    *,
+    spec_reconciliation_field: str = "spec_reconciliation",
+    budget_deviation_field: str = "budget_deviation",
+) -> list[CostConsistencyIssue]:
+    """Validate spec reconciliation, budget deviation disclosure and discount honesty."""
+
+    monthly_estimate = conclusion.get("monthly_estimate")
+    deployment_parameters = conclusion.get("deployment_parameters")
+    preview_validation = conclusion.get("preview_validation")
+    preview_parameters = preview_validation.get("parameters") if isinstance(preview_validation, dict) else None
+
+    issues = reconcile_instance_spec(planned_compute, deployment_parameters, preview_parameters)
+    issues.extend(evaluate_discount_disclosure(monthly_estimate))
+
+    if isinstance(planned_compute, dict) and any(
+        isinstance(planned_compute.get(field), str) and planned_compute[field].strip()
+        for field in _SPEC_PARAMETER_NAMES
+    ):
+        reconciliation = conclusion.get(spec_reconciliation_field)
+        if not isinstance(reconciliation, dict):
+            issues.append(CostConsistencyIssue("missing_spec_reconciliation"))
+        else:
+            reported_match = reconciliation.get("matches_plan")
+            actual_match = not any(issue.code == "spec_deviates_from_plan" for issue in issues)
+            if reported_match is not actual_match:
+                issues.append(CostConsistencyIssue("spec_reconciliation_mismatch", str(reported_match)))
+            if not actual_match and not str(reconciliation.get("deviation_note") or "").strip():
+                issues.append(CostConsistencyIssue("missing_spec_deviation_note"))
+
+    status, actual = evaluate_budget_deviation(planned_budget, monthly_estimate)
+    if status != BUDGET_UNKNOWN:
+        deviation = conclusion.get(budget_deviation_field)
+        if not isinstance(deviation, dict):
+            issues.append(CostConsistencyIssue("missing_budget_deviation"))
+        else:
+            if deviation.get("status") != status:
+                issues.append(CostConsistencyIssue("budget_deviation_status_mismatch", f"{status}"))
+            reported_actual = _decimal_or_none(deviation.get("actual_monthly"))
+            if actual is not None and (reported_actual is None or reported_actual != actual):
+                issues.append(CostConsistencyIssue("budget_deviation_amount_mismatch", f"{actual}"))
+            if status in {BUDGET_ABOVE, BUDGET_BELOW} and not str(deviation.get("note") or "").strip():
+                issues.append(CostConsistencyIssue("missing_budget_deviation_note", status))
+    return issues
+
+
+def _first_parameter(parameters: dict[str, Any], names: tuple[str, ...]) -> str | None:
+    for name in names:
+        value = parameters.get(name)
+        if isinstance(value, str) and value.strip():
+            return value
+    return None
+
+
+def _spec_equal(left: str, right: str) -> bool:
+    return left.strip().casefold() == right.strip().casefold()
+
+
+def _decimal_or_none(value: Any) -> Decimal | None:
+    if isinstance(value, bool):
+        return None
+    if not isinstance(value, (int, float, Decimal, str)):
+        return None
+    text = str(value).strip().replace(",", "")
+    if not text:
+        return None
+    try:
+        number = Decimal(text)
+    except InvalidOperation:
+        return None
+    return number if number.is_finite() else None
diff --git a/src/iac_code/pipeline/engine/loader.py b/src/iac_code/pipeline/engine/loader.py
index 88f5e298..82539919 100644
--- a/src/iac_code/pipeline/engine/loader.py
+++ b/src/iac_code/pipeline/engine/loader.py
@@ -43,6 +43,7 @@
     "message_key",
     "require_conclusion_sha256",
     "require_context_constraint_coverage",
+    "require_cost_consistency",
     "require_tool",
     "require_tool_result",
     "required_conclusion_any_of",
@@ -343,9 +344,7 @@ def _parse_surface_overrides(raw: object, step_id: str) -> dict[str, StepSurface
 
         conclusion_schema = override.get("conclusion_schema")
         if conclusion_schema is not None and not isinstance(conclusion_schema, dict):
-            raise ValueError(
-                f"Step '{step_id}': surface_overrides.{surface}.conclusion_schema must be a mapping"
-            )
+            raise ValueError(f"Step '{step_id}': surface_overrides.{surface}.conclusion_schema must be a mapping")
 
         overrides[surface] = StepSurfaceOverride(
             prompt_file=prompt,
diff --git a/src/iac_code/pipeline/engine/pipeline_runner.py b/src/iac_code/pipeline/engine/pipeline_runner.py
index 59b22a03..a7b4c075 100644
--- a/src/iac_code/pipeline/engine/pipeline_runner.py
+++ b/src/iac_code/pipeline/engine/pipeline_runner.py
@@ -93,6 +93,8 @@
 
 def _is_a2a_surface(surface: str) -> bool:
     return surface == "a2a" or surface.startswith("a2a_")
+
+
 _SIDECAR_ROOT_DIRS = {"a2a", "image-cache", "pipeline", "tool-results"}
 _SIDECAR_ROOT_FILES = {
     ".backup-state.json",
diff --git a/src/iac_code/pipeline/selling/pipeline.yaml b/src/iac_code/pipeline/selling/pipeline.yaml
index af868cb4..a0b60b42 100644
--- a/src/iac_code/pipeline/selling/pipeline.yaml
+++ b/src/iac_code/pipeline/selling/pipeline.yaml
@@ -356,6 +356,13 @@ sub_pipelines:
               checks_field: hard_constraint_checks
               deployment_parameters_field: deployment_parameters
             message_key: hard_constraint_verification_required
+          - always: true
+            require_cost_consistency:
+              planned_compute_field: candidate.planned_compute
+              planned_budget_field: candidate.planned_budget
+              spec_reconciliation_field: spec_reconciliation
+              budget_deviation_field: budget_deviation
+            message_key: cost_plan_consistency_required
         tools:
           include: []
           exclude:
diff --git a/src/iac_code/pipeline/selling/prompts/confirm_and_select.a2a.rich.md b/src/iac_code/pipeline/selling/prompts/confirm_and_select.a2a.rich.md
index 04219ce6..ff66d6e5 100644
--- a/src/iac_code/pipeline/selling/prompts/confirm_and_select.a2a.rich.md
+++ b/src/iac_code/pipeline/selling/prompts/confirm_and_select.a2a.rich.md
@@ -28,6 +28,8 @@
 
 不要使用 `evaluated_candidates[i].candidate.monthly_estimate`，它只是架构规划阶段的粗略估算。不要重新询价或自行补算价格。若费用明细没有规格字段，省略 `spec`，不要猜测。
 
+如果 `evaluated_candidates[i].cost.budget_deviation.status` 是 `above` 或 `below`，必须在 `summary` 末尾追加一句偏离标注，说明实际费用相对规划预算区间的方向和 `note` 内容。同理，`cost.spec_reconciliation.matches_plan` 为 `false` 时，在 `summary` 中说明最终规格与规划规格的差异。`status` 为 `within` 或字段缺失时不要添加任何偏离描述。
+
 `user_prompt`、`summary` 和 Mermaid 节点标签应使用用户当前语言；字段名、枚举值和其他协议字段保持上面的固定形式。
 
 `complete_step.conclusion.user_prompt` 必须是展示给用户的选择提示，例如“请选择要部署的方案：”。
diff --git a/src/iac_code/pipeline/selling/prompts/confirm_and_select.md b/src/iac_code/pipeline/selling/prompts/confirm_and_select.md
index 22c0d6af..7f929dbc 100644
--- a/src/iac_code/pipeline/selling/prompts/confirm_and_select.md
+++ b/src/iac_code/pipeline/selling/prompts/confirm_and_select.md
@@ -50,6 +50,8 @@
 
 不要使用 `evaluated_candidates[i].candidate.monthly_estimate`，该字段是架构规划阶段的粗略估算。不要重新询价，也不要重新估算价格。
 
+如果 `evaluated_candidates[i].cost.budget_deviation.status` 是 `above` 或 `below`，必须在 `summary` 末尾追加一句偏离标注，说明实际费用相对规划预算区间的方向和 `note` 内容（例如"实际 ¥443.29/月，超出规划的 ¥200-300/月 区间"）。同理，`cost.spec_reconciliation.matches_plan` 为 `false` 时，在 `summary` 中说明最终规格与规划规格的差异。`status` 为 `within` 或字段缺失时不要添加任何偏离描述。
+
 如果多个方案都需要展示，必须对每个方案都调用一次“架构图 + 方案详情”的并行展示；不要为了架构图优化额外阻塞方案详情展示。
 
 - 不要用文字输出对比表格或方案信息 — 所有展示数据通过上述工具传递
diff --git a/src/iac_code/pipeline/selling/prompts/cost_estimating.md b/src/iac_code/pipeline/selling/prompts/cost_estimating.md
index 559c4bc2..d48c93fa 100644
--- a/src/iac_code/pipeline/selling/prompts/cost_estimating.md
+++ b/src/iac_code/pipeline/selling/prompts/cost_estimating.md
@@ -6,6 +6,14 @@
 
 ## 当前候选方案
 - 名称：`{candidate.name}`
+- 规划计算规格：
+```json
+{candidate.planned_compute}
+```
+- 规划月度预算：
+```json
+{candidate.planned_budget}
+```
 - 资源生命周期：
 ```json
 {candidate.resource_intents}
@@ -36,12 +44,15 @@
 ## 输出
 API 调用完成后调用 `complete_step` 提交费用预估。
 
-`complete_step.conclusion.monthly_estimate` 必须保留两个价格口径：
+`complete_step.conclusion.monthly_estimate` 的价格口径：
 - `OriginalAmount` 是原价，按统一月度周期换算并汇总为列表价。
 - `TradeAmount` 是合同优惠后的最终价，按与原价相同的月度周期换算并汇总。
-- 两个字段都存在时，使用 `¥<原价>/月（列表价，合同优惠后约¥<最终价>/月）` 格式；即使数值相同也保留两个价格口径。
+- `TradeAmount` 低于 `OriginalAmount` 时，使用 `¥<原价>/月（列表价，合同优惠后约¥<最终价>/月）` 格式。
+- 两者相等表示账户没有产生任何减免，只输出 `¥<原价>/月（列表价）`，不得标注合同优惠后价格。
 - 任一字段缺失时只展示可用价格，并在 `api_raw_summary` 中说明缺失字段；询价失败时仍填写 `"询价失败"`。
 
+上方给出了 `candidate.planned_compute` 和 `candidate.planned_budget` 时，还必须按技能的「规划基线一致性核对」输出 `spec_reconciliation` 和 `budget_deviation`：最终规格必须与规划规格一致，确有阻碍才允许偏离并说明原因；实际月度费用超出或低于规划区间时必须显式标注。
+
 若 `ros_preview_template` 成功，在 `complete_step.conclusion.preview_validation` 写入 PreviewStack 成功证明：`succeeded: true`、`template_url: "{template.file_path}"`、`parameters: <预览通过的同一参数字典>`；失败或未执行时写入 `succeeded: false`、`error: "<原因>"`。
 
 ## 注意事项
diff --git a/src/iac_code/pipeline/selling/skills/iac-aliyun-architecture/SKILL.md b/src/iac_code/pipeline/selling/skills/iac-aliyun-architecture/SKILL.md
index d46d3fb7..e413339e 100644
--- a/src/iac_code/pipeline/selling/skills/iac-aliyun-architecture/SKILL.md
+++ b/src/iac_code/pipeline/selling/skills/iac-aliyun-architecture/SKILL.md
@@ -83,6 +83,33 @@ conclusion_schema:
             type: string
           monthly_estimate:
             type: string
+          planned_compute:
+            type: object
+            additionalProperties: false
+            description: 本方案规划的 ECS 计算规格基线；后续成本步骤必须与它 reconcile。方案不包含 ECS 计算资源时省略
+            properties:
+              instance_type:
+                type: string
+                description: 规划采用的 ECS 实例类型，如 ecs.t6-c1m1.large；尚未确定具体规格时省略
+              image_id:
+                type: string
+                description: 规划采用的镜像 ID；尚未确定时省略
+          planned_budget:
+            type: object
+            required: [monthly_min, monthly_max, currency]
+            additionalProperties: false
+            description: monthly_estimate 的结构化区间；后续成本步骤按它判定预算偏离
+            properties:
+              monthly_min:
+                type: number
+                description: 规划月度费用下界，与 monthly_estimate 文本一致
+              monthly_max:
+                type: number
+                description: 规划月度费用上界，与 monthly_estimate 文本一致
+              currency:
+                type: string
+                enum: [CNY]
+                description: 预算币种，固定为 CNY
           pros:
             type: array
             items:
@@ -134,6 +161,13 @@ conclusion_schema:
 - 月度费用估算范围
 - 优势和局限
 
+## 规划基线：规格与预算
+
+`planned_compute` 和 `planned_budget` 是本步骤向成本估算步骤传递的**可核对基线**，成本步骤会按它们判定规格漂移和预算偏离。
+
+- `planned_compute`：方案包含 ECS 计算资源时，填写你规划采用的 `instance_type`（必要时含 `image_id`）。这是后续 `cost_estimating`、`ros_preview_template` 必须沿用的同一规格；不要在这里填一个小规格却预期后续用大规格。方案不含 ECS 计算资源时省略该字段。
+- `planned_budget`：把 `monthly_estimate` 的文本区间同步为结构化的 `monthly_min` / `monthly_max`，两者必须一致（如 `monthly_estimate: "¥50-100/月"` 对应 `monthly_min: 50`、`monthly_max: 100`）。区间要覆盖你对该方案的真实预期，不要为了让后续询价结果落在区间内而人为放宽。
+
 ## 资源生命周期约束
 
 如果 intent 中存在 `resource_intents`，它是架构设计的硬约束：
diff --git a/src/iac_code/pipeline/selling/skills/iac-aliyun-cost/SKILL.md b/src/iac_code/pipeline/selling/skills/iac-aliyun-cost/SKILL.md
index dd4ce0a1..30406317 100644
--- a/src/iac_code/pipeline/selling/skills/iac-aliyun-cost/SKILL.md
+++ b/src/iac_code/pipeline/selling/skills/iac-aliyun-cost/SKILL.md
@@ -17,7 +17,44 @@ conclusion_schema:
   properties:
     monthly_estimate:
       type: string
-      description: 月度费用估算；询价同时返回 OriginalAmount 与 TradeAmount 时，必须同时包含列表价和合同优惠后价格（如 ¥96.80/月（列表价，合同优惠后约¥13.76/月））；询价失败时填 "询价失败"
+      description: 月度费用估算；询价返回的 TradeAmount 低于 OriginalAmount 时必须同时包含列表价和合同优惠后价格（如 ¥96.80/月（列表价，合同优惠后约¥13.76/月）），两者相等表示账户无合同折扣、只填列表价（如 ¥96.80/月（列表价））；询价失败时填 "询价失败"
+    spec_reconciliation:
+      type: object
+      required: [matches_plan]
+      additionalProperties: false
+      description: 最终采用的计算规格与 candidate.planned_compute 的核对结果；candidate 提供了规划规格时必填
+      properties:
+        instance_type:
+          type: string
+          description: 最终用于询价和预览的 ECS 实例类型
+        image_id:
+          type: string
+          description: 最终用于询价和预览的镜像 ID
+        matches_plan:
+          type: boolean
+          description: 最终规格是否与 candidate.planned_compute 完全一致
+        deviation_note:
+          type: string
+          description: matches_plan 为 false 时必填，说明偏离的原因与影响
+    budget_deviation:
+      type: object
+      required: [status]
+      additionalProperties: false
+      description: 实际询价结果与 candidate.planned_budget 的比对结论；candidate 提供了规划预算时必填
+      properties:
+        status:
+          type: string
+          enum: [within, above, below, unknown]
+          description: 实际月度费用相对规划区间的位置
+        planned_range:
+          type: string
+          description: 规划预算区间的可读表示，如 "¥50-100/月"
+        actual_monthly:
+          type: number
+          description: 从 monthly_estimate 归一化出的实际月度列表价数值
+        note:
+          type: string
+          description: status 为 above 或 below 时必填，显式说明偏离幅度与原因
     currency:
       type: string
       enum: [CNY]
@@ -198,7 +235,22 @@ conclusion_schema:
 5. **按需修复问题** — 仅当询价失败且错误指向模板问题，或你必须修复/改写模板时，修改模板并写回原文件路径
 6. **修改后校验并重新询价** — 调用 `ros_validate_template` 校验改动；通过后调用 `ros_estimate_template_cost` 重新询价；失败则修复重试（最多 7 轮）
 7. **结构化传递参数** — 在 `complete_step.conclusion.deployment_parameters` 输出当前已选或已用于询价的参数字典；在 `preview_validation` 输出预览成功证明；在 `missing_deployment_parameters` 输出仍未补齐的完整部署参数缺口
-8. **输出结果** — 汇总费用并调用 `complete_step`
+8. **核对规划基线** — 按「规划基线一致性核对」比对 `candidate.planned_compute` 与 `candidate.planned_budget`，输出 `spec_reconciliation` 和 `budget_deviation`
+9. **输出结果** — 汇总费用并调用 `complete_step`
+
+## 规划基线一致性核对
+
+`candidate.planned_compute` 和 `candidate.planned_budget` 是 architecture_planning 传下来的规划基线。`complete_step` 会用代码逐项核对，删字段或改写结论都无法绕过。
+
+**规格一致性**：规划给出 `instance_type`（或 `image_id`）时，它就是本步骤的默认规格。同一个值必须贯穿 `ros_preview_template`、`ros_estimate_template_cost`、`deployment_parameters` 和 `preview_validation.parameters`，不得在询价时换成另一档规格。
+
+- 规划规格能满足全部硬约束和库存可用性时，直接沿用，`spec_reconciliation.matches_plan` 填 `true`。
+- 规划规格确实不可用时（违反用户硬约束、库存不足、`AllowedValues` 不包含），才允许改用其它规格，此时 `matches_plan` 填 `false` 并在 `deviation_note` 说明真实原因（如"规划的 ecs.t6-c1m1.large 不满足用户 4 vCPU 硬约束，改用 ecs.c6.xlarge"）。不得在没有实际阻碍时"顺手升配"。
+- `spec_reconciliation.instance_type` / `image_id` 必须填最终真实用于询价和预览的值，不是规划值。
+
+**预算偏离标注**：规划给出 `planned_budget` 时，把询价得到的月度列表价与 `[monthly_min, monthly_max]` 比对，在 `budget_deviation.status` 填 `within` / `above` / `below`，`actual_monthly` 填归一化后的数值。超出或低于区间时必须在 `note` 里显式说明偏离幅度与原因，不得静默通过。
+
+**合同优惠真实性**：`TradeAmount` 等于 `OriginalAmount` 表示该账户没有产生任何减免，此时 `monthly_estimate` 只输出列表价口径，禁止再写"合同优惠后约¥X"——那会让用户误以为存在不存在的折扣。只有 `TradeAmount` 确实低于 `OriginalAmount` 时才输出两个口径。
 
 ## 按需校验模板
 
@@ -306,4 +358,6 @@ aliyun_api(product="ros", action="GetResourceType", params={"ResourceType": "<
 - `preview_validation` 填 `ros_preview_template` 的结构化状态：成功时填 `{"succeeded": true, "template_url": "<当前模板文件路径>", "parameters": <预览通过的同一参数字典>}`；失败或未执行时填 `{"succeeded": false, "error": "<原因>"}`
 - `missing_deployment_parameters` 填完整部署或 PreviewStack 仍缺少的参数及原因；没有缺口时可省略或填 `[]`
 - `parameter_set_summary` 可简要说明参数来源、可用性筛选、PreviewStack 验证结果以及是否使用软门禁继续询价
+- `spec_reconciliation` 在 candidate 提供 `planned_compute` 时必填，`matches_plan` 为 `false` 时同时填 `deviation_note`
+- `budget_deviation` 在 candidate 提供 `planned_budget` 时必填，`status` 为 `above` 或 `below` 时同时填 `note`
 - 询价失败时 `monthly_estimate` 填 "询价失败"，`resources` 为空数组，`error` 说明原因
diff --git a/src/iac_code/services/providers/aliyun.py b/src/iac_code/services/providers/aliyun.py
index a66f6d59..e0119007 100644
--- a/src/iac_code/services/providers/aliyun.py
+++ b/src/iac_code/services/providers/aliyun.py
@@ -298,9 +298,7 @@ def refresh_oauth_if_needed(
 
             owns_client = oauth_client is None
             client = (
-                AliyunOAuthClient(get_oauth_site(credential.oauth_site_type))
-                if oauth_client is None
-                else oauth_client
+                AliyunOAuthClient(get_oauth_site(credential.oauth_site_type)) if oauth_client is None else oauth_client
             )
 
             try:
diff --git a/src/iac_code/services/session_backup_staging.py b/src/iac_code/services/session_backup_staging.py
index 6d061c47..e5d933a3 100644
--- a/src/iac_code/services/session_backup_staging.py
+++ b/src/iac_code/services/session_backup_staging.py
@@ -149,8 +149,7 @@ def backup_session(
                         existing = self._read_existing_snapshot_state(destination, session_id)
                         if existing is not None:
                             completed_next = (
-                                base_state.status == "succeeded"
-                                and existing.parent_generation == base_state.generation
+                                base_state.status == "succeeded" and existing.parent_generation == base_state.generation
                             )
                             if not completed_next and not existing.same_lineage(committed_state):
                                 raise SessionBackupConflict(
diff --git a/tests/a2a/test_executor.py b/tests/a2a/test_executor.py
index 4c48e2b7..4e0a31e1 100644
--- a/tests/a2a/test_executor.py
+++ b/tests/a2a/test_executor.py
@@ -3788,12 +3788,10 @@ def test_accepts_skill_rich_presentation_metadata(self) -> None:
         executor = self._make_executor()
 
         assert (
-            executor._resolve_candidate_presentation({"iac_code": {"candidatePresentation": " rich-v1 "}})
-            == "rich-v1"
+            executor._resolve_candidate_presentation({"iac_code": {"candidatePresentation": " rich-v1 "}}) == "rich-v1"
         )
         assert (
-            executor._resolve_candidate_presentation({"iac_code": {"candidate_presentation": "RICH-V1"}})
-            == "rich-v1"
+            executor._resolve_candidate_presentation({"iac_code": {"candidate_presentation": "RICH-V1"}}) == "rich-v1"
         )
 
     def test_rejects_unknown_or_missing_presentation(self) -> None:
diff --git a/tests/a2a/test_input_required.py b/tests/a2a/test_input_required.py
index b7615e0e..0f310aa7 100644
--- a/tests/a2a/test_input_required.py
+++ b/tests/a2a/test_input_required.py
@@ -240,9 +240,7 @@ def test_ros_deployment_permission_is_localized_and_preserves_safe_plan_summary(
                         "stackName": "demo-stack",
                         "template": "templates/demo.yml",
                         "totalMonthlyCost": "¥88/月",
-                        "resources": [
-                            {"name": "ECS", "spec": "2 vCPU / 4 GiB", "monthlyCost": "¥88/月"}
-                        ],
+                        "resources": [{"name": "ECS", "spec": "2 vCPU / 4 GiB", "monthlyCost": "¥88/月"}],
                     },
                 },
             ),
@@ -388,9 +386,7 @@ async def record_before_enqueue(envelope):
 async def test_sub_pipeline_permissions_stay_working_and_resolve_independently(monkeypatch, tmp_path) -> None:
     registry = PermissionInputRegistry()
     store = A2ATaskStore()
-    await store.save(
-        Task(id="task-1", context_id="ctx-1", status=TaskStatus(state=TaskState.TASK_STATE_WORKING))
-    )
+    await store.save(Task(id="task-1", context_id="ctx-1", status=TaskStatus(state=TaskState.TASK_STATE_WORKING)))
     queue = FakeEventQueue()
     publisher = PipelineA2AEventPublisher(
         event_queue=queue,
@@ -466,9 +462,7 @@ async def test_sub_pipeline_permissions_stay_working_and_resolve_independently(m
     task = await store.get("task-1")
     assert task is not None
     task_metadata = MessageToDict(task.metadata, preserving_proto_field_name=False)
-    assert [item["inputId"] for item in task_metadata["iac_code"]["pendingPermissions"]] == [
-        requests[1]["inputId"]
-    ]
+    assert [item["inputId"] for item in task_metadata["iac_code"]["pendingPermissions"]] == [requests[1]["inputId"]]
     remaining = task_metadata["iac_code"]["pendingPermissions"][0]
     assert remaining["language"] == "zh"
     assert remaining["prompt"] == "是否允许本次操作：运行本地 Shell 命令？"
diff --git a/tests/pipeline/engine/test_complete_step_tool.py b/tests/pipeline/engine/test_complete_step_tool.py
index b47ee908..63f57498 100644
--- a/tests/pipeline/engine/test_complete_step_tool.py
+++ b/tests/pipeline/engine/test_complete_step_tool.py
@@ -1460,6 +1460,94 @@ async def test_disallowed_file_mutation_matches_relative_and_absolute_paths(self
         assert result.is_error
         assert "rerun validation" in result.content
 
+    @staticmethod
+    def _cost_consistency_tool(planned_compute, planned_budget):
+        return CompleteStepTool(
+            StepConfig(step_id="cost_estimating", conclusion_field="cost", forward=None),
+            completion_guards=[
+                {
+                    "always": True,
+                    "require_cost_consistency": {
+                        "planned_compute_field": "candidate.planned_compute",
+                        "planned_budget_field": "candidate.planned_budget",
+                        "spec_reconciliation_field": "spec_reconciliation",
+                        "budget_deviation_field": "budget_deviation",
+                    },
+                    "message_key": "cost_plan_consistency_required",
+                }
+            ],
+            completion_guard_state={
+                "context_snapshot": {
+                    "candidate": {"planned_compute": planned_compute, "planned_budget": planned_budget}
+                }
+            },
+        )
+
+    def test_cost_consistency_guard_accepts_reconciled_conclusion(self):
+        tool = self._cost_consistency_tool(
+            {"instance_type": "ecs.t6-c1m1.large"},
+            {"monthly_min": 200, "monthly_max": 300, "currency": "CNY"},
+        )
+
+        error = tool.validate_completion_input(
+            {
+                "conclusion": {
+                    "monthly_estimate": "¥281.77/月（列表价）",
+                    "deployment_parameters": {"InstanceType": "ecs.t6-c1m1.large"},
+                    "spec_reconciliation": {"instance_type": "ecs.t6-c1m1.large", "matches_plan": True},
+                    "budget_deviation": {"status": "within", "actual_monthly": 281.77},
+                }
+            }
+        )
+
+        assert error is None
+
+    def test_cost_consistency_guard_blocks_unreconciled_spec_drift(self):
+        tool = self._cost_consistency_tool(
+            {"instance_type": "ecs.t6-c1m1.large"},
+            {"monthly_min": 50, "monthly_max": 100, "currency": "CNY"},
+        )
+
+        error = tool.validate_completion_input(
+            {
+                "conclusion": {
+                    "monthly_estimate": "¥281.77/月（列表价）",
+                    "deployment_parameters": {"InstanceType": "ecs.g5.large"},
+                    "spec_reconciliation": {"instance_type": "ecs.g5.large", "matches_plan": True},
+                    "budget_deviation": {"status": "within", "actual_monthly": 281.77},
+                }
+            }
+        )
+
+        assert error is not None
+        assert "multiple_cost_consistency_issues" in error
+        assert "spec_deviates_from_plan" in error
+        assert "budget_deviation_status_mismatch" in error
+
+    def test_cost_consistency_guard_blocks_ineffective_contract_discount(self):
+        tool = self._cost_consistency_tool(None, None)
+
+        error = tool.validate_completion_input(
+            {"conclusion": {"monthly_estimate": "¥443.29/月（列表价，合同优惠后约¥443.29/月）"}}
+        )
+
+        assert error is not None
+        assert "discount_without_reduction" in error
+
+    def test_cost_consistency_guard_skips_candidates_without_baseline(self):
+        tool = self._cost_consistency_tool(None, None)
+
+        error = tool.validate_completion_input(
+            {
+                "conclusion": {
+                    "monthly_estimate": "¥443.29/月（列表价）",
+                    "deployment_parameters": {"InstanceType": "ecs.g5.large"},
+                }
+            }
+        )
+
+        assert error is None
+
 
 class TestSchemaValidation:
     def test_missing_conclusion_validation_error_includes_current_step_schema(self):
diff --git a/tests/pipeline/engine/test_cost_consistency.py b/tests/pipeline/engine/test_cost_consistency.py
new file mode 100644
index 00000000..0ba9f47c
--- /dev/null
+++ b/tests/pipeline/engine/test_cost_consistency.py
@@ -0,0 +1,215 @@
+from decimal import Decimal
+
+from iac_code.pipeline.engine.cost_consistency import (
+    BUDGET_ABOVE,
+    BUDGET_BELOW,
+    BUDGET_UNKNOWN,
+    BUDGET_WITHIN,
+    evaluate_budget_deviation,
+    evaluate_discount_disclosure,
+    parse_monthly_amounts,
+    reconcile_instance_spec,
+    validate_cost_consistency,
+)
+
+
+def _codes(issues):
+    return [issue.code for issue in issues]
+
+
+class TestParseMonthlyAmounts:
+    def test_parses_list_and_discounted_prices(self):
+        assert parse_monthly_amounts("¥96.80/月（列表价，合同优惠后约¥13.76/月）") == (
+            Decimal("96.80"),
+            Decimal("13.76"),
+        )
+
+    def test_single_amount_has_no_discounted_price(self):
+        assert parse_monthly_amounts("¥281.77/月（列表价）") == (Decimal("281.77"), None)
+
+    def test_thousands_separator_is_normalized(self):
+        assert parse_monthly_amounts("¥1,234.50/月（列表价）") == (Decimal("1234.50"), None)
+
+    def test_pricing_failure_yields_no_amount(self):
+        assert parse_monthly_amounts("询价失败") == (None, None)
+        assert parse_monthly_amounts(None) == (None, None)
+
+
+class TestEvaluateBudgetDeviation:
+    def test_within_planned_range(self):
+        status, actual = evaluate_budget_deviation(
+            {"monthly_min": 50, "monthly_max": 100, "currency": "CNY"},
+            "¥96.80/月（列表价）",
+        )
+        assert status == BUDGET_WITHIN
+        assert actual == Decimal("96.80")
+
+    def test_session_4a2b12c_exceeds_planned_range(self):
+        status, actual = evaluate_budget_deviation(
+            {"monthly_min": 50, "monthly_max": 100, "currency": "CNY"},
+            "¥281.77/月（列表价）",
+        )
+        assert status == BUDGET_ABOVE
+        assert actual == Decimal("281.77")
+
+    def test_session_64ceeb9b_exceeds_planned_range(self):
+        status, actual = evaluate_budget_deviation(
+            {"monthly_min": 200, "monthly_max": 300, "currency": "CNY"},
+            "¥443.29/月（列表价）",
+        )
+        assert status == BUDGET_ABOVE
+        assert actual == Decimal("443.29")
+
+    def test_below_planned_range(self):
+        status, _actual = evaluate_budget_deviation({"monthly_min": 200, "monthly_max": 300}, "¥12/月（列表价）")
+        assert status == BUDGET_BELOW
+
+    def test_missing_baseline_is_unknown(self):
+        assert evaluate_budget_deviation(None, "¥100/月")[0] == BUDGET_UNKNOWN
+        assert evaluate_budget_deviation({"monthly_min": 200}, "¥100/月")[0] == BUDGET_UNKNOWN
+
+    def test_unparsable_estimate_is_unknown(self):
+        assert evaluate_budget_deviation({"monthly_min": 1, "monthly_max": 2}, "询价失败") == (BUDGET_UNKNOWN, None)
+
+
+class TestReconcileInstanceSpec:
+    def test_matching_spec_has_no_issue(self):
+        assert (
+            reconcile_instance_spec(
+                {"instance_type": "ecs.t6-c1m1.large"},
+                {"InstanceType": "ecs.t6-c1m1.large"},
+                {"InstanceType": "ecs.t6-c1m1.large"},
+            )
+            == []
+        )
+
+    def test_session_43672b3_spec_drift_is_reported(self):
+        issues = reconcile_instance_spec(
+            {"instance_type": "ecs.t6-c1m1.large"},
+            {"InstanceType": "ecs.c5.large"},
+            {"InstanceType": "ecs.g5.large"},
+        )
+        assert _codes(issues) == ["spec_deviates_from_plan", "spec_preview_mismatch"]
+
+    def test_missing_pricing_parameter_is_reported(self):
+        issues = reconcile_instance_spec({"instance_type": "ecs.g7.large"}, {}, {})
+        assert _codes(issues) == ["spec_missing_in_deployment_parameters"]
+
+    def test_no_planned_compute_skips_check(self):
+        assert reconcile_instance_spec(None, {"InstanceType": "ecs.g7.large"}, {}) == []
+        assert reconcile_instance_spec({}, {"InstanceType": "ecs.g7.large"}, {}) == []
+
+    def test_case_and_whitespace_are_normalized(self):
+        assert reconcile_instance_spec({"instance_type": " ECS.G7.Large "}, {"InstanceType": "ecs.g7.large"}, {}) == []
+
+    def test_image_id_is_reconciled(self):
+        issues = reconcile_instance_spec(
+            {"image_id": "centos_stream_9_x64_20G_alibase_20260414.vhd"},
+            {"ImageId": "aliyun_3_x64_20G_alibase_20260101.vhd"},
+            {},
+        )
+        assert _codes(issues) == ["spec_deviates_from_plan"]
+
+
+class TestEvaluateDiscountDisclosure:
+    def test_real_reduction_is_accepted(self):
+        assert evaluate_discount_disclosure("¥96.80/月（列表价，合同优惠后约¥13.76/月）") == []
+
+    def test_session_64ceeb9b_identical_prices_are_rejected(self):
+        issues = evaluate_discount_disclosure("¥443.29/月（列表价，合同优惠后约¥443.29/月）")
+        assert _codes(issues) == ["discount_without_reduction"]
+
+    def test_list_price_only_is_accepted(self):
+        assert evaluate_discount_disclosure("¥443.29/月（列表价）") == []
+
+    def test_pricing_failure_is_accepted(self):
+        assert evaluate_discount_disclosure("询价失败") == []
+
+
+class TestValidateCostConsistency:
+    def _conclusion(self, **overrides):
+        conclusion = {
+            "monthly_estimate": "¥96.80/月（列表价，合同优惠后约¥13.76/月）",
+            "deployment_parameters": {"InstanceType": "ecs.t6-c1m1.large"},
+            "preview_validation": {"succeeded": True, "parameters": {"InstanceType": "ecs.t6-c1m1.large"}},
+            "spec_reconciliation": {"instance_type": "ecs.t6-c1m1.large", "matches_plan": True},
+            "budget_deviation": {"status": "within", "actual_monthly": 96.80},
+        }
+        conclusion.update(overrides)
+        return conclusion
+
+    def test_consistent_conclusion_passes(self):
+        assert (
+            validate_cost_consistency(
+                {"instance_type": "ecs.t6-c1m1.large"},
+                {"monthly_min": 50, "monthly_max": 100, "currency": "CNY"},
+                self._conclusion(),
+            )
+            == []
+        )
+
+    def test_no_baseline_skips_all_checks(self):
+        assert validate_cost_consistency(None, None, {"monthly_estimate": "¥281.77/月（列表价）"}) == []
+
+    def test_missing_spec_reconciliation_is_reported(self):
+        conclusion = self._conclusion()
+        del conclusion["spec_reconciliation"]
+        issues = validate_cost_consistency({"instance_type": "ecs.t6-c1m1.large"}, None, conclusion)
+        assert _codes(issues) == ["missing_spec_reconciliation"]
+
+    def test_spec_drift_claimed_as_matching_is_reported(self):
+        conclusion = self._conclusion(
+            deployment_parameters={"InstanceType": "ecs.c5.large"},
+            preview_validation={"succeeded": True, "parameters": {"InstanceType": "ecs.c5.large"}},
+        )
+        issues = validate_cost_consistency({"instance_type": "ecs.t6-c1m1.large"}, None, conclusion)
+        assert _codes(issues) == [
+            "spec_deviates_from_plan",
+            "spec_reconciliation_mismatch",
+            "missing_spec_deviation_note",
+        ]
+
+    def test_declared_spec_deviation_with_note_is_accepted(self):
+        conclusion = self._conclusion(
+            deployment_parameters={"InstanceType": "ecs.c6.xlarge"},
+            preview_validation={"succeeded": True, "parameters": {"InstanceType": "ecs.c6.xlarge"}},
+            spec_reconciliation={
+                "instance_type": "ecs.c6.xlarge",
+                "matches_plan": False,
+                "deviation_note": "规划规格不满足用户 4 vCPU 硬约束",
+            },
+        )
+        issues = validate_cost_consistency({"instance_type": "ecs.t6-c1m1.large"}, None, conclusion)
+        assert _codes(issues) == ["spec_deviates_from_plan"]
+
+    def test_missing_budget_deviation_is_reported(self):
+        conclusion = self._conclusion()
+        del conclusion["budget_deviation"]
+        issues = validate_cost_consistency(None, {"monthly_min": 50, "monthly_max": 100}, conclusion)
+        assert _codes(issues) == ["missing_budget_deviation"]
+
+    def test_unreported_budget_overrun_is_reported(self):
+        conclusion = self._conclusion(monthly_estimate="¥443.29/月（列表价）")
+        issues = validate_cost_consistency(None, {"monthly_min": 200, "monthly_max": 300}, conclusion)
+        assert _codes(issues) == [
+            "budget_deviation_status_mismatch",
+            "budget_deviation_amount_mismatch",
+            "missing_budget_deviation_note",
+        ]
+
+    def test_declared_budget_overrun_with_note_is_accepted(self):
+        conclusion = self._conclusion(
+            monthly_estimate="¥443.29/月（列表价）",
+            budget_deviation={
+                "status": "above",
+                "planned_range": "¥200-300/月",
+                "actual_monthly": 443.29,
+                "note": "通用方案实际费用超出规划区间约 48%",
+            },
+        )
+        assert validate_cost_consistency(None, {"monthly_min": 200, "monthly_max": 300}, conclusion) == []
+
+    def test_ineffective_discount_is_reported(self):
+        conclusion = self._conclusion(monthly_estimate="¥443.29/月（列表价，合同优惠后约¥443.29/月）")
+        issues = validate_cost_consistency(None, None, conclusion)
+        assert _codes(issues) == ["discount_without_reduction"]
diff --git a/tests/pipeline/engine/test_step_executor.py b/tests/pipeline/engine/test_step_executor.py
index 2d596efe..c85dc073 100644
--- a/tests/pipeline/engine/test_step_executor.py
+++ b/tests/pipeline/engine/test_step_executor.py
@@ -2776,11 +2776,15 @@ def test_rich_candidate_resume_uses_compact_schema_and_preserves_first_conclusio
             surface="a2a_rich",
         )
 
-        tool_schema = executor._build_step_tools(
-            step,
-            context,
-            compact_candidate_selection=True,
-        ).get("complete_step").input_schema
+        tool_schema = (
+            executor._build_step_tools(
+                step,
+                context,
+                compact_candidate_selection=True,
+            )
+            .get("complete_step")
+            .input_schema
+        )
         conclusion_schema = tool_schema["properties"]["conclusion"]
         assert conclusion_schema["required"] == [
             "selected_candidate_name",
@@ -2863,11 +2867,15 @@ def test_stale_candidate_conclusion_cannot_enable_compact_resume_schema(self, tm
             context,
             resume_candidate_selection=True,
         )
-        tool_schema = executor._build_step_tools(
-            step,
-            context,
-            compact_candidate_selection=preserved is not None,
-        ).get("complete_step").input_schema
+        tool_schema = (
+            executor._build_step_tools(
+                step,
+                context,
+                compact_candidate_selection=preserved is not None,
+            )
+            .get("complete_step")
+            .input_schema
+        )
 
         assert preserved is None
         conclusion_schema = tool_schema["properties"]["conclusion"]
diff --git a/tests/pipeline/selling/skills/test_iac_aliyun_architecture_skill.py b/tests/pipeline/selling/skills/test_iac_aliyun_architecture_skill.py
index d133f258..9e54cc3b 100644
--- a/tests/pipeline/selling/skills/test_iac_aliyun_architecture_skill.py
+++ b/tests/pipeline/selling/skills/test_iac_aliyun_architecture_skill.py
@@ -48,6 +48,31 @@ def test_architecture_hard_constraint_schema_describes_every_field():
     assert all(value.get("description") for value in properties.values())
 
 
+def test_architecture_candidate_schema_declares_planning_baselines():
+    body = (SKILL_DIR / "SKILL.md").read_text(encoding="utf-8")
+    end = body.index("---", 3)
+    schema = yaml.safe_load(body[3:end])["conclusion_schema"]
+    properties = schema["properties"]["candidates"]["items"]["properties"]
+
+    planned_compute = properties["planned_compute"]
+    assert set(planned_compute["properties"]) == {"instance_type", "image_id"}
+    assert planned_compute["additionalProperties"] is False
+
+    planned_budget = properties["planned_budget"]
+    assert planned_budget["required"] == ["monthly_min", "monthly_max", "currency"]
+    assert planned_budget["properties"]["currency"]["enum"] == ["CNY"]
+    assert all(value.get("description") for value in planned_budget["properties"].values())
+
+
+def test_architecture_body_requires_truthful_planning_baselines():
+    body = (SKILL_DIR / "SKILL.md").read_text(encoding="utf-8")
+
+    assert "## 规划基线：规格与预算" in body
+    assert "planned_compute" in body
+    assert "planned_budget" in body
+    assert "monthly_estimate" in body
+
+
 def test_architecture_prompt_guides_optional_memory_lookup_for_planning_context():
     body = PROMPT_FILE.read_text(encoding="utf-8")
 
diff --git a/tests/pipeline/selling/skills/test_iac_aliyun_cost_skill.py b/tests/pipeline/selling/skills/test_iac_aliyun_cost_skill.py
index 2d46be8d..c3073bdf 100644
--- a/tests/pipeline/selling/skills/test_iac_aliyun_cost_skill.py
+++ b/tests/pipeline/selling/skills/test_iac_aliyun_cost_skill.py
@@ -172,6 +172,30 @@ def test_monthly_estimate_schema_describes_list_and_discounted_prices(self):
         assert "OriginalAmount" in description
         assert "TradeAmount" in description
 
+    def test_monthly_estimate_schema_forbids_discount_caliber_without_reduction(self):
+        content = SKILL_MD.read_text(encoding="utf-8")
+        fm = _parse_frontmatter(content)
+        description = fm["conclusion_schema"]["properties"]["monthly_estimate"]["description"]
+
+        assert "两者相等" in description
+        assert "只填列表价" in description
+
+    def test_conclusion_schema_carries_plan_reconciliation_fields(self):
+        content = SKILL_MD.read_text(encoding="utf-8")
+        fm = _parse_frontmatter(content)
+        properties = fm["conclusion_schema"]["properties"]
+
+        spec = properties["spec_reconciliation"]
+        assert spec["required"] == ["matches_plan"]
+        assert set(spec["properties"]) == {"instance_type", "image_id", "matches_plan", "deviation_note"}
+        assert "planned_compute" in spec["description"]
+
+        budget = properties["budget_deviation"]
+        assert budget["required"] == ["status"]
+        assert budget["properties"]["status"]["enum"] == ["within", "above", "below", "unknown"]
+        assert set(budget["properties"]) == {"status", "planned_range", "actual_monthly", "note"}
+        assert "planned_budget" in budget["description"]
+
     def test_conclusion_schema_requires_full_preview_validation_when_succeeded(self):
         content = SKILL_MD.read_text(encoding="utf-8")
         fm = _parse_frontmatter(content)
@@ -539,6 +563,21 @@ def test_prompt_requires_list_and_discounted_monthly_prices(self):
         assert "合同优惠后" in body
         assert "monthly_estimate" in body
 
+    def test_prompt_forbids_reporting_a_discount_that_does_not_reduce_the_price(self):
+        body = COST_PROMPT_MD.read_text(encoding="utf-8")
+
+        assert "即使数值相同也保留两个价格口径" not in body
+        assert "两者相等表示账户没有产生任何减免" in body
+        assert "不得标注合同优惠后价格" in body
+
+    def test_prompt_passes_planning_baselines_and_requires_reconciliation(self):
+        body = COST_PROMPT_MD.read_text(encoding="utf-8")
+
+        assert "{candidate.planned_compute}" in body
+        assert "{candidate.planned_budget}" in body
+        assert "spec_reconciliation" in body
+        assert "budget_deviation" in body
+
     def test_prompt_receives_only_required_candidate_fields_without_repeating_skill_rules(self):
         body = COST_PROMPT_MD.read_text(encoding="utf-8")
 
diff --git a/tests/pipeline/selling/test_pipeline_reviewing.py b/tests/pipeline/selling/test_pipeline_reviewing.py
index 531ced8a..2f0a2016 100644
--- a/tests/pipeline/selling/test_pipeline_reviewing.py
+++ b/tests/pipeline/selling/test_pipeline_reviewing.py
@@ -44,7 +44,17 @@ def test_review_enabled_loads_infraguard_repair_step_before_cost() -> None:
                 "deployment_parameters_field": "deployment_parameters",
             },
             "message_key": "hard_constraint_verification_required",
-        }
+        },
+        {
+            "always": True,
+            "require_cost_consistency": {
+                "planned_compute_field": "candidate.planned_compute",
+                "planned_budget_field": "candidate.planned_budget",
+                "spec_reconciliation_field": "spec_reconciliation",
+                "budget_deviation_field": "budget_deviation",
+            },
+            "message_key": "cost_plan_consistency_required",
+        },
     ]
 
     assert review_step.tools is not None
diff --git a/tests/pipeline/selling/test_terminal_ui_contract.py b/tests/pipeline/selling/test_terminal_ui_contract.py
index c4e2cd6a..ad9382b7 100644
--- a/tests/pipeline/selling/test_terminal_ui_contract.py
+++ b/tests/pipeline/selling/test_terminal_ui_contract.py
@@ -133,9 +133,7 @@ def test_confirm_prompt_tells_model_to_preserve_parameter_overrides():
 def test_confirm_prompts_share_selection_contract_structure():
     repl_prompt = (_selling_pipeline_dir() / "prompts" / "confirm_and_select.md").read_text(encoding="utf-8")
     a2a_prompt = (_selling_pipeline_dir() / "prompts" / "confirm_and_select.a2a.md").read_text(encoding="utf-8")
-    rich_prompt = (_selling_pipeline_dir() / "prompts" / "confirm_and_select.a2a.rich.md").read_text(
-        encoding="utf-8"
-    )
+    rich_prompt = (_selling_pipeline_dir() / "prompts" / "confirm_and_select.a2a.rich.md").read_text(encoding="utf-8")
 
     shared_fragments = [
         "## 首次执行",
diff --git a/tests/providers/test_manager.py b/tests/providers/test_manager.py
index 8ff0fc1a..54057e5f 100644
--- a/tests/providers/test_manager.py
+++ b/tests/providers/test_manager.py
@@ -1676,9 +1676,7 @@ async def test_complete_records_chat_span_event_and_total_metric(self):
 
     async def test_complete_attributes_bailian_openai_endpoint_to_dashscope_on_all_signals(self):
         mock_provider = AsyncMock()
-        mock_provider._base_url = (
-            "https://llm-testworkspace000000.cn-beijing.maas.aliyuncs.com/compatible-mode/v1"
-        )
+        mock_provider._base_url = "https://llm-testworkspace000000.cn-beijing.maas.aliyuncs.com/compatible-mode/v1"
         mock_provider.complete = AsyncMock(
             return_value=NonStreamingResponse(
                 message_id="complete-response",
diff --git a/tests/skill_bridge/test_iac_code_bridge.py b/tests/skill_bridge/test_iac_code_bridge.py
index b7e9280f..1eafe79e 100644
--- a/tests/skill_bridge/test_iac_code_bridge.py
+++ b/tests/skill_bridge/test_iac_code_bridge.py
@@ -542,12 +542,10 @@ def finish_cleanup(_args):
 
     assert len(captured_payloads) == 2
     assert all(
-        payload["params"]["message"]["metadata"]["iac_code"]["cleanupOnly"] is True
-        for payload in captured_payloads
+        payload["params"]["message"]["metadata"]["iac_code"]["cleanupOnly"] is True for payload in captured_payloads
     )
     assert all(
-        payload["params"]["message"]["metadata"]["iac_code"]["channel"] == "skill/host"
-        for payload in captured_payloads
+        payload["params"]["message"]["metadata"]["iac_code"]["channel"] == "skill/host" for payload in captured_payloads
     )
     assert captured_payloads[0]["params"]["message"]["contextId"] == "ctx-pipeline-1"
     assert result["state"] == "completed"
@@ -1138,9 +1136,7 @@ def test_candidate_presentation_survives_bounded_bridge_projection() -> None:
                                     "summary": "单 ECS 低成本方案。",
                                     "architectureDiagram": "flowchart LR\nU[用户] --> E[ECS]",
                                     "totalMonthlyCost": "¥88/月",
-                                    "costItems": [
-                                        {"name": "ECS", "spec": "2核4G", "monthlyCost": "¥88/月"}
-                                    ],
+                                    "costItems": [{"name": "ECS", "spec": "2核4G", "monthlyCost": "¥88/月"}],
                                 }
                             ],
                             "required": True,
diff --git a/tests/skill_bridge/test_runtime_release.py b/tests/skill_bridge/test_runtime_release.py
index 140ebb59..d6de0578 100644
--- a/tests/skill_bridge/test_runtime_release.py
+++ b/tests/skill_bridge/test_runtime_release.py
@@ -104,9 +104,7 @@ def test_runtime_archive_and_version_marker_are_rooted_consistently(tmp_path: Pa
     assert "artifactRevision" not in marker
 
 
-def test_runtime_a2a_smoke_checks_health_and_agent_card(
-    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
-) -> None:
+def test_runtime_a2a_smoke_checks_health_and_agent_card(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
     module = _load_module("skill_runtime_smoke", BUILD_SCRIPT)
     server = tmp_path / "server.py"
     server.write_text(
diff --git a/tests/web/test_frontend_static.py b/tests/web/test_frontend_static.py
index c3c6f30f..fe9e6f69 100644
--- a/tests/web/test_frontend_static.py
+++ b/tests/web/test_frontend_static.py
@@ -2547,7 +2547,7 @@ def test_completed_turn_collapses_process_into_summary() -> None:
     # 「已处理」组的展开态必须跨重建保留:openKey 让 toggle 记录器登记用户操作、
     # applyDetailsOpenOverrides 在重建后恢复;键取 turnId,缺 turnId 时回退首条消息 id。
     assert 'const turnKey = turnId || text(agentMessages[0]?.messageId || agentMessages[0]?.id || "");' in app_source
-    assert 'details.dataset.openKey = `turnproc:${turnKey}`;' in app_source
+    assert "details.dataset.openKey = `turnproc:${turnKey}`;" in app_source
 
     # 只有最后一次工具调用之后的文本才是「最终回答」;此前每个步骤的文本旁白
     # (夹在工具调用之间的 text delta)连同思考、工具一起折进「已处理」,不平铺成答案。