fix(pipeline): fail deploying step when ROS deployment fails

src/iac_code/i18n/locales/de/LC_MESSAGES/messages.po

@@ -4077,6 +4077,15 @@ msgstr ""
 "Eine erfolgreiche Bereitstellung muss warten, bis ros_deploy "
 "CREATE_COMPLETE zurückgibt."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"A failed deployment must report the ROS status_reason to the user and "
+"then retry, roll back, or return to candidate selection."
+msgstr ""
+"Eine fehlgeschlagene Bereitstellung muss den ROS-status_reason an den "
+"Benutzer melden und danach erneut versuchen, zurückrollen oder zur "
+"Kandidatenauswahl zurückkehren."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Every explicit user hard constraint must be covered by a satisfied check "
@@ -4603,6 +4612,15 @@ msgstr "Der Template-Dateipfad muss relativ zum Arbeitsverzeichnis sein"
 msgid "Template file path cannot escape the working directory"
 msgstr "Der Template-Dateipfad darf das Arbeitsverzeichnis nicht verlassen"
 
+#: src/iac_code/pipeline/engine/step_executor.py
+msgid "no failure reason reported"
+msgstr "kein Fehlergrund gemeldet"
+
+#: src/iac_code/pipeline/engine/step_executor.py
+#, python-brace-format
+msgid "Step {step_id} reported a failed conclusion: {reason}"
+msgstr "Schritt {step_id} hat ein fehlgeschlagenes Ergebnis gemeldet: {reason}"
+
 #: src/iac_code/pipeline/engine/user_input.py
 msgid "[Image input]"
 msgstr "[Bildeingabe]"

src/iac_code/i18n/locales/es/LC_MESSAGES/messages.po

@@ -4055,6 +4055,14 @@ msgstr ""
 "Un despliegue correcto debe esperar hasta que ros_deploy devuelva "
 "CREATE_COMPLETE."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"A failed deployment must report the ROS status_reason to the user and "
+"then retry, roll back, or return to candidate selection."
+msgstr ""
+"Un despliegue fallido debe informar al usuario del status_reason de ROS y"
+" después reintentar, revertir o volver a la selección de candidatos."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Every explicit user hard constraint must be covered by a satisfied check "
@@ -4570,6 +4578,15 @@ msgstr ""
 msgid "Template file path cannot escape the working directory"
 msgstr "La ruta del archivo de plantilla no puede salir del directorio de trabajo"
 
+#: src/iac_code/pipeline/engine/step_executor.py
+msgid "no failure reason reported"
+msgstr "no se informó ningún motivo de fallo"
+
+#: src/iac_code/pipeline/engine/step_executor.py
+#, python-brace-format
+msgid "Step {step_id} reported a failed conclusion: {reason}"
+msgstr "El paso {step_id} informó una conclusión fallida: {reason}"
+
 #: src/iac_code/pipeline/engine/user_input.py
 msgid "[Image input]"
 msgstr "[Entrada de imagen]"

src/iac_code/i18n/locales/fr/LC_MESSAGES/messages.po

@@ -4063,6 +4063,15 @@ msgstr ""
 "Un déploiement réussi doit attendre que ros_deploy retourne "
 "CREATE_COMPLETE."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"A failed deployment must report the ROS status_reason to the user and "
+"then retry, roll back, or return to candidate selection."
+msgstr ""
+"Un déploiement en échec doit signaler le status_reason ROS à "
+"l'utilisateur, puis réessayer, revenir en arrière ou retourner à la "
+"sélection des candidats."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Every explicit user hard constraint must be covered by a satisfied check "
@@ -4580,6 +4589,15 @@ msgstr "Le chemin du fichier de modèle doit être relatif au répertoire de tra
 msgid "Template file path cannot escape the working directory"
 msgstr "Le chemin du fichier de modèle ne peut pas sortir du répertoire de travail"
 
+#: src/iac_code/pipeline/engine/step_executor.py
+msgid "no failure reason reported"
+msgstr "aucune raison d'échec signalée"
+
+#: src/iac_code/pipeline/engine/step_executor.py
+#, python-brace-format
+msgid "Step {step_id} reported a failed conclusion: {reason}"
+msgstr "L'étape {step_id} a signalé une conclusion en échec : {reason}"
+
 #: src/iac_code/pipeline/engine/user_input.py
 msgid "[Image input]"
 msgstr "[Entrée d'image]"

src/iac_code/i18n/locales/ja/LC_MESSAGES/messages.po

@@ -3868,6 +3868,12 @@ msgid ""
 "CREATE_COMPLETE."
 msgstr "デプロイ成功には、ros_deploy が CREATE_COMPLETE を返すまで待つ必要があります。"
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"A failed deployment must report the ROS status_reason to the user and "
+"then retry, roll back, or return to candidate selection."
+msgstr "デプロイ失敗時は ROS の status_reason をユーザーに報告し、その後リトライ、ロールバック、または候補選択に戻る必要があります。"
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Every explicit user hard constraint must be covered by a satisfied check "
@@ -4335,6 +4341,15 @@ msgstr "テンプレートファイルのパスは作業ディレクトリから
 msgid "Template file path cannot escape the working directory"
 msgstr "テンプレートファイルのパスは作業ディレクトリの外に出られません"
 
+#: src/iac_code/pipeline/engine/step_executor.py
+msgid "no failure reason reported"
+msgstr "失敗理由が報告されていません"
+
+#: src/iac_code/pipeline/engine/step_executor.py
+#, python-brace-format
+msgid "Step {step_id} reported a failed conclusion: {reason}"
+msgstr "ステップ {step_id} が失敗の結論を報告しました: {reason}"
+
 #: src/iac_code/pipeline/engine/user_input.py
 msgid "[Image input]"
 msgstr "[画像入力]"

src/iac_code/i18n/locales/pt/LC_MESSAGES/messages.po

@@ -4036,6 +4036,14 @@ msgstr ""
 "Uma implantação bem-sucedida deve aguardar até que ros_deploy retorne "
 "CREATE_COMPLETE."
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"A failed deployment must report the ROS status_reason to the user and "
+"then retry, roll back, or return to candidate selection."
+msgstr ""
+"Uma implantação com falha deve informar o status_reason do ROS ao usuário"
+" e depois tentar novamente, reverter ou voltar à seleção de candidatos."
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Every explicit user hard constraint must be covered by a satisfied check "
@@ -4544,6 +4552,15 @@ msgstr ""
 msgid "Template file path cannot escape the working directory"
 msgstr "O caminho do arquivo de template não pode sair do diretório de trabalho"
 
+#: src/iac_code/pipeline/engine/step_executor.py
+msgid "no failure reason reported"
+msgstr "nenhum motivo de falha informado"
+
+#: src/iac_code/pipeline/engine/step_executor.py
+#, python-brace-format
+msgid "Step {step_id} reported a failed conclusion: {reason}"
+msgstr "A etapa {step_id} informou uma conclusão com falha: {reason}"
+
 #: src/iac_code/pipeline/engine/user_input.py
 msgid "[Image input]"
 msgstr "[Entrada de imagem]"

src/iac_code/i18n/locales/zh/LC_MESSAGES/messages.po

@@ -3832,6 +3832,12 @@ msgid ""
 "CREATE_COMPLETE."
 msgstr "部署成功必须等待 ros_deploy 返回 CREATE_COMPLETE。"
 
+#: src/iac_code/pipeline/engine/complete_step_tool.py
+msgid ""
+"A failed deployment must report the ROS status_reason to the user and "
+"then retry, roll back, or return to candidate selection."
+msgstr "部署失败必须向用户报告 ROS 返回的 status_reason，然后重试、回滚或返回方案选择。"
+
 #: src/iac_code/pipeline/engine/complete_step_tool.py
 msgid ""
 "Every explicit user hard constraint must be covered by a satisfied check "
@@ -4280,6 +4286,15 @@ msgstr "模板文件路径必须是相对于工作目录的路径"
 msgid "Template file path cannot escape the working directory"
 msgstr "模板文件路径不能跳出工作目录"
 
+#: src/iac_code/pipeline/engine/step_executor.py
+msgid "no failure reason reported"
+msgstr "未报告失败原因"
+
+#: src/iac_code/pipeline/engine/step_executor.py
+#, python-brace-format
+msgid "Step {step_id} reported a failed conclusion: {reason}"
+msgstr "步骤 {step_id} 提交了失败结论：{reason}"
+
 #: src/iac_code/pipeline/engine/user_input.py
 msgid "[Image input]"
 msgstr "[图片输入]"

src/iac_code/pipeline/engine/complete_step_tool.py

@@ -56,6 +56,10 @@
         "or confirm that it should not be handled for now."
     ),
     "deploy_wait_create_complete": ("A successful deployment must wait until ros_deploy returns CREATE_COMPLETE."),
+    "deploy_report_failure_reason": (
+        "A failed deployment must report the ROS status_reason to the user and then retry, "
+        "roll back, or return to candidate selection."
+    ),
     "hard_constraint_verification_required": (
         "Every explicit user hard constraint must be covered by a satisfied check with matching parameters "
         "and evidence."
@@ -96,6 +100,10 @@ def _completion_guard_message_i18n_markers() -> tuple[str, ...]:
             "or confirm that it should not be handled for now."
         ),
         _("A successful deployment must wait until ros_deploy returns CREATE_COMPLETE."),
+        _(
+            "A failed deployment must report the ROS status_reason to the user and then retry, "
+            "roll back, or return to candidate selection."
+        ),
         _(
             "Every explicit user hard constraint must be covered by a satisfied check with matching parameters "
             "and evidence."

src/iac_code/pipeline/engine/loader.py

@@ -299,6 +299,7 @@ def _parse_steps(raw_steps: list[dict]) -> list[StepSpec]:
                 completion_guards=_parse_completion_guards(raw.get("completion_guards"), step_id),
                 description=raw.get("description", ""),
                 exit_condition=_parse_exit_condition(raw.get("exit_condition"), step_id),
+                failure_condition=_parse_failure_condition(raw.get("failure_condition"), step_id),
                 a2a_artifacts=_parse_a2a_artifacts(raw.get("a2a_artifacts"), step_id),
                 surface_overrides=_parse_surface_overrides(raw.get("surface_overrides"), step_id),
                 config=_parse_mapping(raw.get("config"), "config", step_id),
@@ -410,6 +411,31 @@ def _parse_exit_condition(raw: dict | None, step_id: str) -> dict | None:
     return raw
 
 
+def _parse_failure_condition(raw: dict | None, step_id: str) -> dict | None:
+    if raw is None:
+        return None
+    if not isinstance(raw, dict) or "field" not in raw or "value" not in raw:
+        raise ValueError(
+            f"Step '{step_id}': failure_condition must be a dict with 'field' and 'value' keys, got {raw!r}"
+        )
+    unsupported = set(raw) - {"field", "value", "reason_fields"}
+    if unsupported:
+        unknown = ", ".join(sorted(str(key) for key in unsupported))
+        raise ValueError(
+            f"Step '{step_id}': failure_condition has unsupported keys: {unknown}; "
+            "supported: field, reason_fields, value"
+        )
+    reason_fields = raw.get("reason_fields")
+    if reason_fields is not None and (
+        not isinstance(reason_fields, list) or not all(isinstance(name, str) and name for name in reason_fields)
+    ):
+        raise ValueError(
+            f"Step '{step_id}': failure_condition.reason_fields must be a list of non-empty strings, "
+            f"got {reason_fields!r}"
+        )
+    return raw
+
+
 def _filter_and_relink(steps: list[StepSpec], feature_flags: dict[str, bool]) -> list[StepSpec]:
     """Remove disabled steps and fix forward links."""
     enabled = [s for s in steps if _is_enabled(s, feature_flags)]

src/iac_code/pipeline/engine/step_executor.py

@@ -15,6 +15,7 @@
 
 from iac_code.agent.message import ContentBlock, Message
 from iac_code.agent.system_prompt import SECTION_BUILDERS, build_base_sections
+from iac_code.i18n import _
 from iac_code.mcp.prompt_dispatch import mcp_prompt_command_stream
 from iac_code.pipeline.engine.complete_step_tool import CompleteStepTool
 from iac_code.pipeline.engine.completion_guard_state import (
@@ -79,6 +80,28 @@ def _completion_guard_tool_result_content(event: ToolResultEvent) -> str:
         return event.result
 
 
+def _failure_error_from_conclusion(step: StepSpec, conclusion: Any) -> str | None:
+    """Return a user-facing error when the conclusion matches the step failure condition."""
+    condition = step.failure_condition
+    if not condition or not isinstance(conclusion, dict):
+        return None
+
+    expected = condition.get("value")
+    actual = conclusion.get(str(condition.get("field", "")))
+    matched = actual is expected if isinstance(expected, bool) else actual == expected
+    if not matched:
+        return None
+
+    reason_fields = condition.get("reason_fields") or []
+    reasons = []
+    for name in reason_fields:
+        value = conclusion.get(name)
+        if isinstance(value, str) and value.strip():
+            reasons.append(f"{name}={value.strip()}")
+    reason = "; ".join(reasons) if reasons else _("no failure reason reported")
+    return _("Step {step_id} reported a failed conclusion: {reason}").format(step_id=step.step_id, reason=reason)
+
+
 @dataclass
 class StepAgentLoopContext:
     """AgentLoop context built by the same path used for step execution."""
@@ -430,15 +453,17 @@ async def consume_complete_step_events(
             conclusion = self._merge_preserved_candidate_selection(preserved_selection, conclusion)
             rollback = complete_step_input.get("rollback_request")
             rollback_tuple = (rollback["target_step"], rollback["reason"]) if rollback else None
+            context.set_conclusion(step.conclusion_field, conclusion)
+            if step.on_exit:
+                step.on_exit(context, conclusion)
+            failure_error = None if rollback_tuple else _failure_error_from_conclusion(step, conclusion)
             step_result = StepResult(
                 step_id=step.step_id,
-                status=StepStatus.COMPLETED,
+                status=StepStatus.FAILED if failure_error else StepStatus.COMPLETED,
                 conclusion=conclusion,
                 rollback_request=rollback_tuple,
+                error=failure_error,
             )
-            context.set_conclusion(step.conclusion_field, conclusion)
-            if step.on_exit:
-                step.on_exit(context, conclusion)
         else:
             step_result = StepResult(
                 step_id=step.step_id,
@@ -610,11 +635,14 @@ def _restore_completed_step_result(
         if isinstance(rollback, dict) and rollback.get("target_step") and rollback.get("reason"):
             rollback_tuple = (str(rollback["target_step"]), str(rollback["reason"]))
 
+        conclusion = conclusion if isinstance(conclusion, dict) else {}
+        failure_error = None if rollback_tuple else _failure_error_from_conclusion(step, conclusion)
         return StepResult(
             step_id=step.step_id,
-            status=StepStatus.COMPLETED,
-            conclusion=conclusion if isinstance(conclusion, dict) else {},
+            status=StepStatus.FAILED if failure_error else StepStatus.COMPLETED,
+            conclusion=conclusion,
             rollback_request=rollback_tuple,
+            error=failure_error,
         )
 
     def _build_full_system_prompt(self, step: StepSpec, context: PipelineContext) -> str:

src/iac_code/pipeline/engine/step_spec.py

@@ -85,6 +85,7 @@ class StepSpec:
     completion_guards: list[dict] = field(default_factory=list)
     description: str = ""
     exit_condition: dict | None = None
+    failure_condition: dict | None = None
     a2a_artifacts: list[A2AArtifactSpec] = field(default_factory=list)
     surface_overrides: dict[str, StepSurfaceOverride] = field(default_factory=dict)
     config: dict[str, Any] = field(default_factory=dict)

src/iac_code/pipeline/selling/hooks/deploying.py

@@ -250,6 +250,19 @@ def on_enter(ctx: PipelineContext) -> None:
     ctx.set_conclusion("selected_plan", normalized)
 
 
+def on_exit(ctx: PipelineContext, conclusion: dict[str, Any]) -> None:
+    """Make sure a failed deployment conclusion always carries a ROS failure reason."""
+    _ = ctx
+    if not isinstance(conclusion, dict) or conclusion.get("status") != "failed":
+        return
+    status_reason = conclusion.get("status_reason")
+    if isinstance(status_reason, str) and status_reason.strip():
+        return
+    error = conclusion.get("error")
+    if isinstance(error, str) and error.strip():
+        conclusion["status_reason"] = error.strip()
+
+
 def on_resource_observed(
     ctx: PipelineContext,
     event: ResourceObservedEvent,

src/iac_code/pipeline/selling/pipeline.yaml

@@ -589,6 +589,10 @@ steps:
     interrupt_judge_failure: pause
     context_fields: [intent, selected_plan, evaluated_candidates]
     hooks_file: hooks/deploying.py
+    failure_condition:
+      field: status
+      value: failed
+      reason_fields: [status_reason, error]
     completion_guards:
       - when_conclusion_field_equals:
           status: success
@@ -600,6 +604,10 @@ steps:
           status_in: [CREATE_COMPLETE]
           match_conclusion_field: stack_id
         message_key: deploy_wait_create_complete
+      - when_conclusion_field_equals:
+          status: failed
+        required_conclusion_field: status_reason
+        message_key: deploy_report_failure_reason
     tools:
       include: []
       exclude:

src/iac_code/pipeline/selling/prompts/deploying.md

@@ -46,8 +46,9 @@
 
 ## 错误处理
 - 模板校验失败 → 就地修复模板后重试（最多 5 轮）
-- 部署失败或等待超时 → 按技能的参数补全与 `ros_deploy` 恢复策略处理
-- 架构层面必须变更（如产品组合不可行）→ rollback_request 到 `architecture_planning`
+- 部署失败或等待超时 → 先把 `ros_deploy` 结果中的 `status` 与 `status_reason` 原文报告给用户，再按技能的参数补全与 `ros_deploy` 恢复策略重试
+- 恢复动作用尽仍不能创建成功 → 用 rollback_request 回到 `confirm_and_select` 重新选择方案/参数，或在架构层面必须变更（如产品组合不可行）时回到 `architecture_planning`
+- 确实无法恢复时才提交 `status: failed`，并同时填写 `status_reason`（ROS 原文）与 `error`（结论摘要）；不得用 `status: success` 或省略 `status_reason` 掩盖失败
 
 ## 注意事项
 - 不要读取项目文件或记忆，所需的上下文已在上方提供。