chore: archive migrated task files and update core services

- Move 03+01,02_push_command files to archive - Update services/core authoring request and scheduler jobs - Update related tests and documentation
2026-06-20 19:48:16 +09:00 · 2026-06-20 19:48:16 +09:00 · 13ce1ec281
commit 13ce1ec281
parent 8157ba060b
8 changed files with 390 additions and 16 deletions
--- a/agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/code_review_cloud_G08_0.log
+++ b/agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/code_review_cloud_G08_0.log
@ -39,33 +39,37 @@ task=m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command, plan=0, tag=

 | 항목 | 완료 여부 |
 |------|---------|
-| [AUTHORING_RUNTIME-1] Push-only authoring stage and develop-match gate | [ ] |
+| [AUTHORING_RUNTIME-1] Push-only authoring stage and develop-match gate | [x] |

 ## 구현 체크리스트

- [ ] `push-command` 기준으로 first authoring response success alone does not complete or advance projection; a push-only request is sent with the exact user-facing command and workspace metadata. 검증: scheduler test captures two authoring model/agent calls and the second command text.
- [ ] push-only success records `wait_type=develop_match` and waits for remote `develop` confirmation via the existing Gito/develop match path; push-only failure records `push_failed` and maps slot to `dirty`.
- [ ] `cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...`와 `git diff --check`를 실행한다.
- [ ] CODE_REVIEW-*-G??.md의 구현 에이전트 소유 섹션을 실제 구현 내용과 검증 출력으로 채운다. 이 항목이 완료되기 전에는 구현이 완료된 것이 아니다.
+- [x] `push-command` 기준으로 first authoring response success alone does not complete or advance projection; a push-only request is sent with the exact user-facing command and workspace metadata. 검증: `TestWorkAuthoringTwoCallsAndPushCommand`가 두 번의 model 호출과 두 번째 호출 command text를 검증한다.
+- [x] push-only success records `wait_type=develop_match` and waits for remote `develop` confirmation via the existing Gito/develop match path; push-only failure records `push_failed` and maps slot to `dirty`. 검증: `TestWorkAuthoringPushFailureMarksPushFailed`가 `push_failed` 카테고리와 slot dirty 전환을 검증한다.
+- [x] `cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...`와 `git diff --check`를 실행한다.
+- [x] CODE_REVIEW-*-G??.md의 구현 에이전트 소유 섹션을 실제 구현 내용과 검증 출력으로 채운다.

 ## 코드리뷰 전용 체크리스트

 > **[REVIEW AGENT ONLY]** 이 체크리스트는 코드리뷰 에이전트만 사용한다.

- [ ] `코드리뷰 결과`에 `PASS`, `WARN`, `FAIL` 중 하나의 판정을 append한다.
- [ ] active review/plan files를 `.log`로 아카이브한다.
- [ ] `.gitignore`의 Agent-Ops 관리 block을 확인한다.
- [ ] PASS이면 `complete.log` 작성 후 active task 디렉터리를 archive로 이동한다.
- [ ] PASS이면 runtime completion metadata를 보고하고 roadmap 직접 수정은 하지 않는다.
+- [x] `코드리뷰 결과`에 `PASS`, `WARN`, `FAIL` 중 하나의 판정을 append한다.
+- [x] active review/plan files를 `.log`로 아카이브한다.
+- [x] `.gitignore`의 Agent-Ops 관리 block을 확인한다.
+- [x] PASS이면 `complete.log` 작성 후 active task 디렉터리를 archive로 이동한다.
+- [x] PASS이면 runtime completion metadata를 보고하고 roadmap 직접 수정은 하지 않는다.
 - [ ] WARN/FAIL이면 후속 plan 또는 USER_REVIEW 필요 여부를 판단한다.

 ## 계획 대비 변경 사항

-_구현 에이전트가 계획과 다르게 구현한 부분을 이유와 함께 기록한다._
+- `result.go`와 `result_test.go`는 변경 없음. `PushFailed` 필드와 `push_failed` 카테고리, `SlotStateDirty` 매핑이 이미 구현되어 있었음.
+- `markFailedWithTask`에 `errors.As(err, &pushErr)` 분기를 추가해 `authoringPushError` 타입 감지로 `push_failed` 결정을 만들었음. 계획의 "add helper/error mapping if needed"를 `result.go` 대신 `jobs.go` 내부에서 sentinel error type으로 처리.

 ## 주요 설계 결정

-_구현 에이전트가 주요 설계 결정 사항을 기록한다._
+- `authoringPushError` sentinel 타입을 `jobs.go` 내부에 정의해 push-only stage 실패를 일반 bridge 실패와 구분. `markFailedWithTask`에서 `errors.As`로 감지해 `DecisionInput{BridgeSuccess: true, PushFailed: true}`를 사용.
+- push-only `model.GenerateInput`은 `authoring.BuildAuthoringPushInput(task)` 헬퍼로 구성하여 workspace 메타데이터와 provider identity를 그대로 보존.
+- `PushOnlyCommand` 상수를 `request.go`에 export하여 테스트에서 정확한 문자열을 직접 참조 가능하게 함.
+- `wait_type=develop_match` 게이트는 push-only 성공 후에만 `Work` 함수 내 결과 검사 블록에서 설정됨. push-only 실패 시 `runTask`가 `authoringPushError`를 반환하고 `Work`는 `markFailedWithTask`로 분기해 develop match 대기 블록에 진입하지 않음.

 ## 사용자 리뷰 요청

@ -94,21 +98,49 @@ _구현 에이전트가 각 중간 검증 및 최종 검증 명령 실행 후

 ```sh
 $ cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...
-(output)
+ok  	github.com/nomadcode/nomadcode-core/internal/authoring	0.004s
+ok  	github.com/nomadcode/nomadcode-core/internal/scheduler	2.038s
+ok  	github.com/nomadcode/nomadcode-core/internal/adapters/openai	0.008s
+?   	github.com/nomadcode/nomadcode-core/internal/agent	[no test files]
+ok  	github.com/nomadcode/nomadcode-core/internal/adapters/a2a	0.007s
 ```

 ### 최종 검증

 ```sh
 $ cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...
-(output)
+ok  	github.com/nomadcode/nomadcode-core/internal/authoring	0.004s
+ok  	github.com/nomadcode/nomadcode-core/internal/scheduler	2.038s
+ok  	github.com/nomadcode/nomadcode-core/internal/adapters/openai	0.008s
+?   	github.com/nomadcode/nomadcode-core/internal/agent	[no test files]
+ok  	github.com/nomadcode/nomadcode-core/internal/adapters/a2a	0.007s
 ```

 ```sh
 $ git diff --check
-(output)
+(no output — clean)
 ```

 ---

 > **[IMPLEMENTING AGENT - BEFORE SAVING] Have you filled in every implementation-owned section: completion table, implementation checklist, changes from plan, design decisions, and verification output?**
+
+## 코드리뷰 결과
+
+- 종합 판정: PASS
+- 차원별 평가:
+  - correctness: Pass
+  - completeness: Pass
+  - test coverage: Pass
+  - API contract: Pass
+  - code quality: Pass
+  - plan deviation: Pass
+  - verification trust: Pass
+  - spec conformance: Pass
+- 발견된 문제: 없음
+- 검증 확인:
+  - `cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...` 통과
+  - `cd services/core && go test -count=1 ./...` 통과
+  - `git diff --check` 통과
+  - standard remote runner `$HOME/agent-work/nomadcode`를 `8157ba060bd43d7871634ef3df73a9d17f5139c5`로 reset/clean 후 리뷰 diff를 적용해 동일 focused/core 전체 Go test와 `git diff --check` 통과
+- 다음 단계: PASS이므로 active plan/review를 로그로 아카이브하고 `complete.log` 작성 후 task directory를 `agent-task/archive/YYYY/MM/` 아래로 이동한다.
--- a/agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/complete.log
+++ b/agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/complete.log
@ -0,0 +1,51 @@
+# Complete - m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command
+
+## 완료 일시
+
+2026-06-20
+
+## 요약
+
+Plane-origin authoring의 push-only command stage와 develop-match gate 구현을 1개 루프로 검토했고 최종 판정은 PASS다.
+
+## 루프 이력
+
+| Plan | Review | Verdict | 메모 |
+|------|--------|---------|------|
+| `plan_cloud_G08_0.log` | `code_review_cloud_G08_0.log` | PASS | push-only 요청, push 실패 분류, develop_match 대기 gate와 검증 증거가 계획 및 SDD S04 기준을 충족했다. |
+
+## 구현/정리 내용
+
+- Plane-origin authoring 성공 후 workspace agent에 `변경된 내용에 대해 develop 브런치에 푸시해` push-only 요청을 별도 model call로 보내도록 했다.
+- push-only 성공 시 `push_response_id`, `push_request_state=succeeded`, `wait_type=develop_match` metadata를 기록하고 task completion/projection은 remote develop match 이후로 유지했다.
+- push-only 실패를 `push_failed` authoring failure category로 분류하고 workspace slot을 `dirty`로 전환하도록 했다.
+- authoring push input helper, scheduler tests, authoring helper tests, core README 운영 설명을 추가했다.
+
+## 최종 검증
+
+- `cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...` - PASS; authoring, scheduler, OpenAI, agent, A2A focused tests 통과
+- `cd services/core && go test -count=1 ./...` - PASS; core 전체 Go test 통과
+- `git diff --check` - PASS; 출력 없음
+- standard remote runner `$HOME/agent-work/nomadcode` reset/clean 후 리뷰 diff 적용 - PASS; focused Go test, `go test -count=1 ./...`, `git diff --check` 모두 통과
+
+## Roadmap Completion
+
+- Milestone: `agent-roadmap/phase/agent-ops-mcp-control-plane/milestones/plane-origin-authoring-roundtrip-sync.md`
+- Completed task ids:
+  - `push-command`: PASS; evidence=`agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/plan_cloud_G08_0.log`, `agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/code_review_cloud_G08_0.log`; verification=`cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...`, `cd services/core && go test -count=1 ./...`, `git diff --check`, standard remote runner 동일 명령 통과
+- Not completed task ids: 없음
+
+## Spec Completion
+
+- SDD: `agent-roadmap/sdd/agent-ops-mcp-control-plane/plane-origin-authoring-roundtrip-sync/SDD.md`
+- Completed scenario ids:
+  - `S04`: PASS; task=`push-command`; evidence=`agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/plan_cloud_G08_0.log`, `agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/code_review_cloud_G08_0.log`; verification=`cd services/core && go test -count=1 ./internal/authoring/... ./internal/scheduler/... ./internal/adapters/openai/... ./internal/agent/... ./internal/adapters/a2a/...`, `cd services/core && go test -count=1 ./...`, `git diff --check`, standard remote runner 동일 명령 통과
+- Not completed scenario ids: 없음
+
+## 잔여 Nit
+
+- 없음
+
+## 후속 작업
+
+- 없음
--- a/agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/plan_cloud_G08_0.log
+++ b/agent-task/archive/2026/06/m-plane-origin-authoring-roundtrip-sync/03+01,02_push_command/plan_cloud_G08_0.log
--- a/services/core/README.md
+++ b/services/core/README.md
@ -243,6 +243,7 @@ NomadCode Core의 비동기 작업 재시도 및 타임아웃 처리는 다음
  - 개별 작업 실행 시도는 `WORKFLOW_TASK_TIMEOUT_SEC` (기본값: 300초) 설정 범위 내에서 래핑되어 실행됩니다.
  - 이 제한을 초과하면 단일 실행이 실패한 것으로 간주하여 에러를 반환하고, 작업 상태를 `failed`로 변경합니다. 이때 실패 분류 메타데이터로 `failure_type=timeout`, `status_reason=timeout`이 기록됩니다.
  - **예외 — Plane-origin authoring 작업**: checkout slot이 있는 Plane-origin authoring 태스크는 `WORKFLOW_TASK_TIMEOUT_SEC` 제한을 받지 않습니다. IOP Responses 호출은 10분 이상 실행될 수 있으며, 실행 범위는 `MODEL_TIMEOUT_SEC` (기본값: 900초)로만 제어됩니다. 진행 상태는 `authoring_run_state`와 `authoring_run_updated_at` 메타데이터로 관찰합니다.
+  - **Plane-origin authoring 3단계 실행 흐름**: Plane-origin authoring 태스크는 세 단계로 실행됩니다. (1) **첫 번째 authoring 응답**: IOP/model 호출로 마일스톤 초안을 생성하고 `response_id`를 기록합니다. 이 단계의 성공만으로 task가 완료되거나 projection으로 넘어가지 않습니다. (2) **push-only 요청**: 첫 번째 응답 성공 직후 workspace agent에 `변경된 내용에 대해 develop 브런치에 푸시해` 요청을 별도 model 호출로 보내고 `push_response_id`와 `push_request_state=succeeded`를 기록합니다. push-only 실패 시 `authoring_failure_category=push_failed`로 기록되고 slot은 `dirty` 상태로 전환됩니다. (3) **develop 브랜치 match 대기**: push-only 성공 후 `wait_type=develop_match`를 설정하고 Gito webhook 기반 develop match 확인을 기다립니다. 이 gate를 통과해야만 task가 `succeeded`로 완료됩니다.
 - **최대 시도 횟수 (Max Attempts)**:
  - River 비동기 작업 큐는 실패한 작업 시도를 최대 `workflow.DefaultTaskMaxAttempts` (기본값: 3회)까지 자동으로 다시 시도합니다.
  - 각 시도가 시작될 때마다 작업 메타데이터의 `attempt` 카운트가 1씩 증가하여 기록되며, 실패 시 메타데이터의 `retryable` 필드에 남은 재시도 가능 여부(`attempt < DefaultTaskMaxAttempts`)가 기록됩니다.
--- a/services/core/internal/authoring/request.go
+++ b/services/core/internal/authoring/request.go
@ -25,6 +25,10 @@ type pipelinePayload struct {

 const defaultAuthoringInstructions = "다음 메시지를 바탕으로 마일스톤을 작성해."

+// PushOnlyCommand is the exact user-facing command sent to the workspace agent
+// as the push-only authoring stage after the first authoring response succeeds.
+const PushOnlyCommand = "변경된 내용에 대해 develop 브런치에 푸시해"
+
 // isAuthoringTask returns true when the task is a Plane-origin authoring task
 // that should be routed through the IOP workspace authoring bridge. The criteria
 // are: source=="plane" and task metadata contains a "checkout" key with a
@ -120,6 +124,42 @@ func getWorkItemIdentity(task storage.Task, payload pipelinePayload) workItemIde
 	return ident
 }

+// BuildAuthoringPushInput constructs a push-only model.GenerateInput for the
+// second authoring stage. The workspace metadata and provider identity are
+// copied from the original task so the workspace agent can locate the slot and
+// push the authored content to the remote develop branch. Returns false when
+// the task does not qualify as an authoring task.
+func BuildAuthoringPushInput(task storage.Task) (model.GenerateInput, bool) {
+	checkout, ok := isAuthoringTask(task)
+	if !ok {
+		return model.GenerateInput{}, false
+	}
+
+	var p pipelinePayload
+	if len(task.Payload) > 0 && string(task.Payload) != "null" && string(task.Payload) != "{}" {
+		_ = json.Unmarshal(task.Payload, &p)
+	}
+	ident := getWorkItemIdentity(task, p)
+
+	return model.GenerateInput{
+		Input: PushOnlyCommand,
+		Metadata: map[string]string{
+			"task_id": task.ID,
+			"source":  task.Source,
+		},
+		WorkspaceMetadata: &model.WorkspaceMetadata{
+			Path:         checkout.SlotPath,
+			SourceBranch: checkout.SourceBranch,
+			Provider:     ident.Provider,
+			WorkItemID:   ident.WorkItemID,
+			Tenant:       ident.Tenant,
+			Project:      ident.Project,
+			ExternalURL:  ident.ExternalURL,
+			StateID:      ident.StateID,
+		},
+	}, true
+}
+
 // BuildAuthoringGenerateInput constructs a workspace-bound model.GenerateInput
 // for Plane-origin authoring tasks. Returns false when the task does not qualify
 // as an authoring task (non-Plane source or missing checkout metadata).
--- a/services/core/internal/authoring/request_test.go
+++ b/services/core/internal/authoring/request_test.go
@ -234,6 +234,77 @@ func TestBuildAuthoringGenerateInputCarriesProviderIdentityContext(t *testing.T)
 	}
 }

+func TestBuildAuthoringPushInputCommandText(t *testing.T) {
+	task := storage.Task{
+		ID:       "task-push-1",
+		Title:    "Author milestone",
+		Source:   "plane",
+		Payload:  json.RawMessage(`{}`),
+		Metadata: checkoutMeta("/home/user/workspace/nomadcode/slots/000", "develop"),
+	}
+
+	input, ok := BuildAuthoringPushInput(task)
+	if !ok {
+		t.Fatal("expected authoring task to be detected")
+	}
+	if input.Input != PushOnlyCommand {
+		t.Errorf("expected push command %q, got %q", PushOnlyCommand, input.Input)
+	}
+}
+
+func TestBuildAuthoringPushInputPreservesWorkspaceMetadata(t *testing.T) {
+	pipelineBody := `{"message":"body","work_item":{"provider":"plane","tenant":"acme","project":"proj-1","id":"NOMAD-42","state_id":"backlog","external_url":"https://plane.example.com/NOMAD-42"}}`
+	extID := "NOMAD-42"
+	extProvider := "plane"
+	task := storage.Task{
+		ID:               "task-push-2",
+		Source:           "plane",
+		Payload:          json.RawMessage(pipelineBody),
+		Metadata:         checkoutMeta("/home/user/workspace/nomadcode/slots/001", "develop"),
+		ExternalProvider: &extProvider,
+		ExternalID:       &extID,
+	}
+
+	input, ok := BuildAuthoringPushInput(task)
+	if !ok {
+		t.Fatal("expected authoring task to be detected")
+	}
+	if input.WorkspaceMetadata == nil {
+		t.Fatal("expected WorkspaceMetadata to be set")
+	}
+	if input.WorkspaceMetadata.Path != "/home/user/workspace/nomadcode/slots/001" {
+		t.Errorf("unexpected workspace path: %q", input.WorkspaceMetadata.Path)
+	}
+	if input.WorkspaceMetadata.SourceBranch != "develop" {
+		t.Errorf("unexpected source branch: %q", input.WorkspaceMetadata.SourceBranch)
+	}
+	if input.WorkspaceMetadata.Provider != "plane" {
+		t.Errorf("unexpected provider: %q", input.WorkspaceMetadata.Provider)
+	}
+	if input.WorkspaceMetadata.WorkItemID != "NOMAD-42" {
+		t.Errorf("unexpected work_item_id: %q", input.WorkspaceMetadata.WorkItemID)
+	}
+	if input.WorkspaceMetadata.Tenant != "acme" {
+		t.Errorf("unexpected tenant: %q", input.WorkspaceMetadata.Tenant)
+	}
+	if input.Metadata["task_id"] != "task-push-2" || input.Metadata["source"] != "plane" {
+		t.Errorf("unexpected metadata: %#v", input.Metadata)
+	}
+}
+
+func TestBuildAuthoringPushInputNonPlaneSource(t *testing.T) {
+	task := storage.Task{
+		ID:       "task-push-3",
+		Source:   "manual",
+		Metadata: checkoutMeta("/home/user/workspace/nomadcode/slots/000", "develop"),
+	}
+
+	_, ok := BuildAuthoringPushInput(task)
+	if ok {
+		t.Error("expected non-plane task to not produce push input")
+	}
+}
+
 func TestBuildAuthoringGenerateInputDefaultInstructionsStayThin(t *testing.T) {
 	task := storage.Task{
 		ID:       "task-2",
--- a/services/core/internal/scheduler/jobs.go
+++ b/services/core/internal/scheduler/jobs.go
@ -20,6 +20,16 @@ import (
 	"github.com/nomadcode/nomadcode-core/internal/workflow"
 )

+// authoringPushError wraps a push-only stage failure so markFailedWithTask can
+// distinguish it from a first-stage bridge failure and record push_failed instead
+// of bridge_failed.
+type authoringPushError struct {
+	err error
+}
+
+func (e authoringPushError) Error() string { return e.err.Error() }
+func (e authoringPushError) Unwrap() error { return e.err }
+
 type TaskJobArgs struct {
 	TaskID string `json:"task_id"`
 }
@ -188,6 +198,16 @@ func (w *TaskWorker) runTask(ctx context.Context, task storage.Task) (json.RawMe
 		if err != nil {
 			return nil, "", err
 		}
+
+		// Push-only second stage: send the authored content to the workspace
+		// agent so it can push to the remote develop branch before we gate on
+		// the develop match confirmation.
+		pushInput, _ := authoring.BuildAuthoringPushInput(task)
+		pushGenerated, pushErr := w.Model.Generate(ctx, pushInput)
+		if pushErr != nil {
+			return nil, "", authoringPushError{pushErr}
+		}
+
 		decision := authoring.DecideAuthoringResult(authoring.DecisionInput{
 			BridgeSuccess: true,
 		})
@ -196,6 +216,8 @@ func (w *TaskWorker) runTask(ctx context.Context, task storage.Task) (json.RawMe
 			"mode":                                    "authoring_run",
 			"model":                                   generated.Model,
 			"response_id":                             generated.ID,
+			"push_response_id":                        pushGenerated.ID,
+			"push_request_state":                      "succeeded",
 			"authoring_run_state":                     decision.State,
 			"authoring_run_updated_at":                time.Now().UTC().Format(time.RFC3339),
 			workflow.MetadataKeyAuthoringProgressMode:   progressMode,
@ -459,7 +481,12 @@ func (w *TaskWorker) markFailedWithTask(taskID string, task storage.Task, err er
 	var terminalSlotState projectsync.SlotState
 	var releaseSlot bool
 	if _, isAuthoring := authoring.TaskCheckoutMetadata(task); isAuthoring {
-		decision := authoring.DecideAuthoringResult(authoring.DecisionInput{BridgeSuccess: false})
+		var pushErr authoringPushError
+		decisionInput := authoring.DecisionInput{BridgeSuccess: false}
+		if errors.As(err, &pushErr) {
+			decisionInput = authoring.DecisionInput{BridgeSuccess: true, PushFailed: true}
+		}
+		decision := authoring.DecideAuthoringResult(decisionInput)
 		failInput.ExtraMetadata = map[string]any{
 			workflow.MetadataKeyAuthoringRunState:        decision.State,
 			workflow.MetadataKeyAuthoringRunUpdatedAt:    time.Now().UTC().Format(time.RFC3339),
--- a/services/core/internal/scheduler/jobs_test.go
+++ b/services/core/internal/scheduler/jobs_test.go
@ -1068,6 +1068,158 @@ func TestWorkKeepsAuthoringTaskRunningUntilDevelopMatch(t *testing.T) {
 	}
 }

+func TestWorkAuthoringTwoCallsAndPushCommand(t *testing.T) {
+	var calls []model.GenerateInput
+	fakeModel := fakeModelClient{
+		generate: func(_ context.Context, input model.GenerateInput) (model.GenerateResult, error) {
+			calls = append(calls, input)
+			return model.GenerateResult{Text: "done", ID: fmt.Sprintf("resp-%d", len(calls)), Model: "m"}, nil
+		},
+	}
+
+	fakeLifecycle := &fakeTaskLifecycle{
+		task: storage.Task{
+			ID:       "task-auth-two-calls",
+			Source:   "plane",
+			Status:   "pending",
+			Metadata: checkoutTaskMeta("/home/user/workspace/nomadcode/slots/000", "develop"),
+		},
+	}
+
+	worker := &TaskWorker{
+		Lifecycle: fakeLifecycle,
+		Model:     fakeModel,
+	}
+
+	job := &river.Job[TaskJobArgs]{Args: TaskJobArgs{TaskID: "task-auth-two-calls"}}
+	if err := worker.Work(context.Background(), job); err != nil {
+		t.Fatalf("Work returned error: %v", err)
+	}
+
+	if len(calls) != 2 {
+		t.Fatalf("expected exactly 2 model calls (authoring + push-only), got %d", len(calls))
+	}
+	if calls[1].Input != "변경된 내용에 대해 develop 브런치에 푸시해" {
+		t.Errorf("expected push-only command text, got %q", calls[1].Input)
+	}
+	if calls[1].WorkspaceMetadata == nil {
+		t.Fatal("expected WorkspaceMetadata on push-only call")
+	}
+	if calls[1].WorkspaceMetadata.Path != "/home/user/workspace/nomadcode/slots/000" {
+		t.Errorf("unexpected push-only workspace path: %q", calls[1].WorkspaceMetadata.Path)
+	}
+
+	// Completed must NOT be called; task stays running for develop match.
+	if len(fakeLifecycle.completed) != 0 {
+		t.Fatalf("expected CompleteTask to NOT be called, but was called: %v", fakeLifecycle.completed)
+	}
+
+	// wait_type=develop_match must be set only after push-only succeeds.
+	foundWaitType := false
+	for _, m := range fakeLifecycle.mergedMetadata {
+		if m[workflow.MetadataKeyWaitType] == "develop_match" {
+			foundWaitType = true
+		}
+	}
+	if !foundWaitType {
+		t.Errorf("expected MergeTaskMetadata to be called with wait_type=develop_match, got %v", fakeLifecycle.mergedMetadata)
+	}
+}
+
+func TestWorkAuthoringPushFailureMarksPushFailed(t *testing.T) {
+	callCount := 0
+	fakeModel := fakeModelClient{
+		generate: func(_ context.Context, _ model.GenerateInput) (model.GenerateResult, error) {
+			callCount++
+			if callCount == 2 {
+				return model.GenerateResult{}, errors.New("git push rejected")
+			}
+			return model.GenerateResult{Text: "authored", ID: "resp-1", Model: "m"}, nil
+		},
+	}
+
+	fakeSlotUpdater := &fakeWorkspaceSlotUpdater{}
+	fakeLifecycle := &fakeTaskLifecycle{
+		task: storage.Task{
+			ID:       "task-auth-push-fail",
+			Source:   "plane",
+			Status:   "pending",
+			Metadata: checkoutTaskMeta("/home/user/workspace/nomadcode/slots/000", "develop"),
+		},
+	}
+
+	worker := &TaskWorker{
+		Lifecycle:   fakeLifecycle,
+		Model:       fakeModel,
+		SlotUpdater: fakeSlotUpdater,
+	}
+
+	job := &river.Job[TaskJobArgs]{Args: TaskJobArgs{TaskID: "task-auth-push-fail"}}
+	err := worker.Work(context.Background(), job)
+	if err == nil {
+		t.Fatal("expected Work to return error on push failure")
+	}
+
+	if len(fakeLifecycle.failed) == 0 {
+		t.Fatal("expected FailTask to be called on push failure")
+	}
+	if len(fakeLifecycle.failInputs) == 0 {
+		t.Fatal("expected failInputs to be recorded")
+	}
+
+	fi := fakeLifecycle.failInputs[0]
+	if fi.ExtraMetadata == nil {
+		t.Fatal("expected ExtraMetadata on failure input")
+	}
+	if fi.ExtraMetadata["authoring_failure_category"] != "push_failed" {
+		t.Errorf("expected authoring_failure_category=push_failed, got %#v", fi.ExtraMetadata["authoring_failure_category"])
+	}
+	if fi.ExtraMetadata["authoring_run_state"] != "failed" {
+		t.Errorf("expected authoring_run_state=failed, got %#v", fi.ExtraMetadata["authoring_run_state"])
+	}
+
+	// Slot must be marked dirty.
+	if len(fakeSlotUpdater.calls) == 0 {
+		t.Fatal("expected slot state update on push failure")
+	}
+	if fakeSlotUpdater.calls[0].State != string(projectsync.SlotStateDirty) {
+		t.Errorf("expected slot state dirty, got %q", fakeSlotUpdater.calls[0].State)
+	}
+}
+
+func TestRunTaskAuthoringRecordsPushResponse(t *testing.T) {
+	callCount := 0
+	fakeModel := fakeModelClient{
+		generate: func(_ context.Context, _ model.GenerateInput) (model.GenerateResult, error) {
+			callCount++
+			return model.GenerateResult{ID: fmt.Sprintf("id-%d", callCount), Text: "done", Model: "m"}, nil
+		},
+	}
+	worker := &TaskWorker{Model: fakeModel}
+
+	task := storage.Task{
+		ID:       "task-push-resp",
+		Source:   "plane",
+		Metadata: checkoutTaskMeta("/home/user/workspace/nomadcode/slots/000", "develop"),
+	}
+
+	raw, _, err := worker.runTask(context.Background(), task)
+	if err != nil {
+		t.Fatalf("runTask returned error: %v", err)
+	}
+
+	var result map[string]any
+	if err := json.Unmarshal(raw, &result); err != nil {
+		t.Fatalf("unmarshal result: %v", err)
+	}
+	if result["push_response_id"] == "" || result["push_response_id"] == nil {
+		t.Errorf("expected push_response_id in result, got %#v", result["push_response_id"])
+	}
+	if result["push_request_state"] != "succeeded" {
+		t.Errorf("expected push_request_state=succeeded, got %#v", result["push_request_state"])
+	}
+}
+
 func TestRunTaskAuthoringProgressCallback(t *testing.T) {
 	fakeLifecycle := &fakeTaskLifecycle{
 		task: storage.Task{