diff --git a/scripts/main_loop.py b/scripts/main_loop.py
index a2fd8ad4..084e4dc4 100644
--- a/scripts/main_loop.py
+++ b/scripts/main_loop.py
@@ -31,7 +31,14 @@ PLAN_LINE_RE = re.compile(
 )
 FINISH_STATUSES = {"done", "blocked", "skipped"}
 ENV_BLOCKED_RE = re.compile(r"^env:([^:]+):(.+)$")
-WORKFLOW_PHASES = {"brainstorming", "planning", "executing", "done", "blocked"}
+WORKFLOW_PHASES = {
+    "brainstorming",
+    "planning",
+    "executing",
+    "done",
+    "blocked",
+    "skipped",
+}
 THREAD_LOCKS: dict[str, threading.Lock] = {}
 THREAD_LOCKS_GUARD = threading.Lock()
 
@@ -51,7 +58,7 @@ def usage() -> str:
         "  -plan PATH\n"
         "  -status done|blocked|skipped\n"
         "  -progress FILE\n"
-        "  -phase brainstorming|planning|executing|done|blocked\n"
+        "  -phase brainstorming|planning|executing|done|blocked|skipped\n"
         "  -spec PATH\n"
         "  -executor NAME\n"
         "  -constraints CSV\n"
@@ -365,13 +372,19 @@ def update_workflow_state(
 
 
 def ensure_all_plans_present(
-    lines: list[str], start_idx: int, end_idx: int, progress_path: Path, plan_keys: list[str]
+    lines: list[str],
+    start_idx: int,
+    end_idx: int,
+    progress_path: Path,
+    plan_keys: list[str],
 ) -> list[tuple[str, str, Optional[str], int]]:
     entries = parse_entries(lines, start_idx, end_idx)
     existing = {plan_key for plan_key, _, _, _ in entries}
     missing = [plan_key for plan_key in plan_keys if plan_key not in existing]
     if missing:
-        insert_lines = [render_plan_line(plan_key, "pending", None) for plan_key in missing]
+        insert_lines = [
+            render_plan_line(plan_key, "pending", None) for plan_key in missing
+        ]
         lines[end_idx:end_idx] = insert_lines
         write_progress_lines(progress_path, lines)
         end_idx += len(insert_lines)
@@ -387,23 +400,29 @@ def filter_existing_entries(
 
 
 def choose_claim_entry(
-    entries: list[tuple[str, str, Optional[str], int]], current_env: Optional[str]
+    entries: list[tuple[str, str, Optional[str], int]],
+    current_env: Optional[str],
+    plan_keys: list[str],
 ) -> Optional[tuple[str, Optional[str], int]]:
-    for plan_key, status, note, idx in entries:
+    entry_by_plan: dict[str, tuple[str, str, Optional[str], int]] = {}
+    for entry in entries:
+        entry_by_plan.setdefault(entry[0], entry)
+    ordered_entries = [
+        entry_by_plan[plan_key] for plan_key in plan_keys if plan_key in entry_by_plan
+    ]
+
+    for plan_key, status, note, idx in ordered_entries:
         if status == "in-progress":
             return plan_key, note, idx
 
-    for plan_key, status, note, idx in entries:
+    for plan_key, status, note, idx in ordered_entries:
         if status == "pending":
             return plan_key, note, idx
-
-    if current_env:
-        for plan_key, status, note, idx in entries:
-            if status != "blocked":
-                continue
-            env_info = parse_env_blocked_note(note)
-            if env_info and env_info[0] == current_env:
-                return plan_key, note, idx
+        if status != "blocked" or not current_env:
+            continue
+        env_info = parse_env_blocked_note(note)
+        if env_info and env_info[0] == current_env:
+            return plan_key, note, idx
 
     return None
 
@@ -419,13 +438,17 @@ def claim_plan(plans_dir: Path, progress_path: Path) -> tuple[int, str]:
     with locked_progress(progress_path):
         lines = load_progress_lines(progress_path)
         try:
-            lines, start_idx, end_idx = ensure_plan_block(lines, progress_path, plan_keys)
+            lines, start_idx, end_idx = ensure_plan_block(
+                lines, progress_path, plan_keys
+            )
         except ValueError as exc:
             return 2, f"ERROR: {exc}"
 
-        entries = ensure_all_plans_present(lines, start_idx, end_idx, progress_path, plan_keys)
+        entries = ensure_all_plans_present(
+            lines, start_idx, end_idx, progress_path, plan_keys
+        )
         entries = filter_existing_entries(entries, plan_keys)
-        chosen = choose_claim_entry(entries, detect_env())
+        chosen = choose_claim_entry(entries, detect_env(), plan_keys)
         if not chosen:
             return 0, "NOOP: no claimable plans"
 
@@ -457,18 +480,24 @@ def finish_plan(
         lines = load_progress_lines(progress_path)
 
         try:
-            lines, start_idx, end_idx = ensure_plan_block(lines, progress_path, [plan_key])
+            lines, start_idx, end_idx = ensure_plan_block(
+                lines, progress_path, [plan_key]
+            )
         except ValueError as exc:
             return 2, f"ERROR: {exc}"
 
         entries = parse_entries(lines, start_idx, end_idx)
         rendered_note = normalize_note(note) if note else None
         updated_line = render_plan_line(plan_key, status, rendered_note)
+        workflow_phase = {
+            "done": "done",
+            "blocked": "blocked",
+            "skipped": "skipped",
+        }[status]
 
         for entry_plan, _, _, idx in entries:
             if entry_plan == plan_key:
                 lines[idx] = updated_line
-                workflow_phase = "done" if status == "done" else "blocked"
                 lines = update_workflow_state(
                     lines,
                     phase=workflow_phase,
@@ -478,7 +507,6 @@ def finish_plan(
                 return 0, updated_line
 
         lines[end_idx:end_idx] = [updated_line]
-        workflow_phase = "done" if status == "done" else "blocked"
         lines = update_workflow_state(
             lines,
             phase=workflow_phase,
diff --git a/templates/AGENT_RULES.template.md b/templates/AGENT_RULES.template.md
index f80a0c68..5044a910 100644
--- a/templates/AGENT_RULES.template.md
+++ b/templates/AGENT_RULES.template.md
@@ -83,7 +83,7 @@
 - `$subagent-driven-development` 仅在 Plan 或平台明确要求时使用，
   不是默认执行器
 - 执行完成后，必须先运行 `main_loop.py finish` 写回状态，
-  再更新 `progress.md` 上半部分摘要
+  再更新 `progress.md` 上半部分摘要，并按主循环收尾要求提交
 
 ### Plan 要求
 
@@ -116,7 +116,8 @@
 - `in-progress`：执行中，用于恢复中断任务
 - `done`：已完成
 - `blocked`：阻塞，需人工介入或切换环境
-- `skipped`：永久跳过，不再执行
+- `skipped`：永久跳过，不再执行，`workflow-state.phase` 也写为
+  `skipped`
 
 `skipped` 如需恢复，必须手动改回 `pending`。
 
@@ -139,8 +140,9 @@ python {{PLAYBOOK_SCRIPTS}}/main_loop.py claim \
 该命令会在锁保护下串行完成三件事：
 
 - 自动识别当前环境：`windows`、`linux`、`darwin`
-- 按顺序选择可执行 Plan：
-  `in-progress` > `pending` > `blocked: env:<当前环境>:...`
+- 已有 `in-progress` 优先恢复
+- 如无 `in-progress`，按 Plan 文件顺序选择第一个可执行 Plan：
+  `pending` 或 `blocked: env:<当前环境>:...`
 - 将选中的 Plan 写成 `in-progress`
 
 这里的锁保护的是 `progress.md` 状态块更新，避免多个 session
@@ -218,10 +220,27 @@ python {{PLAYBOOK_SCRIPTS}}/main_loop.py finish \
 8. 触碰安全红线时立即停止，不继续后续 Plan
 9. 常规模式下可对高风险事项向用户确认；无交互模式按本文件
    的“需要确认的场景”自动处理
-10. 每次 `claim` 只领取一个 Plan；写回后再领取下一个
+10. 每次 `claim` 只领取一个 Plan；写回并完成必要提交后再领取下一个
 11. 全部 Plan 处理完后，统一汇总完成项、阻塞项、跳过项、
     环境需求与待确认事项
 
+### Plan 完成提交契约
+
+- Plan `done` 后必须完成一次 commit，然后才能继续领取下一个 Plan
+- 提交顺序：
+  1. 完成 Plan 约定验证
+  2. 运行 `main_loop.py finish -status done` 写回状态
+  3. 必要时更新 `progress.md` 上半部分摘要和相关 memory
+  4. 检查 `git status --short` 与 diff
+  5. 只提交当前 Plan 相关改动
+- 不得由 `main_loop.py finish` 自动执行 `git commit`
+- 不得把用户已有改动或其他 Plan 的改动混入当前 Plan commit
+- 如 Plan `done` 后没有 diff，必须在回复中说明无提交原因
+- `blocked` / `skipped` 不默认提交代码改动；只有状态留痕或已验证的
+  局部成果需要保留时才提交
+- 工作区保持干净后再领取下一个 Plan；如存在不属于当前 Plan 的脏改动，
+  常规模式先向用户确认，无交互模式写入风险并停止继续领取
+
 ## 通用执行约束
 
 ### 代码与配置修改
@@ -297,6 +316,7 @@ python {{PLAYBOOK_SCRIPTS}}/main_loop.py finish \
 - [ ] 换行符与文件格式正确
 - [ ] 无语法错误或明显运行时错误
 - [ ] 已通过 `main_loop.py finish` 写回 Plan 状态
+- [ ] Plan `done` 后已完成对应 commit，或已说明无 diff 无需提交
 
 ---
 
diff --git a/templates/prompts/coding/close-task.template.md b/templates/prompts/coding/close-task.template.md
index 9895e694..84aac65e 100644
--- a/templates/prompts/coding/close-task.template.md
+++ b/templates/prompts/coding/close-task.template.md
@@ -17,7 +17,8 @@
 
 ## 规则
 
-- 如果任务状态变更，优先通过 `main_loop.py finish` 留痕
+- 如本轮来自 `main_loop.py claim` 且任务状态变更，优先通过
+  `main_loop.py finish` 留痕
 - 未验证内容必须显式说明
 - 只写对下一轮仍重要的信息
 - 不手工改写 `workflow-state` 或 `plan-status` 状态块
@@ -26,15 +27,18 @@
 
 1. 核对已完成项与未完成项
 2. 核对已运行验证与未运行验证
-3. 核对 `main_loop.py finish` 是否已经写回 `plan-status`
-4. 核对 `workflow-state.phase` 是否与当前结果一致
+3. 如本轮来自 `main_loop.py claim`，核对 `main_loop.py finish`
+   是否已经写回 `plan-status`
+4. 如本轮来自 `main_loop.py claim`，核对 `workflow-state.phase`
+   是否与当前结果一致
 5. 如需回写上下文，更新 `active-context`、`progress` 上半部分和 `decisions`
 6. 输出本轮摘要与下一步
 
 ## 状态留痕复核
 
-- `main_loop.py finish` 是否已经写回 `plan-status`
-- `workflow-state.phase` 是否与当前结果一致
+- 如本轮来自 `main_loop.py claim`，`main_loop.py finish` 是否已经写回
+  `plan-status`
+- 如本轮来自 `main_loop.py claim`，`workflow-state.phase` 是否与当前结果一致
 - 如为代码类执行，`workflow-state` 中是否保留了
   `executor=executing-plans` 与既定 `constraints`
 
@@ -59,7 +63,7 @@
 
 ## 停止条件
 
-- 如状态未写回，先完成留痕再收尾
+- 如已领取 Plan 但状态未写回，先完成留痕再收尾
 - 如验证不足以支持交付，停止并标记风险
 
 ---
diff --git a/templates/prompts/coding/verify-change.template.md b/templates/prompts/coding/verify-change.template.md
index 96333222..2c8806cf 100644
--- a/templates/prompts/coding/verify-change.template.md
+++ b/templates/prompts/coding/verify-change.template.md
@@ -31,7 +31,8 @@
 2. 运行与本次改动直接相关的验证命令
 3. 记录命令、结果和关键输出
 4. 复核 diff 是否只包含预期修改
-5. 复核 `workflow-state.phase`、`plan-status` 与当前声明一致
+5. 如本轮来自 `main_loop.py claim`，复核 `workflow-state.phase`、
+   `plan-status` 与当前声明一致
 6. 汇总未覆盖项和剩余风险
 
 ## 输出协议
@@ -52,8 +53,9 @@
 
 ## 状态留痕复核
 
-- `workflow-state.phase` 是否与当前声明一致
-- `plan-status` 是否已经通过 `main_loop.py finish` 写回
+- 如本轮来自 `main_loop.py claim`，`workflow-state.phase` 是否与当前声明一致
+- 如本轮来自 `main_loop.py claim`，`plan-status` 是否已经通过
+  `main_loop.py finish` 写回
 - 如为代码类任务，`workflow-state` 中是否保留：
   `executor=executing-plans`
   `constraints=karpathy-guidelines,.agents,AGENT_RULES`
diff --git a/tests/test_main_loop_cli.py b/tests/test_main_loop_cli.py
index 8c9e9192..0a27559d 100644
--- a/tests/test_main_loop_cli.py
+++ b/tests/test_main_loop_cli.py
@@ -245,6 +245,55 @@ class MainLoopCliTests(unittest.TestCase):
             text = progress.read_text(encoding="utf-8")
             self.assertIn(f"`2026-01-05-env.md` in-progress: {note}", text)
 
+    def test_claim_prefers_earlier_env_blocked_plan_over_later_pending_plan(self):
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            root = Path(tmp_dir)
+            plans_dir = root / "docs" / "superpowers" / "plans"
+            plans_dir.mkdir(parents=True)
+            (plans_dir / "2026-01-01-env.md").write_text("env", encoding="utf-8")
+            (plans_dir / "2026-01-02-pending.md").write_text(
+                "pending", encoding="utf-8"
+            )
+
+            progress = root / "memory-bank" / "progress.md"
+            progress.parent.mkdir(parents=True)
+            env = self._current_env()
+            note = f"env:{env}:Task2"
+            progress.write_text(
+                "\n".join(
+                    [
+                        "# Plan 状态",
+                        "",
+                        "<!-- plan-status:start -->",
+                        f"- [ ] `2026-01-01-env.md` blocked: {note}",
+                        "- [ ] `2026-01-02-pending.md` pending",
+                        "<!-- plan-status:end -->",
+                        "",
+                    ]
+                ),
+                encoding="utf-8",
+            )
+
+            result = run_cli(
+                "claim",
+                "-plans",
+                "docs/superpowers/plans",
+                "-progress",
+                "memory-bank/progress.md",
+                cwd=root,
+            )
+
+            self.assertEqual(result.returncode, 0, msg=result.stderr)
+            self.assertEqual(
+                result.stdout.strip(),
+                "\n".join(
+                    [
+                        "PLAN=docs/superpowers/plans/2026-01-01-env.md",
+                        f"NOTE={note}",
+                    ]
+                ),
+            )
+
     def test_finish_updates_line(self):
         with tempfile.TemporaryDirectory() as tmp_dir:
             root = Path(tmp_dir)
@@ -329,13 +378,62 @@ class MainLoopCliTests(unittest.TestCase):
             self.assertEqual(result.returncode, 0, msg=result.stderr)
             text = progress.read_text(encoding="utf-8")
             self.assertIn("phase: done", text)
-            self.assertIn("spec: docs/superpowers/specs/2026-05-18-demo-design.md", text)
+            self.assertIn(
+                "spec: docs/superpowers/specs/2026-05-18-demo-design.md", text
+            )
             self.assertIn("executor: executing-plans", text)
             self.assertIn(
                 "constraints: karpathy-guidelines,.agents,AGENT_RULES",
                 text,
             )
 
+    def test_finish_skipped_updates_workflow_phase_to_skipped(self):
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            root = Path(tmp_dir)
+            progress = root / "memory-bank" / "progress.md"
+            progress.parent.mkdir(parents=True)
+            progress.write_text(
+                "\n".join(
+                    [
+                        "# 当前进展",
+                        "",
+                        "## Workflow State",
+                        "",
+                        "<!-- workflow-state:start -->",
+                        "phase: executing",
+                        "plan: docs/superpowers/plans/2026-05-18-demo.md",
+                        "<!-- workflow-state:end -->",
+                        "",
+                        "## Plan Status",
+                        "",
+                        "<!-- plan-status:start -->",
+                        "- [ ] `2026-05-18-demo.md` in-progress",
+                        "<!-- plan-status:end -->",
+                        "",
+                    ]
+                )
+                + "\n",
+                encoding="utf-8",
+            )
+
+            result = run_cli(
+                "finish",
+                "-plan",
+                "docs/superpowers/plans/2026-05-18-demo.md",
+                "-status",
+                "skipped",
+                "-progress",
+                "memory-bank/progress.md",
+                "-note",
+                "obsolete",
+                cwd=root,
+            )
+
+            self.assertEqual(result.returncode, 0, msg=result.stderr)
+            text = progress.read_text(encoding="utf-8")
+            self.assertIn("phase: skipped", text)
+            self.assertIn("- [ ] `2026-05-18-demo.md` skipped: obsolete", text)
+
     def test_record_updates_workflow_state_block(self):
         with tempfile.TemporaryDirectory() as tmp_dir:
             root = Path(tmp_dir)
@@ -378,7 +476,9 @@ class MainLoopCliTests(unittest.TestCase):
             text = progress.read_text(encoding="utf-8")
             self.assertIn("<!-- workflow-state:start -->", text)
             self.assertIn("phase: planning", text)
-            self.assertIn("spec: docs/superpowers/specs/2026-05-18-demo-design.md", text)
+            self.assertIn(
+                "spec: docs/superpowers/specs/2026-05-18-demo-design.md", text
+            )
             self.assertIn("plan: docs/superpowers/plans/2026-05-18-demo.md", text)
             self.assertIn("executor: executing-plans", text)
             self.assertIn(
@@ -450,7 +550,9 @@ class MainLoopCliTests(unittest.TestCase):
 
             text = progress.read_text(encoding="utf-8")
             self.assertIn("phase: done", text)
-            self.assertIn("spec: docs/superpowers/specs/2026-05-18-demo-design.md", text)
+            self.assertIn(
+                "spec: docs/superpowers/specs/2026-05-18-demo-design.md", text
+            )
             self.assertIn("plan: docs/superpowers/plans/2026-05-18-demo.md", text)
             self.assertIn("executor: executing-plans", text)
             self.assertIn(
@@ -513,7 +615,9 @@ class MainLoopCliTests(unittest.TestCase):
 
             text = progress.read_text(encoding="utf-8")
             self.assertIn("phase: planning", text)
-            self.assertIn("spec: docs/superpowers/specs/2026-05-18-demo-design.md", text)
+            self.assertIn(
+                "spec: docs/superpowers/specs/2026-05-18-demo-design.md", text
+            )
             self.assertIn("plan: docs/superpowers/plans/2026-05-18-demo.md", text)
             self.assertIn("executor: executing-plans", text)
             self.assertIn(
@@ -571,7 +675,9 @@ class MainLoopCliTests(unittest.TestCase):
 
             text = progress.read_text(encoding="utf-8")
             self.assertIn("phase: planning", text)
-            self.assertIn("spec: docs/superpowers/specs/2026-05-18-demo-design.md", text)
+            self.assertIn(
+                "spec: docs/superpowers/specs/2026-05-18-demo-design.md", text
+            )
             self.assertIn("plan: docs/superpowers/plans/2026-05-18-demo.md", text)
             self.assertIn("executor: executing-plans", text)
             self.assertIn(
diff --git a/tests/test_sync_templates_placeholders.py b/tests/test_sync_templates_placeholders.py
index 895b4868..15e7512c 100644
--- a/tests/test_sync_templates_placeholders.py
+++ b/tests/test_sync_templates_placeholders.py
@@ -76,6 +76,7 @@ class SyncTemplatesPlaceholdersTests(unittest.TestCase):
             ROOT / "templates" / "prompts" / "coding" / "close-task.template.md"
         ).read_text(encoding="utf-8")
         self.assertIn("main_loop.py finish", close_task_template)
+        self.assertIn("如本轮来自 `main_loop.py claim`", close_task_template)
         self.assertIn("workflow-state.phase", close_task_template)
         self.assertIn("## Completed", close_task_template)
         self.assertIn("## Not Completed", close_task_template)
@@ -86,6 +87,7 @@ class SyncTemplatesPlaceholdersTests(unittest.TestCase):
         verify_change_template = (
             ROOT / "templates" / "prompts" / "coding" / "verify-change.template.md"
         ).read_text(encoding="utf-8")
+        self.assertIn("如本轮来自 `main_loop.py claim`", verify_change_template)
         self.assertIn("workflow-state.phase", verify_change_template)
         self.assertIn("plan-status", verify_change_template)
         self.assertIn("## Validated", verify_change_template)
@@ -108,9 +110,9 @@ class SyncTemplatesPlaceholdersTests(unittest.TestCase):
         self.assertIn("## Open Questions", code_review_template)
         self.assertIn("## Residual Risk", code_review_template)
 
-        prompts_readme = (
-            ROOT / "templates" / "prompts" / "README.md"
-        ).read_text(encoding="utf-8")
+        prompts_readme = (ROOT / "templates" / "prompts" / "README.md").read_text(
+            encoding="utf-8"
+        )
         self.assertIn("AGENT_RULES.md", prompts_readme)
         self.assertIn("docs/superpowers/", prompts_readme)
         self.assertIn("不是流程权威", prompts_readme)
@@ -132,16 +134,27 @@ class SyncTemplatesPlaceholdersTests(unittest.TestCase):
         self.assertNotIn("main_loop.py claim", agent_behavior_template)
         self.assertNotIn("playbook.py -record-spec", agent_behavior_template)
         self.assertNotIn("playbook.py -record-plan", agent_behavior_template)
-        self.assertIn("项目上下文与执行状态写入 `memory-bank/`", agent_behavior_template)
+        self.assertIn(
+            "项目上下文与执行状态写入 `memory-bank/`", agent_behavior_template
+        )
 
-        rules_template = (
-            ROOT / "templates" / "AGENT_RULES.template.md"
-        ).read_text(encoding="utf-8")
+        rules_template = (ROOT / "templates" / "AGENT_RULES.template.md").read_text(
+            encoding="utf-8"
+        )
         self.assertIn("唯一流程约束中心", rules_template)
         self.assertIn("唯一设计与计划产物中心", rules_template)
         self.assertIn("memory-bank/progress.md", rules_template)
+        self.assertIn("已有 `in-progress` 优先恢复", rules_template)
+        self.assertIn("按 Plan 文件顺序选择第一个可执行 Plan", rules_template)
+        self.assertIn("Plan `done` 后必须完成一次 commit", rules_template)
+        self.assertIn(
+            "不得由 `main_loop.py finish` 自动执行 `git commit`", rules_template
+        )
+        self.assertIn("工作区保持干净后再领取下一个 Plan", rules_template)
 
-    def test_sync_templates_replaces_playbook_scripts_without_main_language_support(self):
+    def test_sync_templates_replaces_playbook_scripts_without_main_language_support(
+        self,
+    ):
         with tempfile.TemporaryDirectory() as tmp_dir:
             config_body = f"""
 [playbook]