Improve plan review UX with enriched rows, TUI filters, and execute gate.

Make review-plan easier to act on: Chinese risk labels, relative paths,
verdict/next-step footer, optional identity/analysis enrichment, grouping,
spot-check sampling, and structure preview. Extend the TUI with filters,
duplicate-group reject, and quality context. Persist review_context on
plan operations and add --require-review for confirmed execute.

Co-authored-by: Cursor <cursoragent@cursor.com>

src/vlm/commands/execute.py

@@ -42,6 +42,8 @@ def execute_cmd(
     verbose_ops: bool,
     preserve_directories: bool,
     safe_mode: bool,
+    require_review: bool = False,
+    review_csv: Path | None = None,
 ) -> None:
     """Execute an execution plan in dry-run or execute mode."""
     config = ctx.config
@@ -53,6 +55,15 @@ def execute_cmd(
 
     execution_plan = load_plan(plan)
 
+    if require_review and confirm:
+        from vlm.plan_review import check_review_requirements
+
+        csv_path = review_csv or plan.parent / "plan_manual_review.csv"
+        review_errors = check_review_requirements(execution_plan, plan, csv_path)
+        if review_errors:
+            formatted = "\n".join(f"  - {e}" for e in review_errors)
+            raise ValueError(f"REVIEW REQUIRED BEFORE EXECUTE:\n{formatted}")
+
     if preserve_directories:
         click.echo("Directory preservation is enabled (plan metadata/operations will be honored).")
         click.echo()