diff --git a/.github/workflows/ci-guards.yml b/.github/workflows/ci-guards.yml
index 9ca07a3f911..4394b04f5f3 100644
--- a/.github/workflows/ci-guards.yml
+++ b/.github/workflows/ci-guards.yml
@@ -618,6 +618,7 @@ jobs:
           python3 tests/test_ci_focused_test_selectors.py
           python3 tests/test_ci_ui_tests_dispatch.py
           python3 tests/test_ci_pr_media.py
+          python3 tests/test_ci_guard_tests_emit_no_workflow_commands.py
 
       - name: Validate manual macOS package cache recovery
         if: ${{ matrix.group == 'app-host-execution' }}
diff --git a/tests/test-execution.toml b/tests/test-execution.toml
index 69dcb0e8028..03da13d385a 100644
--- a/tests/test-execution.toml
+++ b/tests/test-execution.toml
@@ -1429,3 +1429,7 @@ lane = "linux-guard"
 [[test]]
 path = "tests/test_ci_pr_media.py"
 lane = "linux-guard"
+
+[[test]]
+path = "tests/test_ci_guard_tests_emit_no_workflow_commands.py"
+lane = "linux-guard"
diff --git a/tests/test_ci_guard_tests_emit_no_workflow_commands.py b/tests/test_ci_guard_tests_emit_no_workflow_commands.py
new file mode 100755
index 00000000000..9b9c038bd9f
--- /dev/null
+++ b/tests/test_ci_guard_tests_emit_no_workflow_commands.py
@@ -0,0 +1,83 @@
+#!/usr/bin/env python3
+"""Passing guard tests print no GitHub workflow commands.
+
+The runner reads every line of a step's output that starts with `::` as a
+workflow command, so a fixture that makes a script under test print
+`::error::...` becomes a red annotation on the pull request's run page even
+though the test passed. PR 15160's run 36420353579 showed "No ci-ui-tests.yml
+run titled 'UI tests for CI run 100 attempt 1' appeared" and "https://x/900
+ended success without running the UI tests" from test_ci_ui_tests_dispatch.py,
+which read as real CI failures.
+
+Each module below exercises code that annotates on purpose. It runs here the
+way the guard job runs it, with GITHUB_ACTIONS set, and its output must hold
+no line the runner would parse as a command. A module that fails is left to
+its own step to report: a failing test's captured output may annotate.
+"""
+
+from __future__ import annotations
+
+import os
+import re
+import subprocess
+import sys
+import tempfile
+import unittest
+from concurrent.futures import ThreadPoolExecutor
+from pathlib import Path
+
+ROOT = Path(__file__).resolve().parents[1]
+
+# Guard tests whose scripts under test print annotations.
+ANNOTATING_MODULES = (
+    "tests/test_ci_helper_prebuild_lifecycle.py",
+    "tests/test_ci_late_placement.py",
+    "tests/test_ci_main_regression_bisect.py",
+    "tests/test_ci_parallel_artifact_transport.py",
+    "tests/test_ci_pr_media.py",
+    "tests/test_ci_prune_pr_media.py",
+    "tests/test_ci_ui_tests_dispatch.py",
+    "tests/test_ios_upload_batching.py",
+)
+
+# unittest's progress characters share stderr with the script's output, so a
+# command may follow them on one line of the combined log; the runner reads
+# stdout and stderr as separate streams, where it starts the line.
+COMMAND = re.compile(r"^\s*[.EFsxu]*::(?:error|warning|notice|debug|group|endgroup|add-mask|stop-commands|echo"
+                     r"|set-output|save-state|set-env|add-path|add-matcher|remove-matcher)\b")
+
+
+def run_module(relative: str) -> tuple[int, str]:
+    """Exit code and combined output. Files, not pipes: a descendant a test
+    leaves behind would hold a pipe open and hang the read."""
+    env = {**os.environ, "GITHUB_ACTIONS": "true"}
+    env.pop("GITHUB_OUTPUT", None)
+    env.pop("GITHUB_STEP_SUMMARY", None)
+    with tempfile.TemporaryFile("w+") as output:
+        code = subprocess.run([sys.executable, str(ROOT / relative)], cwd=ROOT, env=env, stdin=subprocess.DEVNULL,
+                              stdout=output, stderr=output, timeout=600, check=False).returncode
+        output.seek(0)
+        return code, output.read()
+
+
+class GuardTestsEmitNoWorkflowCommandsTests(unittest.TestCase):
+    def test_passing_guard_tests_print_no_workflow_commands(self) -> None:
+        with ThreadPoolExecutor(max_workers=len(ANNOTATING_MODULES)) as pool:
+            results = dict(zip(ANNOTATING_MODULES, pool.map(run_module, ANNOTATING_MODULES)))
+        checked = 0
+        for relative, result in results.items():
+            code, output = result
+            with self.subTest(module=relative):
+                if code != 0:
+                    # Its own step reports the failure; say which module this check skipped.
+                    self.skipTest(f"{relative} failed here (exit {code}), so its output was not checked")
+                checked += 1
+                leaked = [line for line in output.splitlines() if COMMAND.match(line)]
+                # repr() quotes each line, so this report is never read as a command itself.
+                self.assertFalse(leaked, f"{relative} printed workflow commands:\n"
+                                 + "\n".join(repr(line) for line in leaked[:10]))
+        self.assertGreater(checked, 0, "no annotating guard test passed, so nothing was checked")
+
+
+if __name__ == "__main__":
+    unittest.main(buffer=True)
diff --git a/tests/test_ci_helper_prebuild_lifecycle.py b/tests/test_ci_helper_prebuild_lifecycle.py
index 460796b336c..107f66fe372 100644
--- a/tests/test_ci_helper_prebuild_lifecycle.py
+++ b/tests/test_ci_helper_prebuild_lifecycle.py
@@ -101,4 +101,4 @@ def test_cancellation_kills_helper_descendants(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)
diff --git a/tests/test_ci_late_placement.py b/tests/test_ci_late_placement.py
index 8f5b84c248d..ac96cd3c5bf 100644
--- a/tests/test_ci_late_placement.py
+++ b/tests/test_ci_late_placement.py
@@ -439,4 +439,4 @@ def test_moved_jobs_leave_the_marker_the_rescue_watch_looks_for(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)
diff --git a/tests/test_ci_main_regression_bisect.py b/tests/test_ci_main_regression_bisect.py
index 988d99a3200..75459b505cf 100644
--- a/tests/test_ci_main_regression_bisect.py
+++ b/tests/test_ci_main_regression_bisect.py
@@ -479,4 +479,4 @@ def test_advances_on_a_schedule_and_after_each_report(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)
diff --git a/tests/test_ci_parallel_artifact_transport.py b/tests/test_ci_parallel_artifact_transport.py
index 23d33c9b0d9..942439c757d 100644
--- a/tests/test_ci_parallel_artifact_transport.py
+++ b/tests/test_ci_parallel_artifact_transport.py
@@ -458,4 +458,4 @@ def test_unknown_kind_cannot_choose_a_destination(self):
 
 
 if __name__ == "__main__":
-    unittest.main(verbosity=2)
+    unittest.main(verbosity=2, buffer=True)
diff --git a/tests/test_ci_pr_media.py b/tests/test_ci_pr_media.py
index ee5284f8119..e051cc11667 100644
--- a/tests/test_ci_pr_media.py
+++ b/tests/test_ci_pr_media.py
@@ -722,4 +722,4 @@ def test_adopts_on_matches_the_product_family(self) -> None:
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)
diff --git a/tests/test_ci_prune_pr_media.py b/tests/test_ci_prune_pr_media.py
index 8a657e16651..55265df0d99 100644
--- a/tests/test_ci_prune_pr_media.py
+++ b/tests/test_ci_prune_pr_media.py
@@ -154,4 +154,4 @@ def test_only_main_prunes_and_a_dry_run_is_the_default(self) -> None:
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)
diff --git a/tests/test_ci_ui_tests_dispatch.py b/tests/test_ci_ui_tests_dispatch.py
index cb135071428..424a8f371e9 100644
--- a/tests/test_ci_ui_tests_dispatch.py
+++ b/tests/test_ci_ui_tests_dispatch.py
@@ -458,4 +458,4 @@ def test_the_dispatching_workflow_runs_from_the_default_branch(self) -> None:
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)
diff --git a/tests/test_ios_upload_batching.py b/tests/test_ios_upload_batching.py
index a9d9d5b0bbb..8bfea6e439e 100644
--- a/tests/test_ios_upload_batching.py
+++ b/tests/test_ios_upload_batching.py
@@ -408,4 +408,4 @@ def test_batching_steps_continue_on_error(self):
 
 
 if __name__ == "__main__":
-    unittest.main()
+    unittest.main(buffer=True)