[CAT-UNVERIFIED tag](6) show-me-your-work names the tag, and guard the whole repo

product/skills/show-me-your-work/SKILL.md

@@ -128,7 +128,7 @@ the audit trail and the run's transcript, then flags what the user should
 pay attention to. Not a redo of the work — a scan for what's risky.
 
 If the harness cannot spawn a different-family model, still list flags and
-mark the Attention line `UNVERIFIED: same-model self-review`.
+mark the Attention line `{{CAT-UNVERIFIED: the review -- cannot verify: same-model self-review}}`.
 
 - Decisions logged with weak or absent evidence.
 - Verification steps skipped or claimed without proof in the transcript.

tests/test_escape_hatch_vocabulary.py

@@ -0,0 +1,50 @@
+#!/usr/bin/env python3
+"""No rule or skill tells an author to write the retired bare `UNVERIFIED:`.
+
+Every evidence hook clears on the `{{CAT-UNVERIFIED}}` tag and nothing else,
+so prose that still says "write `UNVERIFIED:`" sends an agent into a block.
+
+Run: python3 -m unittest discover -s tests -v
+"""
+import os
+import sys
+import unittest
+
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+import escape_hatch_vocab as vocab  # noqa: E402
+
+
+class InstructionalProseUsesTheTag(unittest.TestCase):
+    def test_no_instructional_file_names_the_retired_marker(self):
+        offenders = []
+        for path in vocab.instructional_files():
+            with open(path, encoding="utf-8") as handle:
+                for line in vocab.instructs_retired_marker(handle.read()):
+                    offenders.append(f"{os.path.relpath(path, vocab.REPO_ROOT)}: {line[:100]}")
+        self.assertEqual(offenders, [])
+
+    def test_scan_covers_the_always_loaded_rules(self):
+        scanned = {os.path.relpath(p, vocab.REPO_ROOT) for p in vocab.instructional_files()}
+        for rel in ("engine/CLAUDE.core.md", "corpus/CLAUDE.learned.md", "always-on/evidence-check.md"):
+            self.assertIn(rel, scanned)
+
+
+class DetectorShape(unittest.TestCase):
+    def test_bare_marker_instruction_is_flagged(self):
+        self.assertEqual(len(vocab.instructs_retired_marker("- Prefix `UNVERIFIED:` until then.")), 1)
+
+    def test_tag_is_not_flagged(self):
+        text = "Tag it `{{CAT-UNVERIFIED: <claim> -- cannot verify: <reason>}}`."
+        self.assertEqual(vocab.instructs_retired_marker(text), [])
+
+    def test_line_calling_the_marker_retired_is_not_flagged(self):
+        self.assertEqual(vocab.instructs_retired_marker("Bare `UNVERIFIED:` is retired."), [])
+
+    def test_other_uses_of_the_word_are_not_flagged(self):
+        text = "| Codex | UNVERIFIED schema | UNVERIFIED end-to-end |"
+        self.assertEqual(vocab.instructs_retired_marker(text), [])
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_show_me_your_work.py

@@ -129,5 +129,15 @@ def test_strips_tabs_and_newlines_from_cells(self):
             self.assertNotIn("\n", cells[3])
 
 
+class TestEscapeHatchVocabulary(unittest.TestCase):
+    def test_attention_line_uses_the_tag_not_the_retired_marker(self):
+        import sys
+        sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+        import escape_hatch_vocab as vocab
+        text = read_skill()
+        self.assertEqual(vocab.instructs_retired_marker(text), [])
+        self.assertIn("{{CAT-UNVERIFIED: the review -- cannot verify: same-model self-review}}", text)
+
+
 if __name__ == "__main__":
     unittest.main()