diff --git a/scripts/make-detection-fixture.py b/scripts/make-detection-fixture.py
index 8b4399604..a5b4e9110 100755
--- a/scripts/make-detection-fixture.py
+++ b/scripts/make-detection-fixture.py
@@ -71,6 +71,30 @@ def canonical_tail(content: str, limit: int = DETECTOR_TAIL_LIMIT) -> str:
     return "\n".join(lines[start:])
 
 
+# The runtime names `DetectedAgent` carries, which are also the `<runtime>` path
+# component of a fixture. Two of them are not the case name (`cursor-agent`,
+# `qodercli`), so an unconstrained flag is easy to get wrong — and every wrong
+# value would take the bounded-tail branch below without saying so.
+# `test_agent_vocabulary_matches_the_swift_enum` fails when this list drifts.
+DETECTED_AGENTS = (
+    "pi",
+    "omp",
+    "claude",
+    "codex",
+    "gemini",
+    "cursor-agent",
+    "cline",
+    "opencode",
+    "copilot",
+    "kimi",
+    "droid",
+    "amp",
+    "qodercli",
+    "qwen",
+    "grok",
+)
+
+
 def detection_screen_text(text: str, agent: str) -> str:
     """Port of `DetectedAgent.detectionScreenText(from:)`.
 
@@ -234,10 +258,11 @@ def main() -> int:
     parser.add_argument(
         "--agent",
         required=True,
+        choices=DETECTED_AGENTS,
         metavar="AGENT",
         help="detector the fixture targets (the <runtime> path component, e.g. claude "
         "or codex); claude keeps the full screen, every other agent takes the "
-        "bounded 24-line tail",
+        "bounded 24-line tail. One of: " + ", ".join(DETECTED_AGENTS),
     )
     parser.add_argument(
         "--redact",
diff --git a/scripts/test_make_detection_fixture.py b/scripts/test_make_detection_fixture.py
index 90875f899..da3707619 100644
--- a/scripts/test_make_detection_fixture.py
+++ b/scripts/test_make_detection_fixture.py
@@ -15,12 +15,16 @@
 import importlib.util
 import json
 import pathlib
+import re
 import subprocess
 import sys
 import tempfile
 import unittest
 
 SCRIPT = pathlib.Path(__file__).resolve().parent / "make-detection-fixture.py"
+DETECTED_AGENT_SWIFT = (
+    SCRIPT.parent.parent / "supacode/Domain/AgentDetection/DetectedAgent.swift"
+)
 
 _spec = importlib.util.spec_from_file_location("make_detection_fixture", SCRIPT)
 fixture = importlib.util.module_from_spec(_spec)
@@ -116,6 +120,18 @@ def test_several_amounts_on_one_row(self):
         self.assertEqual(len(applied), 2)
 
 
+
+class AgentVocabulary(unittest.TestCase):
+    def test_agent_vocabulary_matches_the_swift_enum(self):
+        # The generator mirrors the detector's dispatch, so its agent names have
+        # to be the detector's. A hand-copied list drifts silently; this fails.
+        source = DETECTED_AGENT_SWIFT.read_text(encoding="utf-8")
+        body = source.split("enum DetectedAgent", 1)[1].split("var id:", 1)[0]
+        cases = re.findall(r'^\s*case (\w+)(?:\s*=\s*"([^"]+)")?\s*$', body, re.M)
+        self.assertTrue(cases, "no cases parsed from DetectedAgent.swift")
+        self.assertEqual(list(fixture.DETECTED_AGENTS), [raw or name for name, raw in cases])
+
+
 class Reduction(unittest.TestCase):
     def test_claude_keeps_the_full_screen(self):
         # Mirrors `DetectedAgent.detectionScreenText(from:)`: trimming a Claude
@@ -175,6 +191,13 @@ def test_summary_never_names_what_it_replaced(self):
         self.assertIn("/Users/usr", result.stderr)
         self.assertIn("1 line", result.stderr)
 
+    def test_unknown_agent_is_rejected(self):
+        # Before the flag was constrained, "Claude" was not "claude" and so took
+        # the bounded-tail branch, silently dropping rows from a Claude capture.
+        result = self.run_script("row", agent="Claude")
+        self.assertEqual(result.returncode, 2)
+        self.assertIn("invalid choice", result.stderr)
+
     def test_undecidable_replacement_fails_the_run(self):
         result = self.run_script("│ owner name │", "--redact", "owner=❤️")
         self.assertEqual(result.returncode, 1)