test: split test_denied_commands_security.py by behaviour (2 of 2)

.github/file-line-cap-baseline.txt

@@ -107,7 +107,6 @@
 3175 test/test_dashboard_approval.py
 2718 test/test_dashboard_server_composition_contract.py
 2478 test/test_dashboard_updates_coverage.py
-11203 test/test_denied_commands_security.py
 2765 test/test_fargate_launch_engine.py
 4437 test/test_github_pull_request_monitor.py
 5410 test/test_governance_distribution.py

src/kiro_crew/builtin_skills/security-conductor/golden-paths.json

@@ -497,6 +497,41 @@
       "command_or_flow": "test/test_denied_commands_security_hosts_file.py",
       "platform": "any",
       "reason": "The deny fence's own assertions, as a behaviour row rather than a classification: the shell rows of this corpus each ask 'is this one command still permitted', and this file asks the complementary question -- that the real refusals still fire and the documented permits still pass. A fix that widened a rule until every golden path happened to survive would still trip it, which is the failure mode a per-command corpus cannot see. (part of the split test_denied_commands_security.py)"
+    },
+    {
+      "kind": "test",
+      "surface": "S1 bash classifier fence",
+      "command_or_flow": "test/test_denied_commands_security_rows.py",
+      "platform": "any",
+      "reason": "The deny fence's own assertions, as a behaviour row rather than a classification: the shell rows of this corpus each ask 'is this one command still permitted', and this file asks the complementary question -- that the real refusals still fire and the documented permits still pass. A fix that widened a rule until every golden path happened to survive would still trip it, which is the failure mode a per-command corpus cannot see. (part of the split test_denied_commands_security.py)"
+    },
+    {
+      "kind": "test",
+      "surface": "S1 bash classifier fence",
+      "command_or_flow": "test/test_denied_commands_security_mentions.py",
+      "platform": "any",
+      "reason": "The deny fence's own assertions, as a behaviour row rather than a classification: the shell rows of this corpus each ask 'is this one command still permitted', and this file asks the complementary question -- that the real refusals still fire and the documented permits still pass. A fix that widened a rule until every golden path happened to survive would still trip it, which is the failure mode a per-command corpus cannot see. (part of the split test_denied_commands_security.py)"
+    },
+    {
+      "kind": "test",
+      "surface": "S1 bash classifier fence",
+      "command_or_flow": "test/test_denied_commands_security_kill.py",
+      "platform": "any",
+      "reason": "The deny fence's own assertions, as a behaviour row rather than a classification: the shell rows of this corpus each ask 'is this one command still permitted', and this file asks the complementary question -- that the real refusals still fire and the documented permits still pass. A fix that widened a rule until every golden path happened to survive would still trip it, which is the failure mode a per-command corpus cannot see. (part of the split test_denied_commands_security.py)"
+    },
+    {
+      "kind": "test",
+      "surface": "S1 bash classifier fence",
+      "command_or_flow": "test/test_denied_commands_security_mint.py",
+      "platform": "any",
+      "reason": "The deny fence's own assertions, as a behaviour row rather than a classification: the shell rows of this corpus each ask 'is this one command still permitted', and this file asks the complementary question -- that the real refusals still fire and the documented permits still pass. A fix that widened a rule until every golden path happened to survive would still trip it, which is the failure mode a per-command corpus cannot see. (part of the split test_denied_commands_security.py)"
+    },
+    {
+      "kind": "test",
+      "surface": "S1 bash classifier fence",
+      "command_or_flow": "test/test_denied_commands_security_stdin.py",
+      "platform": "any",
+      "reason": "The deny fence's own assertions, as a behaviour row rather than a classification: the shell rows of this corpus each ask 'is this one command still permitted', and this file asks the complementary question -- that the real refusals still fire and the documented permits still pass. A fix that widened a rule until every golden path happened to survive would still trip it, which is the failure mode a per-command corpus cannot see. (part of the split test_denied_commands_security.py)"
     }
   ]
 }