diff --git a/.agents/docs/code-review.md b/.agents/docs/code-review.md
index 179858260..3092e419a 100644
--- a/.agents/docs/code-review.md
+++ b/.agents/docs/code-review.md
@@ -20,6 +20,10 @@ line, not as background reading.
 - Never take a claim as fact, neither the PR description's nor a comment's. Verify game-data
   claims (spell/creature/quest ids, mechanics) against the world DB, DBC data, or cited sources;
   verify "fixed in the latest push" against the current diff.
+- When a change touches control flow (early returns, new branches, removed guards), trace the
+  states it can leave behind (null, empty, fall-through) into every consumer of them — including
+  unchanged lines the new flow now reaches — and verify what runs when an acquire/attack/GetX
+  call leaves a null result, not just what enables it.
 - Check the change is still needed against current `master`: the surrounding code may have moved,
   or another change may have landed the same fix.
 - On an existing PR, walk every discussion item one by one: what was raised, whether it was
diff --git a/.agents/docs/self-review-rules.md b/.agents/docs/self-review-rules.md
index e52bb98c5..7e5e427f1 100644
--- a/.agents/docs/self-review-rules.md
+++ b/.agents/docs/self-review-rules.md
@@ -24,4 +24,6 @@ Every change gets the full review; go deeper the farther it can reach:
 PRs are expected to be tested in-game, which the reviewer cannot do. Never guess what the
 author already tested — ask them, and record the answer in the report. Then tell them what
 else to test — especially side effects they might not expect: a fix for X that also reaches Y
-means testing Y too. Invite them to ask questions if anything is unclear.
+means testing Y too. When the author's tested scenario matches the change's main path, probe
+the branches it doesn't take — regressions hide there. Invite them to ask questions if
+anything is unclear.