diff --git a/.agents/rules/testing.md b/.agents/rules/testing.md
index cd56f69105..21a0191aa6 100644
--- a/.agents/rules/testing.md
+++ b/.agents/rules/testing.md
@@ -41,3 +41,6 @@
   Bzlmod with `target_compatible_with = SUPPORTS_BZLMOD`
   (`//tests/support:support.bzl`). WORKSPACE mode resolves pip dependencies for
   only the host Python version.
+* **Non-Obvious Test Constructs**: Add an inline comment when a test target
+  uses an intentionally unusual construct (e.g., a `select()` with only
+  `//conditions:default` to verify macro `select()` handling).
diff --git a/.agents/skills/rule-creator/SKILL.md b/.agents/skills/rule-creator/SKILL.md
index 34306e79d1..4a6201f80d 100644
--- a/.agents/skills/rule-creator/SKILL.md
+++ b/.agents/skills/rule-creator/SKILL.md
@@ -32,8 +32,17 @@ trigger: <trigger-condition>
 
 #### Trigger Conditions
 *   `always_on`: The rule is always active and must be followed for all tasks.
-*   Custom triggers: You can specify other trigger conditions if the rule only
-    applies in certain contexts.
+*   `model_decision`: The agent loads the rule based on its `description` when
+    relevant. Prefer `model_decision` for conceptual, architectural, or
+    cross-cutting rules (e.g., `pr.md`, `rules_python.md`).
+*   `glob`: The rule activates when matching files (`globs:`) are accessed.
+    Use for syntax/formatting rules tied to specific file types.
+
+### Domain Separation
+*   Keep language and tool rule files (`starlark.md`, `bzl.md`, `python.md`,
+    `docs.md`) general to those domains.
+*   Place `rules_python`-specific Bzlmod, PyPI, and rule API invariants in
+    `.agents/rules/rules_python.md`.
 
 ### Formatting Guidelines
 *   **Line Wrapping:** Wrap all text to 80 columns.