Fix skill token limits: scale for 35 reference files after spec growth

Limits were calibrated for 16 patterns but the project grew to 22 patterns
(35 total ref files) through planned spec work. Since skill references load
on-demand, total limits reflect maintainability not context pressure.

Author: mdi-odoo

references/patterns/token-budget.md

@@ -17,8 +17,8 @@ Guidelines for keeping commands efficient and within limits.
 |-----------|-------|
 | SKILL.md | 5,000 tokens / 500 lines |
 | Single reference | 2,000 tokens |
-| All references | 10,000 tokens |
-| Total skill | 15,000 tokens |
+| All references | 30,000 tokens |
+| Total skill | 35,000 tokens |
 
 ## Why Token Budgets Matter
 

scripts/count-tokens.py

@@ -67,21 +67,23 @@ class TokenReport(TypedDict):
     "command_lines": 600,
 }
 
-# Skill-level limits (for self-validation of the generator directory)
+# Skill-level limits (for self-validation of the generator directory).
+# References are loaded on-demand, not all at once, so total limits
+# reflect maintainability rather than context window pressure.
 SKILL_LIMITS = {
     "skill_md_tokens": 5000,
     "skill_md_lines": 500,
     "single_ref_tokens": 2000,
-    "total_ref_tokens": 10000,
-    "total_skill_tokens": 15000,
+    "total_ref_tokens": 30000,
+    "total_skill_tokens": 35000,
 }
 
 SKILL_HARD_LIMITS = {
     "skill_md_tokens": 10000,
     "skill_md_lines": 1000,
     "single_ref_tokens": 4000,
-    "total_ref_tokens": 20000,
-    "total_skill_tokens": 30000,
+    "total_ref_tokens": 60000,
+    "total_skill_tokens": 70000,
 }