Fix subdir guard, qpe cache, migrations. Support latest claude code

.github/workflows/ci.yml

@@ -71,7 +71,6 @@ jobs:
             | MI (normalized) | ${qpe.mi_normalized.toFixed(3)} | Maintainability Index / 100 |
             | Smell Penalty | ${qpe.smell_penalty.toFixed(3)} | Weighted code smell deduction |
             | Adjusted Quality | ${qpe.adjusted_quality.toFixed(3)} | MI × (1 - smell_penalty) + bonuses |
-            | Effort Factor | ${qpe.effort_factor.toFixed(2)} | log(Halstead Effort + 1) |
 
             <details>
             <summary>Code Smell Breakdown</summary>
@@ -82,7 +81,7 @@ jobs:
 
             </details>
 
-            > Higher QPE = better quality per unit effort`;
+            > Higher QPE = better quality`;
 
             const { data: comments } = await github.rest.issues.listComments({
               owner: context.repo.owner,

CLAUDE.md

@@ -174,8 +174,8 @@ You workflow should be incremental with a stepping back review phase after each
 - Note any design choices or decisions you are not sure about and request the user for comments
 - Look for and point out dead or duplicated code branches that could be DRYed
 - Double check the README.md or other documentation to remove outdated sections
-- Be wary of unconstrained and overly generic types in arguments and returns. Introduce ADT using dedicated Pydantic BaseModel and a domain-driven approach, if the current ones are not sufficient. Use pattern matches on these types instead of hasattr/getattr decomposition
+- Be wary of unconstrained and overly generic types in function signatures, and make sure type hints are present in all signatures. Introduce ADT using dedicated Pydantic BaseModel and a domain-driven approach, if the current ones are not sufficient. Use pattern matches on these types instead of hasattr/getattr decomposition
 - **Leverage Pydantic validation**: When adding new configuration parameters or architectural constraints, use Pydantic field validators (`@field_validator`) to catch errors early with helpful messages
 - **Use domain models**: Replace isinstance/hasattr patterns with domain objects that use pydantic's `BaseModel`
 - Always run tests with pytest as final verification step for newly added code
-- any use of `hasattr` and `.get()` with defaults and similar existence checks on objects are code smells (!) and indication that proper configuration or domain object models need to be reviewed
+- any use of `hasattr` and `.get()` with defaults and similar existence checks on objects are code smells (!) and indication that proper configuration or domain object models need to be reviewed