fix(dynamodb-mcp-server): validate and escape schema names in Python generator

[LeeroyHannigan]

Summary

The Python data-access-layer generator (repo_generation_tool) renders source code through
Jinja2 with autoescaping disabled, and nothing downstream escaped schema values, so values
from schema.json were written into the generated code verbatim.

Schema validation did not prevent this. It checked required fields, JSON types, enum values
and uniqueness, but applied no character or identifier constraint to any name — while a
comment in jinja2_generator.py asserted the opposite ("Schema validation ensures all inputs
are safe before template rendering"). That comment is corrected here.

This applies to the Python generator the same discipline #4384 applied to the CDK generator.
Values are constrained during schema validation, which already gates rendering (generate()
returns before create_generator() when validation fails), so an invalid schema is rejected
before any template runs.

Approach

A new core/name_validator.py holds the rules, applied according to where a value lands in
the generated code — the position determines what is possible, so one rule does not fit all:

Position	Rule	Applies to
Identifier (class/def/attribute/import name)	must be a valid non-keyword Python identifier	entity names, field names, access pattern names, parameter names, key template placeholders
Identifier fragment (method-name suffix)	identifier characters only; may start with a digit	GSI names
String literal / docstring / comment	no quotes, backslashes, newlines or braces	table names, key attribute names, entity types, index names
Prose	as above but apostrophes allowed	descriptions

Escaping is impossible in an identifier position, which is why those values must be rejected
rather than escaped. Conversely, DynamoDB permits almost any character in an attribute name,
so string-literal positions are not restricted to an identifier — spaces and non-ASCII text
remain valid.

Braces are rejected in string-literal positions because several of those positions are
f-strings, where a brace is a live expression rather than literal text.

Key templates are validated separately: they are rendered into f-strings, so a brace region
that is not a supported {placeholder} is rejected. The existing placeholder regex matches
only word-character placeholders and silently ignores any other brace region, so such a region
previously reached an f-string unchecked.

Sinks covered

Beyond the names above, these reached generated code and no validator visited them:

• parameter default — rendered into a generated def signature. Validated here, and the
  generator now also escapes it with json.dumps so correctness does not rest on validation
  alone.
• parameter description — reaches the generated method docstring.
• filter_expression param / param2 / params[] — reach a docstring and comments.
• cross-table pattern name and description — reach a def name and its docstring in the
  transaction service.
• entities_involved[].condition — reaches comments in the transaction service.
• table-level gsi_list entries — GSIValidator validated their structure but never their
  characters, while their key attribute names reach generated query code.

Separately, the usage_data string escaper handled backslashes and double quotes but not
newlines, so a newline in a sample value produced an unterminated string literal in the
generated example. It now escapes newlines and carriage returns too.

Confirmed already constrained and deliberately left alone: range_condition and
projection_type are restricted to enum allowlists; item_type is never interpolated and
unknown values map to Any; parameter entity_type and entities_involved[].entity are
membership-checked against entity names, which this change makes safe.

Behaviour change

A schema whose entity or field names are not valid Python identifiers is now rejected with a
clear validation error instead of producing code that cannot be imported. Worth a release note.

Testing

• New tests/repo_generation_tool/unit/test_name_validator.py: unit coverage per rule plus
  end-to-end assertions that a schema targeting each sink fails validation and writes no file.
• Verified per sink that a payload which previously reached generated code is now rejected. For
  the two that produced executable output, executability was confirmed beforehand by parsing
  the generated file and finding the payload as a live AST node rather than as text in a
  comment — and after the change both are rejected during validation.
• Legitimate schemas still generate unchanged, including an attribute name containing a space
  and a description containing an apostrophe (36 of the 447 descriptions in the repo fixtures
  contain apostrophes, which is why prose allows them).
• Full package suite: 1582 passed. ruff check, ruff format --check and pyright clean.
• pre-commit could not run end to end in my environment: its gitleaks hook cannot fetch Go
  modules through my network. The Python hooks it wraps (ruff, pyright) were run directly.

One pre-existing property test asserted that arbitrary text is a valid GSI name, which encoded
the unsafe behaviour; its strategy is narrowed to names usable in generated code, with
rejection covered by the new tests.

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.

[codecov]

Codecov Report

❌ Patch coverage is 93.19728% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 93.27%. Comparing base (e9f2439) to head (eb5cc9d).
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
...rver/repo_generation_tool/core/schema_validator.py	88.13%	1 Missing and 6 partials ⚠️
...repo_generation_tool/core/cross_table_validator.py	71.42%	0 Missing and 2 partials ⚠️
...eneration_tool/core/filter_expression_validator.py	87.50%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##             main    #4579    +/-   ##
========================================
  Coverage   93.27%   93.27%            
========================================
  Files        1053     1054     +1     
  Lines       89208    89344   +136     
  Branches    14385    14441    +56     
========================================
+ Hits        83205    83337   +132     
+ Misses       3626     3624     -2     
- Partials     2377     2383     +6     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.