diff --git a/.claude/skills/tortoise-orm/SKILL.md b/.claude/skills/tortoise-orm/SKILL.md
new file mode 100644
index 000000000..9ab93807e
--- /dev/null
+++ b/.claude/skills/tortoise-orm/SKILL.md
@@ -0,0 +1,259 @@
+---
+name: tortoise-orm
+description: Tortoise ORM project conventions and development workflow. Use when working on Tortoise ORM models, fields, querysets, database backends, migrations, framework integrations, documentation, or tests in this repository.
+---
+
+# Tortoise ORM
+
+Use this skill when changing Tortoise ORM itself, its integrations, docs, examples, or tests.
+
+Tortoise ORM is an async-native Python ORM inspired by Django. Keep the public API explicit,
+typed where practical, and predictable across SQLite, PostgreSQL, MySQL, Microsoft SQL Server,
+and Oracle.
+
+## Use uv and the Makefile
+
+Prefer the repository's Makefile targets over ad hoc commands.
+
+Install dependencies:
+
+```bash
+make deps
+```
+
+Run the default test suite against in-memory SQLite:
+
+```bash
+make test
+```
+
+Run the faster SQLite test target without coverage report generation:
+
+```bash
+make test_fast
+```
+
+Run style and lint checks:
+
+```bash
+make check
+```
+
+Auto-format and auto-fix lint where supported:
+
+```bash
+make style
+```
+
+Run the fuller lint pipeline (style plus mypy, bandit, and `twine check`):
+
+```bash
+make lint
+```
+
+Build the docs:
+
+```bash
+make docs
+```
+
+When running individual commands, keep using `uv run --frozen` so dependencies match `uv.lock`.
+
+## Project layout
+
+Core package code lives in `tortoise/`.
+
+* `tortoise/models.py`: model metadata, model class behavior, relation setup.
+* `tortoise/fields/`: field definitions and DB type mapping.
+* `tortoise/queryset.py`, `tortoise/query_utils.py`, `tortoise/expressions.py`: query construction.
+* `tortoise/connection.py`, `tortoise/transactions.py`, `tortoise/router.py`: connection registry, transactions, and routing.
+* `tortoise/signals.py`, `tortoise/manager.py`, `tortoise/functions.py`, `tortoise/filters.py`, `tortoise/indexes.py`, `tortoise/validators.py`, `tortoise/exceptions.py`: cross-cutting features and DB primitives.
+* `tortoise/backends/`: backend-specific clients, executors, schema generators, and SQL quirks.
+* `tortoise/contrib/`: optional integrations such as FastAPI, Pydantic (`tortoise/contrib/pydantic/`), testing helpers, and linters.
+* `tortoise/migrations/`: built-in migration framework: autodetector, executor, graph, loader, writer, recorder, operations, plus `schema_editor/`, `schema_generator/`, and `api/`.
+* `tortoise/cli/`: CLI entry point and subcommand wiring (`init`, `makemigrations`, `migrate`, `upgrade`, `downgrade`, `history`, `sqlmigrate`, `heads`, `shell`).
+* `tests/`: library tests and backend behavior coverage. The root `conftest.py` is the test entry point and reads `TORTOISE_TEST_DB`.
+* `examples/`: runnable framework and migration examples.
+* `docs/`: Sphinx documentation in reStructuredText.
+
+## API design priorities
+
+Follow the repository's contribution guide priorities:
+
+* Correctness before ease of use.
+* Ease of use before performance.
+* Performance before maintenance cost.
+* Keep behavior explicit and simple.
+* Add tests for behavior changes.
+
+Avoid surprising implicit behavior in models, querysets, migrations, and backend adapters. If a
+change affects generated SQL, schema generation, relation resolution, or initialization order,
+write tests that make the new contract clear.
+
+## Async ORM conventions
+
+Tortoise ORM public I/O APIs are async. Do not hide blocking database work inside sync helpers.
+
+Use `async def` for functions that await database operations. Keep pure metadata and query-building
+helpers sync when they do not perform I/O.
+
+Preserve lazy query construction. QuerySet methods should generally return new or cloned querysets
+until execution is explicitly awaited.
+
+Be careful with connection and transaction context. Do not introduce global state unless it is part
+of the existing connection registry or initialization lifecycle.
+
+## Models and fields
+
+Keep model metadata changes centralized around `MetaInfo`, model initialization, and field setup.
+
+When changing fields:
+
+* Preserve Python type expectations and database serialization/deserialization symmetry.
+* Check backend-specific DB type mappings.
+* Keep `source_field`, `model_field_name`, and relation metadata consistent.
+* Add tests for validation, schema generation, query filters, and serialization when relevant.
+
+When changing table, schema, or relation behavior, cover inheritance and abstract-model cases. Also
+check generated table names, through tables, indexes, constraints, and backward-compatible defaults.
+
+## Query and expression behavior
+
+When changing queries, verify both the Python-facing API and generated SQL.
+
+Prefer existing abstractions in `query_utils`, `expressions`, `functions`, and backend executors
+instead of hand-building SQL strings in high-level code.
+
+Add tests for:
+
+* Filtering and excluding.
+* Related-field traversal.
+* Ordering and distinct behavior.
+* Aggregates, functions, and annotations.
+* Empty inputs and null handling.
+* Backend-specific SQL differences when applicable.
+
+## Database backends
+
+Keep backend-specific behavior inside the backend packages unless the contract truly belongs in
+shared code.
+
+Run targeted tests with the matching Makefile target when a change touches a backend:
+
+```bash
+make test_postgres_asyncpg
+make test_postgres_psycopg
+make test_mysql
+make test_mysql_asyncmy
+make test_mssql
+make test_oracle
+```
+
+Use environment variables from the Makefile for database passwords and drivers. SQLite tests should
+remain the first local verification path for most changes.
+
+## Migrations
+
+For migration framework or CLI changes, update tests and examples that exercise:
+
+* `tortoise init`
+* `tortoise makemigrations`
+* `tortoise migrate` (and its alias `tortoise upgrade`)
+* `tortoise downgrade`
+* `tortoise history`
+* `tortoise sqlmigrate`
+* `tortoise heads`
+
+Migration examples to consult and keep aligned with framework changes:
+
+* `examples/migrations_project/`: minimal single-app migrations.
+* `examples/multiapp_migrations_project/`: cross-app dependencies.
+* `examples/schema_migrations_project/`: schema-focused migrations.
+* `examples/comprehensive_migrations_project/`: end-to-end coverage; CI exercises this one.
+
+Migration changes should preserve deterministic output so generated files are stable across
+repeated runs.
+
+## Pydantic and framework integrations
+
+For Pydantic changes, remember that this project targets Pydantic v2. The integration code lives
+in `tortoise/contrib/pydantic/`. Keep generated schemas, computed fields, relation prefetching,
+and serialization behavior explicit.
+
+For FastAPI, Starlette, Sanic, BlackSheep, aiohttp, or Quart integration changes, keep examples and
+docs aligned with implementation. The CI runs the FastAPI, BlackSheep, and Sanic example tests from
+`.github/workflows/ci.yml`; use those examples as executable documentation.
+
+FastAPI example checks (mirroring CI):
+
+```bash
+PYTHONPATH=examples/fastapi uv run --frozen tortoise -c config.TORTOISE_ORM migrate
+PYTHONPATH=examples/fastapi uv run --frozen pytest -n auto --cov=tortoise --cov-append --cov-branch --tb=native -q examples/fastapi/_tests.py
+```
+
+## Testing guidance
+
+Use the smallest meaningful test first, then broaden when behavior crosses module boundaries. The
+root `conftest.py` is the test entry point and reads the `TORTOISE_TEST_DB` environment variable
+to pick the backend, so set it explicitly for targeted runs.
+
+For local changes, prefer targeted pytest runs:
+
+```bash
+TORTOISE_TEST_DB=sqlite://:memory: uv run --frozen pytest tests/test_file.py::test_name -q
+```
+
+Then run an appropriate Makefile target. For broad ORM behavior changes, run at least SQLite tests
+and mention any database-specific test targets that could not be run locally.
+
+On Windows, the contribution guide notes that running the full suite natively is not supported yet;
+WSL is the recommended path for full test runs.
+
+## Documentation
+
+Update docs when changing public APIs, integration behavior, migrations, or examples.
+
+Docs are Sphinx/reStructuredText. Keep examples importable and consistent with files under
+`examples/`. If docs include literal examples from files, update the source file rather than
+duplicating stale snippets.
+
+Build docs with:
+
+```bash
+make docs
+```
+
+## Style
+
+Follow `pyproject.toml`:
+
+* Python `>=3.10`.
+* Ruff line length 100.
+* Use future annotations where the existing file does.
+* Keep imports sorted by Ruff.
+* Prefer type annotations for public APIs and complex internal helpers.
+
+Use comments sparingly. Add them only when they explain non-obvious ORM, SQL, migration, or
+backend behavior.
+
+## Compatibility
+
+Treat public model, field, queryset, connection, migration, and contrib APIs as compatibility
+sensitive. When changing behavior, check existing docs and tests for implied guarantees.
+
+Do not remove deprecations or change defaults casually. If a change introduces a new option, make
+the default preserve existing behavior unless the task explicitly calls for a breaking change.
+
+## Do not do this
+
+Do not hand-roll SQL in high-level APIs when backend abstractions already exist.
+
+Do not make async APIs perform blocking work.
+
+Do not add broad abstractions without tests that show the shared behavior.
+
+Do not update generated lock files, formatting, docs, or unrelated modules unless the task requires
+it.
+
+Do not assume a single database backend. If behavior differs by backend, make the contract explicit
+and test the difference.