Summary

The recent change in PR #15136 adds ty check fastapi to scripts/lint.sh, but FastAPI now maintains the static-analysis command set in multiple places with slightly different entrypoints:

• scripts/lint.sh
• .pre-commit-config.yaml
• .github/workflows/pre-commit.yml / CI setup

This creates a drift risk: each time a new checker is added or an existing one changes invocation details, contributors and automation may silently diverge.

Why this matters

Right now, ty is present both in scripts/lint.sh and in .pre-commit-config.yaml, but they are not driven from a single source of truth:

• scripts/lint.sh runs bare executables:
  • mypy fastapi
  • ty check fastapi
  • ruff check ...
  • ruff format ... --check
• .pre-commit-config.yaml runs the same tools through uv run ...

This means future updates can easily land in one path but not the others. PR #15136 is a good example of how easy it is for the maintenance burden to increase with every new tool added.

Concrete risk

A contributor can get different results depending on whether they run:

• bash scripts/lint.sh
• prek run -a
• the GitHub Actions pre-commit workflow

Even if they currently happen to pass, the repo is relying on humans to remember to update several places whenever the lint stack changes.

Suggested fix

Use one source of truth for the static checks. For example:

1. Make .pre-commit-config.yaml call bash scripts/lint.sh (or smaller script fragments), or
2. Make scripts/lint.sh invoke the exact uv run ... commands used by pre-commit, or
3. Generate/centralize the command list so adding a new checker only requires one edit.

A smaller improvement would be to at least align invocation style between the script and pre-commit (uv run ... vs bare executables), so local and automated paths behave more consistently.

Context

This issue was noticed while reviewing the latest changes on master, specifically PR #15136 (👷 Add ty check to lint.sh).