kornia-rs build / test workflow

All commands are mirrored in CI. Run them through pixi (preferred) so the toolchain matches CI.

Mandatory pre-PR gate

pixi run pr-check       # skills-check + rust-fmt-check + rust-clippy + rust-check + rust-test

Then self-review: run kornia-code-review on git diff main...HEAD, fix what it finds, and paste its final output (with Verdict: and Skill gaps:) in the PR's Self-review section; pr-validation.yml fails while it is empty. Be ready to explain any line when a reviewer asks.

Equivalent pieces:

pixi run skills-check                 # validate .claude/skills (scripts/check_skills.py)
pixi run rust-fmt-check               # cargo fmt --all -- --check
pixi run rust-clippy                  # clippy --workspace --no-deps --all-targets --features ci -D warnings
pixi run rust-test                    # cargo test --features ci
pixi run rust-test-package <crate>    # single crate (cargo test -p)
cargo test --doc -p <crate>           # doctests for public docs

Conditional gates:

• Python code touched: pixi run py-test (builds with maturin first).
• C++ code touched: pixi run cpp-test (cpp-fmt-check for formatting).
• Cargo.toml version/deps touched: python3 scripts/check_version_pins.py and bash scripts/release_rust.sh --check (also pre-commit hooks).
• New dependency: check advisories (security_audit.yml runs in CI).

Without pixi: cargo test --features ci, cargo test -p kornia-imgproc, cargo clippy --workspace --no-deps --all-targets --features ci -- -D warnings.

PR rules (checked in review; pr-validation.yml fails on a missing Self-review and warns on missing fields)

1. Start the PR body from .github/pull_request_template.md and fill every section (gh pr create --body-file; --body text skips the template). Linked issue: Fixes #123 / Closes #123; say on the issue that you are working on it.
2. Small, one concern, only what the issue asks. Out-of-scope hunks block review (kornia-code-review); split them into their own PRs.
3. Paste the commands you actually ran and their output, including failures and what you did not run.
4. Fill in the AI usage disclosure (see AI_POLICY.md): agent + version, model, effort level, skills applied, and which part a human did. You are the sole responsible author of every line.
5. Perf-sensitive change ⇒ before/after bench numbers (kornia-code-review).
6. Changed a fact the skills describe, hit a non-obvious trap, or found something reusable (a helper, a measured number, a pattern) ⇒ one line in the owning skill (kornia-curation §1, §3). User-visible change ⇒ add changelog.d/<PR>.<type>.md (see changelog.d/README.md); never edit CHANGELOG.md outside a release PR.
7. Reference the existing library implementation (OpenCV, a Rust crate) when porting an algorithm; state deliberate departures from it in the rustdoc.
8. Conventional Commits: feat:, fix:, docs:, refactor:, test:, chore:.
9. New/changed public API needs doc comments; bug fixes need a regression test.
10. An agent posting a comment or review on someone's behalf ends it with Posted on behalf of @<handle> by <Agent> (<actual model>). (not Co-Authored-By:). The PR body is never committed (/PR.md is ignored).

Debugging CI

Reproduce with the exact job command from .github/workflows/rust_lint.yml or rust_test.yml. Clippy warnings are errors.