kta-code-review

Provide the PR number, branch name, or run a review of pending git changes. The reviewer diffs the code, runs a static pre-pass for common defects, and returns prioritized findings with file:line anchors, severity ratings, and minimal fixes.

The review detects GlobalScope usage, swallowed CancellationException, runBlocking in production code, unsynchronized shared mutable state, dropped async Deferreds, and missing cancellation cooperation in CPU loops. Each finding cites the violated structured-concurrency rule and a fix.

No, the review is framework-agnostic and makes no assumptions about DI frameworks, UI toolkits, networking stacks, or build systems. It judges Kotlin language idioms, architecture principles, and stdlib mechanisms rather than recommending specific libraries.

Yes, requesting JSON output produces a structured report containing the target, composite score, recommendation, severity counts, and a findings array with category, file, line, rule, issue, fix, and snippet fields suitable for automation.

Hardcoded credentials, tokens, or keys are flagged as Critical findings. The reviewer redacts the literal value in the report, recommends loading secrets from environment or a secret store, and notes that rotation is required even after deletion.

It is not intended for pure test generation, refactoring without review, or single-function micro-optimization tasks. It also does not execute code, modify files, or post comments to remote systems — it only reads source and produces findings.