Transaction Correctness

Use this skill to review whether a workflow remains correct under retries, concurrency, partial failure, and race conditions.

What To Inspect

• Which writes must succeed or fail together.
• What happens if the same request is retried.
• Which reads need freshness or locking.
• Whether side effects occur inside or outside the transaction boundary.
• How state can be corrupted by concurrent actors.

Correctness Checks

• Atomicity of related changes.
• Isolation level assumptions.
• Idempotency for retries and duplicate delivery.
• Ordering guarantees for dependent updates.
• Recovery behavior after partial failure.

Good Output

• Transaction boundary recommendation.
• Concurrency or race-condition risks.
• Idempotency gaps.
• Safer sequencing or locking suggestions.
• Invariants that must always hold.

Common Failure Modes

• Read-modify-write flows without protection.
• External side effects that cannot be rolled back.
• Assuming default isolation solves business-level races.
• Missing uniqueness or state guards for retries.