python-error-handling

Define a base exception class inheriting from Exception for your package, then derive specific exceptions like ValidationError or APIError from it. This lets callers catch all package errors with one except clause or handle specific cases individually.

Use exception chaining with `raise NewError(...) from e` inside the except block to keep the original exception as the cause. Use `raise NewError(...) from None` when the original context is irrelevant noise, such as converting a ValueError into a domain validation error.

structlog produces structured, key-value log events ideal for machine parsing and production observability, while stdlib logging with basicConfig is simpler for small scripts. This Skill shows both: structlog for contextual fields like order_id, and logging.basicConfig for quick setup.

Wrap the function in a retry decorator that catches transient exceptions, sleeps with exponential backoff (delay multiplied by 2 to the attempt power), and re-raises the last error after exhausting attempts. Only retry errors classified as transient.

Use a Result type (Ok/Err dataclasses) when errors are expected control flow, such as validation pipelines, making failure paths explicit in type signatures. Reserve exceptions for truly exceptional conditions like infrastructure failures.

Logging passwords, tokens, or credentials in error messages creates security leaks since logs are often aggregated and widely accessible. The Skill's checklist requires scrubbing sensitive fields and logging only safe structured context like identifiers and error types.