---

name: gh-breaking-changes
description: "Audits a diff or change for everything that could break callers, dependents, consumers, or existing data."

/gh-breaking-changes

Breaking changes are the ones you didn't know were breaking. You renamed a function, changed a response shape, dropped a database column, or altered default behaviour — and something downstream broke in a way you didn't anticipate. This audit catches them before merge, not after an incident.

Analyse the changes described in this conversation for breaking changes across each layer:

API contracts
Any change to a public-facing API endpoint: URL, method, request shape, response shape, status codes, error formats, or authentication requirements. Partial changes count — adding a required field breaks existing callers just as much as removing one.

Internal interfaces
Function signatures, class constructors, exported types, or module interfaces that other parts of the codebase depend on. Renamed parameters, changed return types, added required arguments — all potentially breaking.

Database schema
Column renames, type changes, dropped columns, new NOT NULL constraints without defaults, index removals, changed foreign key constraints. Any migration that can't be rolled back without data loss.

Behavioural changes
The same input now produces different output. Default values changed. Error conditions that previously succeeded now fail. Timing or ordering of operations changed.

Configuration and environment
New required environment variables. Changed config file formats. Renamed or removed feature flags.

Clients and consumers
Who calls this code? List the known callers and for each: does this change break them? What do they need to update?

For each breaking change found:

• What breaks — specific and concrete
• Who is affected — known callers or consumers
• Migration path — what needs to change on the consumer side
• Can it be made backwards-compatible? — flag if a deprecation period is possible

Rules:

• A change is breaking if any existing code that works today would fail after this change without modification
• Internal-only changes are still breaking if they affect other teams or services
• Flag anything that requires a coordinated deployment — changes that can't be rolled out independently