ApiShift — OpenAPI Release Review Kit

An API contract changed. Which consumers might need attention before release?

ApiShift turns a baseline and candidate OpenAPI contract into a review you can inspect: exact structural changes, referenced operations, request-versus-response impact, unknowns and a focused release checklist. Use it with your own AI agent to prepare a review in English or Spanish.

You receive an original skill workflow, direction-sensitive review rules, a local Python inventory, fictional example contracts and a worked review. The inventory produces changes.json, changes.md, a review template and a manifest with input/output hashes. Your agent completes a five-section release-review.md with evidence, affected consumers, minimal repair or migration options and tests to run.

Designed for developers and small API teams who already have two contracts. It helps discuss new required inputs, response changes, component references, parameters, authentication overrides and generated-client concerns without connecting to the API.

Requirements: buyer-selected AI agent/client with local file access and execution, Python3.10+, and sanitized single-file UTF-8 OpenAPI3.0.0–3.0.4 JSON. Maximum2MiB per file,64nesting levels,10000structural changes. Actual development checks ran on Python3.12/Linux; Python3.10/Windows/macOS and universal client/model compatibility were not executed. A text-only chat cannot run the package. Your agent/model may have separate costs.

Important limits: the script inventories structure and performs only minimum input-shape checks. It is not an OpenAPI validator or exhaustive compatibility detector. Arrays are compared as whole ordered values; reference expansion and semantic review are manual tasks for your agent. Unsupported/unavailable references remain explicit unknowns. Exit0 or no structural differences never approves compatibility. YAML, Swagger2, OpenAPI3.1/3.2, automatic conversion, CI integration, live API/client/runtime/security tests and deployment are outside this release. Mature free tools such as oasdiff can be preferable for automated semantic checks; you may supply their results as supporting evidence.

Data: local inventory has no network, seller backend, API-key requirement or telemetry. Your chosen AI provider may process the specifications under its own policy; full offline processing is not guaranteed. Remove credentials, real customer payloads and private data. Generated inventories/reviews can contain contract fragments, so handle them as confidential documents.

Commercial work and delivery of generated reports to clients are allowed under the included license. Redistribution or resale of the skill/source/package is excluded. Support: [email protected]; send version and sanitized error/input, never credentials. No SLA, consulting, exhaustive detection, security certification or production-safety guarantee is included.