Kinet - API Automation with Draft PRs

Source code: github.com/akhilyad/smasi (product name: Kinet)

Kinet is a prototype SaaS for teams that integrate with payment APIs. It compares the API version a codebase is pinned to with a provider's newer version, reports what is affected, and prepares a draft pull request with the change. The first provider modelled is Stripe, including the Stripe.js method removals in its named major versions.

Payment API upgrades are easy to postpone and hard to scope. Teams do not know which of their files a new version touches, or whether a fix is mechanical or needs a human decision.

A readiness report separates mechanical changes (a fixed rewrite rule) from semantic ones (plan only, no automatic edit). Mechanical changes become a patch that goes through a verification ladder, then a draft pull request whose body states what was and was not verified.

Spec-diff, repository indexer and playbook modules, including a Stripe.js legacy-confirm-methods rule (mechanical) and a charges-create deprecation rule (semantic, plan only)

Patch agent that edits files and derives the diff from the result instead of asking a model to write a diff

Verification ladder (baseline, static, tests, diff audit) with an explicit confidence label and a list of what was not verified

Draft-only pull request publisher: never auto-merges, never pushes to a default branch, PR body includes a change table, verification table and a before-you-merge checklist

Invite-only access request flow, upgrade readiness report, upgrade run history and audit log screens

Next.js App Router with TypeScript, a Prisma schema, and a written build spec with hard rules (no secrets in the sandbox, no customer source stored, repository content treated as untrusted data). The repository's scripted test run passes 31 of 31 checks (run locally), covering playbooks, patch generation, the verification ladder, the draft-PR rules and the readiness report.

Sandbox execution is simulated; no customer code is actually run in an isolated container

The runs and report screens show seeded sample data for a fictional repository

No live GitHub App integration or real provider polling has been verified

A prototype: no certification or production readiness is claimed