adld-ai/appintoss-kit

appintoss-kit is an unofficial toolkit that gives AI coding agents a structured, repeatable workflow for building mini-apps for App in Toss, a Korean platform for hosting small apps inside the Toss app. Instead of relying on scattered one-off prompts, the kit organizes agent instructions, reusable skills, and verification steps into a consistent loop that any agent can follow. The core idea is a root contract file called AGENTS.md, which every agent reads first. It tells the agent to research the codebase before editing, prefer the smallest correct change, follow the project's instruction rules, route each task through a resolver file to find the right guide, and verify the work before claiming it is finished. Because app-in-toss work touches many different areas such as Korean UX writing, Toss-style visual design, store assets, screenshots, and console metadata, the kit keeps the rules for each area separate so an agent loads only what it needs. The .agents folder holds the reusable execution layer: subagent definitions for review, documentation images, lifecycle hooks that gate each step of development, larger reference material, and individual skills such as app naming, SEO keyword generation, image generation, and the Toss design theme. Every meaningful task follows the same loop: resolve the task, read the required rules, make the smallest scoped change, verify with the strictest relevant check, and if verification fails, replan from the root cause before finishing. The repository only tracks the public coordination layer itself. Generated mini-apps, build output, screenshots, and submission bundles are meant to stay local rather than being committed. The kit explicitly states it must not use official Toss trademarks as its own identity, must not invent unsupported behavior, and must not claim a task is ready without fresh verification evidence. It is released under the MIT license.