purewhiter--mobilegym
wehub-resource-sync
55 行
4.1 KiB
Markdown
55 行
4.1 KiB
Markdown
# MobileGym Documentation
|
|
|
|
Welcome. This index points you at the right doc for what you're trying to do. The repository [`README.md`](../README.md) is the project overview — these docs go deeper.
|
|
|
|
All docs here are in English. Pick a track:
|
|
|
|
## Where to start
|
|
|
|
| If you want to… | Read this |
|
|
|---|---|
|
|
| 🚀 Install and run your first task | [`getting-started.md`](getting-started.md) |
|
|
| 🏗️ Understand how the layers fit together | [`platform/architecture.md`](platform/architecture.md) |
|
|
| 📱 Add a new app to the simulator | [`guides/add-an-app.md`](guides/add-an-app.md) |
|
|
| 🧪 Add a new task and judge | [`guides/add-a-task.md`](guides/add-a-task.md) |
|
|
| 🤖 Plug in a new agent (model adapter) | [`guides/add-an-agent.md`](guides/add-an-agent.md) |
|
|
| 📊 Benchmark an agent on MobileGym-Bench | [`guides/bench-an-agent.md`](guides/bench-an-agent.md) |
|
|
| 🔌 Look up the browser-console debug API | [`api/runtime-api.md`](api/runtime-api.md) |
|
|
| 🤝 Contribute code or report a bug | [`../CONTRIBUTING.md`](../CONTRIBUTING.md) |
|
|
|
|
## Platform deep dive
|
|
|
|
For implementation-level rules and contracts — when you're extending the simulator itself, debugging unusual behavior, or building a new app whose needs aren't covered by the tutorial. See [`platform/README.md`](platform/README.md) for the orientation.
|
|
|
|
| Topic | Spec |
|
|
|---|---|
|
|
| OS internals — TaskManager, BackDispatcher, IntentResolver, lifecycle | [`platform/os/overview.md`](platform/os/overview.md) |
|
|
| App module contract — how apps integrate, file conventions, runtime hooks | [`platform/app/module-contract.md`](platform/app/module-contract.md) |
|
|
| State model — layered runtime/world data, snapshots, diffs, side effects | [`platform/state/model.md`](platform/state/model.md) |
|
|
| Declarative navigation — FSM, transitions, actions, graph generation | [`platform/navigation/declaration.md`](platform/navigation/declaration.md) |
|
|
| Intent system — cross-app calls, launchMode, choosers | [`platform/os/intent-system.md`](platform/os/intent-system.md) |
|
|
| Cross-app launch — Task/back-stack traces, `openApp`, `newTask` | [`platform/os/cross-app-launch.md`](platform/os/cross-app-launch.md) |
|
|
| OS services — Time, Location, Network, SMS, Display, Clipboard, etc. | [`platform/os/services/README.md`](platform/os/services/README.md) |
|
|
| Display scaling — system zoom, font size, per-app design width | [`platform/os/services/display-scaling.md`](platform/os/services/display-scaling.md) |
|
|
| Android mapping — which AOSP concepts we model, which we simplify | [`platform/android-mapping.md`](platform/android-mapping.md) |
|
|
|
|
## Benchmark / task authoring
|
|
|
|
Pair `bench_env/` and its docs when authoring tasks:
|
|
|
|
| Doc | Read for |
|
|
|---|---|
|
|
| [`../bench_env/docs/task/TASK_AUTHORING_GUIDE.md`](../bench_env/docs/task/TASK_AUTHORING_GUIDE.md) | Task authoring workflow — audit app state, add helpers, write the task, tune sampling |
|
|
| [`../bench_env/docs/task/TASK_CODE_SPEC.md`](../bench_env/docs/task/TASK_CODE_SPEC.md) | Hard task-code spec — taxonomy, templates, parameters, judge rules, forbidden patterns, checklist |
|
|
| [`../bench_env/docs/REFERENCE.md`](../bench_env/docs/REFERENCE.md) | Formal field reference — `JudgeInput`/`JudgeResult` shapes, CLI flags, metadata vocab |
|
|
| [`../bench_env/docs/task/GROUNDED_MODE.md`](../bench_env/docs/task/GROUNDED_MODE.md) | The AnswerSheet / grounded-mode protocol |
|
|
| [`../bench_env/docs/task/TASK_TESTING_GUIDE.md`](../bench_env/docs/task/TASK_TESTING_GUIDE.md) | Task testing guide (`OFFLINE_JUDGE_POSITIVE_CASES`, `NEGATIVE_CASES`, fixtures) |
|
|
| [`../bench_env/docs/FRAMEWORK.md`](../bench_env/docs/FRAMEWORK.md) | Runner lifecycle, Episode pipeline, sampling, parallel execution |
|
|
|
|
## Conventions used in these docs
|
|
|
|
- **Code paths and identifiers** are real — copy-paste them into `grep` and you'll find what's referenced.
|
|
- **Tables** are the load-bearing format. We avoid prose for things that are really enumerations.
|
|
- **`(advanced)` or **`(internal)`** tags mark sections most readers don't need.
|
|
- These docs are written against the **current main branch**. If a doc disagrees with the code, file an issue — the code wins.
|