ZenlessZoneZero-OneDragon: a Windows automation stack for Zenless Zone Zero

What OneDragon automates inside Zenless Zone Zero

The README lists five capability groups: automated combat with custom logic, skill markers, conditions and variables; a dodge assistant; daily cleanup covering the video store, scratch cards, coffee shop, materials and rewards; Hollow Zero operations with recognition, event and pathfinding modules; and scheduling features such as timed launch, multi-account switching, scheduled tasks and automatic dialogue. The audience is a player who already runs Zenless Zone Zero on Windows and wants the repetitive layer removed without writing their own vision pipeline. The project describes itself as a study in image recognition and automation rather than a game modification, and the disclaimer states it is for learning and exchange only, that the developers offer no after-sales service, and that any risk from using the tool is borne by the user. That framing matters: this is a desktop automation tool that drives the game through the same surfaces a human uses, not a client patch.

Recognition instead of memory reading: the architecture

The mechanism visible in the repository is inference over what the screen and the sound card produce. The dodge assistant is described as a combined sound and image model, which is why soundcard and librosa sit in the dependency list next to opencv-python. Inference runs through onnxruntime-directml, so the models execute on the local machine and can use DirectML for acceleration. pyautogui and pynput provide the input side, sending the key and mouse events the game reacts to, while vgamepad in the optional gamepad group covers controller output. Shapely and pyclipper are geometry libraries, consistent with the pathfinding and region work implied by the Hollow Zero module. The application shell is PySide6 with pyside6-fluent-widgets, which matches the screenshot in the README showing an operation interface. pygit2 is present, which suggests the app can pull its own updates from a Git remote. Nothing in the repository claims the tool reads game memory or injects into the process, and the recognition-first design is the reason the dependency set looks like a computer vision project rather than a game mod.

Installing ZenlessZoneZero-OneDragon on Windows

The README does not contain install instructions. It points to the official website at one-dragon.com, which it describes as covering the project introduction, beginner onboarding, feature documentation and configuration documentation, and to an official document link described as covering common problems, fault lookup, diagnostics and update guidance. Treat those two pages as the install path; the repository itself gives you the environment contract rather than a step-by-step. That contract is strict: requires-python is >=3.11.9,<3.12, so 3.11 is the only supported interpreter line, and pyproject.toml sets environments to sys_platform == 'win32'. A manual environment setup therefore starts from a Python 3.11 interpreter and a uv sync against the repository root, which installs the default dependency group only. The dev and gamepad groups are opt-in, so controller support is not part of a default install. After the sync, the README's own channels are where you look for the first launch procedure, since no run command appears in the repository files listed here.

The pywin32 pin and the redistributable trap

One dependency line carries a comment that explains a real failure mode. pywin32 is pinned below 312 because wheels from 312 onward no longer bundle mfc140u.dll, and on machines without the VC++ 2015-2022 Redistributable (x64) installed, importing win32ui crashes. The comment cites issue #2428 and says the pin will be lifted once an upstream build 313 fixes it. This is the kind of detail that decides whether a first run succeeds, and it is documented only in a TOML comment, not in the README. If you build the environment yourself rather than using a packaged release, install the redistributable or accept the pin. The same file also sets a uv cache directory under .install/uv_cache, so a uv-based setup keeps its cache inside the repository tree, and default-groups is empty, meaning the dev and gamepad groups are opt-in. Controller support is not part of the default install.

Where OneDragon is the wrong tool

Three boundaries stand out. First, platform: the README badge says Windows, and pyproject.toml restricts environments to win32, so macOS and Linux users have no supported path in this repository. Second, form factor: this is a PySide6 desktop application with a GUI, not a headless service, so anyone wanting to run it on a home server or inside a container has nothing to work with here. Third, the automation model itself is a limitation. Because the tool recognises pixels and audio rather than reading game state, it inherits the fragility of that approach: a UI change, a resolution change or a scene the models have not seen can break a task, and the README's own support channels are organised around fault lookup and diagnostics rather than around guarantees. The disclaimer reinforces the point that no support obligation exists. If you need deterministic, contract-based automation against a documented API, this is not that project, and no amount of configuration will make it one.

OneDragon compared with a general-purpose macro tool

The obvious alternative is a general desktop automation or macro recorder such as AutoHotkey-style scripting or a generic vision-based automation framework. The difference is in what ships in the box. A generic tool gives you primitives: find an image on screen, click a coordinate, wait for a pixel. You then build the Zenless Zone Zero logic yourself, including the dodge timing, the Hollow Zero pathfinding and the daily routine ordering. OneDragon inverts that: the game-specific logic is the product, and the primitives are internal. The trade-off is control. With a general tool you can fix a broken step in an afternoon because you wrote it; with OneDragon you depend on the maintainers recognising the same breakage and shipping a release. The release history supports the idea that this happens on a regular cadence, with v2.5.2 on 2026-09-18 following v2.5.1 on 2026-08-12 and v2.4.7 on 2026-08-02, and the last push to the repository was on 2026-09-21. Choose OneDragon for coverage, choose a generic tool for control.

Licence, signature and maintenance cost

The repository is GPL-3.0. If you redistribute the software or a modified version, the licence's source and copyleft conditions apply; if you only run it locally, the practical effect is small. This is not legal advice, and the project's own disclaimer adds a separate layer: it states the project authorises no individual, merchant or media account to sell it, and that any problems arising from paid boosting services using the software are unrelated to the project. On integrity, the README documents a code signing policy: free code signing is provided by SignPath.io with a certificate from SignPath Foundation, and the named approvers are DoctorReid and ShadowLemoon. That is a checkable signal when you download a release. Maintenance cost is mostly on the maintainers' side, since the recognition models and game-specific logic need to track the game. For a user, the cost is update latency: when a game patch breaks a task, you wait for a release rather than patching it yourself. The last push was on 2026-09-21, so the repository is not dormant, but nothing here promises a response time.

Editorial conclusion

Adopt it if you play Zenless Zone Zero on Windows, want dailies, Hollow Zero and combat handled by a GUI tool, and are comfortable with a project whose README points you to an external site for setup. Skip it if you are on macOS or Linux, since pyproject.toml pins the dependency set to win32 and the README advertises Windows only, or if you need a headless server daemon rather than a desktop application. Before trusting it with an account, read PRIVACY.md, confirm the release you download carries the SignPath signature the README describes, and check the GPL-3.0 obligations for anything you redistribute.

Frequently asked questions

Official sources

1. License: GPL-3.0
2. OneDragon-Anything/ZenlessZoneZero-OneDragon on GitHub
3. Project website
4. README
5. Releases