Introduction

The core of DeepSeek Harness is “Everything is a plugin.” For game developers or localization teams, building a pipeline from discovery, terminology, translation to packaging and validation often requires writing scripts and maintaining configuration. dsh-plugin-tav2 integrates this pipeline into the conversational interface. Translation is performed by the model; users only need to issue instructions through conversation. From initialization and translation to packaging and validation, the entire process is driven by standard workflows.

What Is This

This is a plugin for DeepSeek Harness, and the open-source repository is on GitHub (Drhushi/dsh-plugin-tav2). Its core value is connecting the AI translation pipeline to conversation: translation is performed by the model; you only need to talk to the assistant. From initialization and translation to packaging and deployment, the entire process is driven by standard workflows. This plugin uses the MIT license.

Core Features

  • Conversational end-to-end workflow: Say “Initialize game translation” and the assistant detects the game, generates the configuration, and provides a translation plan; it starts only after confirmation.
  • Patch-based, non-invasive operation: It leaves the original game files untouched. The deliverable is a mergeable patch package (including the manifest list and version fingerprints), and deleting the files listed in the manifest fully restores the original state.
  • Incremental migration: Incremental migration is performed by version fingerprint and stable ID. Existing translations are automatically retained, and only changed parts are retranslated.
  • Translation quality infrastructure: Terminology scanning and deliberation-based locking, a worldbook nomination system (recommends setting-level entities worth creating cards for based on three-question criteria), review queue, integrity check, and three-layer runtime validation.
  • Multi-engine adapter architecture: A single workflow serves different game engines; the Ren’Py adapter is currently implemented.

Install and Enable

Before installing, confirm that the environment meets the following requirements:
* DeepSeek Harness ≥ 0.1.5-rc.1 (Web edition, Node.js ≥ 22.19).
* The game directory contains game/ (currently it must be a Ren’Py game; see below for other engine support).
* A usable LLM translation channel is configured (for example, a DeepSeek API key or a local OpenAI-compatible endpoint).

Installation Method A (recommended):

Download the prebuilt .tgz asset from the GitHub release; no local build is required. Run the following in PowerShell:

$ver = '0.2.1'   # 换成你要装的版本
$tgz = "https://github.com/Drhushi/dsh-plugin-tav2/releases/download/v$ver/dsh-plugin-tav2-$ver.tgz"
pnpm -C "$env:USERPROFILE\.dsh\profiles\web" add $tgz

After installation, restart the GUI (the client bundle is a snapshot taken at startup; reopening the chat does not take effect). Go to “Settings → Plugins” and confirm that the card appears.

Typical Usage

  1. Initialization: After setting the workspace folder to the game root directory, tell the assistant “Initialize game translation,” and it will generate a minimal configuration.
  2. Reconnaissance: Run tav2_detect or tav2_status to list the translation plan for confirmation.
  3. Translation: Run tav2_translate_batch. The assistant translates in batches by scene and automatically validates the result after completion.
  4. Packaging and deployment: Run tav2_pack to generate the patch package and merge the game directory into the root directory.
  5. Verification: Select Chinese in the in-game language menu, launch the game and check, or run tav2_verify for rechecking.
  6. Uninstall the patch: Run tav2_uninstall to precisely delete the files listed according to the manifest.

Commonly used tools also include:
* tav2_prepare: Generates the translation template.
* tav2_terms / tav2_deliberate: Terminology scanning and locking.
* tav2_worldbook: Worldbook nomination system.
* tav2_check: Integrity check.
* tav2_migrate: Incremental migration after a game update.

Applicable Scenarios and Notes

The plugin initially supports Ren’Py games. Standard .rpy source games work out of the box; if you encounter compiled .rpyc games, you need to configure the Ren’Py SDK. Other engines such as Unity are currently recognized by layout, but the adapter is still pending implementation; false positives are flagged.

Please note the following:
* Prerequisites: Runtime components such as CJK fonts are prerequisites and must be installed by the user; they are not included in the patch package.
* Version dependency: DSH must be ≥ 0.1.5-rc.1 to load properly.
* Uninstall mechanism: Uninstallation requires removing the mount from the profile and restarting the harness. The “Translation Mode” switch on the settings card is not a loading gate; turning it off does not stop the plugin from taking effect in new sessions.

Short Ending

dsh-plugin-tav2 standardizes the previously cumbersome localization workflow through conversational interaction. If you are using DeepSeek Harness for game localization, especially Ren’Py projects, this plugin provides a one-stop solution from discovery to delivery. For project details, see GitHub.