dsh-galgame-localization-reverse
中文 | English
> [!IMPORTANT] > Unofficial community plugin. Independently developed and maintained; not reviewed or endorsed by DeepSeek.
A packaged DeepSeek Harness skill for evidence-driven Windows galgame and visual-novel localization.

What it does
- detects 25+ visual-novel engine families from files, signatures, and runtime evidence;
- routes text, image, archive, bytecode, font, and encoding work to focused references;
- preserves placeholders, control codes, source hashes, save compatibility, and rollback paths;
- includes
vn_qa.pyfor placeholder/encoding gates; - includes
vn_patch.pyfor hash-bound file replacement, verification, and rollback; - includes dependency-free
vn_image_qa.pyfor PNG/APNG, JPEG, GIF, and BMP inventory plus geometry, alpha, frame-count, and hash comparison; - covers AI6WIN, BGI/Ethornell, Director, RealLive, PSP/Vita workflows, and custom engines.
Install
dsh plugin --profile web add github:julescules/dsh-galgame-localization-reverse#v0.2.0
dsh --profile web --dump-configRestart DSH after installation. The bundled provider registers galgame-localization-reverse with ctx.skills and exposes its references and scripts through resourceBase.
Use
Ask DSH to use galgame-localization-reverse, then provide the game directory and goal. The workflow starts with read-only inventory and engine evidence before selecting an extraction or writeback route.
Use galgame-localization-reverse to identify this engine, build a text-owner map,
and propose a reversible localization workflow for D:\Games\Example.Built-in utilities
python -X utf8 skill\scripts\vn_qa.py --selftest
python -X utf8 skill\scripts\vn_patch.py --selftest
python -X utf8 skill\scripts\vn_image_qa.py --selftest
# Compare original and localized raster trees.
python -X utf8 skill\scripts\vn_image_qa.py compare D:\project\original\images D:\project\working\images -o D:\project\logs\image-qa.jsonThe repository contains no game assets, decrypted archives, credentials, or title-specific keys. Public releases should contain tooling, manifests, documentation, and differential or replacement patches only.
Verified
- official packaged Skill Provider shape:
inject = ['skills']andctx.skills.registerProvider; - official
dsh.bundle.patchinstall path; - frontmatter removal and resource-base resolution;
- direct-reference integrity across the skill bundle;
- Node provider tests plus all three Python utility self-tests;
- package allowlist and public-content scan.
- real
@deepseek-ai/dsh@0.1.1-rc.2profile installation and config composition with clean host-peer resolution.
npm run check
npm pack --dry-runLimits
- Engine formats and binary constants remain title/version specific until proven against exact hashes.
- Static validation does not replace runtime UI, save/load, audio, and rollback tests.
- DeepSeek Harness is in developer preview; pin a reviewed plugin release.
License
[MIT](LICENSE)