Plugin directory / Developer / dsh-galgame-localization-reverse
dsh-galgame-localization-reverse
Verified · install-tested on dsh julescules
What it does
DeepSeek Harness Galgame Doctor — engine evidence, strict translation QA, and reversible localization workflows.
Works — verified, early-stage project
DeepSeek Harness Galgame Doctor — engine evidence, strict translation QA, and reversible localization workflows. It installs cleanly and boots without issues in our testing. It's early-stage but functional.
“Verified” means our automated CI actually ran dsh plugin add in a clean profile and it booted — nothing more. Feature descriptions and version compatibility are the author’s claims. This is not a security audit and not an endorsement of third-party code.
README
Galgame Doctor for DeepSeek Harness
中文 | English
[!IMPORTANT]
Unofficial community plugin. Independently developed and maintained; not reviewed or endorsed by DeepSeek.
Audit first, reuse safely, localize second. Give DeepSeek Harness a visual-novel folder and get an evidence-backed engine candidate, safe translation migration, encoding clues, strict QA, and a reversible next step. The audit is offline and read-only; only report paths outside the audited tree are written.
Start in 30 seconds
dsh plugin --profile web add github:julescules/dsh-galgame-localization-reverse#v0.6.0
Restart DSH, then ask:
Use galgame-localization-reverse to audit D:\Games\Example read-only.
Show the engine evidence, localization risks, and the safest next step.
No API key is required. The plugin contains no game assets, decrypted archives, credentials, or title-specific keys.
What you get
One-command localization plan
vn_project_audit.py plan inventories a bounded set of files and reports:
- engine candidates with confidence, signal groups, and concrete path evidence;
- script, archive, image, font, and executable candidates;
- bounded encoding samples and CJK-path-safe relative paths;
- a cautious classification of
candidate,weak-candidate, orunknown; - an evidence status, recommended route, exact next commands, risks, and release gates.
The classification is routing evidence, not a claim that static inspection proves runtime behavior.
python -X utf8 skill\scripts\vn_project_audit.py plan D:\Games\Example `
--json D:\project\logs\project-audit.json `
--markdown D:\project\logs\project-audit.md
Lossless source-script adapters
vn_script_adapter.py extracts plain KiriKiri .ks, Ren'Py .rpy, and NScripter 0.txt text into UTF-8 JSONL while preserving source hash, line hash, path, line number, encoding, BOM, newline, prefix, and suffix. Apply always writes to a separate output root and fails if the source changed.
python -X utf8 skill\scripts\vn_script_adapter.py extract D:\Games\Example `
--engine kirikiri --output D:\project\translations.jsonl
python -X utf8 skill\scripts\vn_script_adapter.py apply D:\Games\Example `
D:\project\translations.jsonl --output-root D:\project\rebuilt
It does not open XP3/NSA archives or claim compiled-script support.
Safe translation migration after game updates
vn_translation_memory.py carries reviewed translations into a newly extracted catalog without silently guessing. It auto-reuses only exact ID/source pairs or relocated exact source matches with one consistent, placeholder-safe target. Fuzzy matches are review-only; conflicts stay empty and make the command return a non-zero status. Both inputs remain unchanged and their SHA-256 hashes are recorded in the report.
python -X utf8 skill\scripts\vn_translation_memory.py migrate `
--current D:\project\v2\translations.jsonl `
--previous D:\project\v1\translations.jsonl `
--output D:\project\v2\migrated.jsonl `
--report D:\project\logs\translation-memory.json `
--markdown D:\project\logs\translation-memory.md
Strict translation preflight
vn_qa.py rejects missing, added, or reordered control tokens by default. v3 adds glossary and forbidden-term gates, repeated-source consistency, character/line budgets, punctuation balance, empty translations, residual Japanese, and target-encoding checks. Generic JSON/JSONL, script-adapter JSONL, and GalTransl cache fields are supported without third-party packages.
python -X utf8 skill\scripts\vn_qa.py check-jsonl D:\project\translations.jsonl `
--encoding gbk --require-complete --require-consistent `
--glossary D:\project\glossary.json --max-chars 80 --max-lines 3 --check-punctuation `
--report D:\project\logs\translation-qa.json `
--markdown D:\project\logs\translation-qa.md
An intentionally broken synthetic fixture produces a concise failure report:
FAIL 2 blockers · 3 segments
E control-token-added scene01:2 added [wait]
E empty-translation scene01:3 translation required
Release evidence, not another translator
The bundled workflow also provides:
vn_patch.py— hash-bound file replacement, verification, backup, and rollback;vn_image_qa.py— PNG/APNG, JPEG, GIF, and BMP geometry/alpha/frame comparison;vn_slot_qa.py— fixed-slot terminator, padding, marker, and outside-region checks;- focused routes for AI6WIN, BGI/Ethornell, Director, RealLive, Delphi/DirectDraw, PSP/Vita, Ren'Py, KiriKiri, and custom engines.
GalTransl, VNTextPatch, GARbro, LunaTranslator, and engine-specific tools can handle translation or extraction. Galgame Doctor complements them by deciding what the project appears to be, checking their text output, and preserving release evidence.
Try the synthetic examples
python -X utf8 skill\scripts\vn_project_audit.py plan examples\synthetic-project\game `
--json test_outputs\synthetic-audit.json `
--markdown test_outputs\synthetic-audit.md
python -X utf8 skill\scripts\vn_qa.py check-jsonl examples\translations.jsonl `
--encoding gbk --require-complete --glossary examples\glossary.json `
--report test_outputs\translation-qa.json `
--markdown test_outputs\translation-qa.md
All example text is synthetic and redistributable.
DSH integration
The bundle uses the official packaged Skill Provider shape:
inject = ['skills'];ctx.skills.registerProvider(...);- directory
resourceBasefor on-demand references and scripts; - the official
dsh.bundle.patchinstall path.
The registered name remains galgame-localization-reverse, so existing prompts continue to work.
Verify the package
npm run check
npm pack --dry-run
node scripts/build-release-metadata.mjs .\dsh-galgame-localization-reverse-0.6.0.tgz .\builds\v0.6.0
.\scripts\verify-release.ps1 -PackagePath .\dsh-galgame-localization-reverse-0.6.0.tgz -ChecksumsPath .\builds\v0.6.0\SHA256SUMS
The check runs provider tests plus deterministic self-tests for planning, script adapters, translation-memory migration, translation QA, and all bundled patch/asset utilities. Each release also publishes SHA-256 and CycloneDX SBOM evidence.
Limits
- Engine formats and binary constants remain title/version specific until proven against exact hashes.
- Script adapters support plain source files only and never imply archive/compiled-script compatibility.
- Static audit and QA do not replace a zero-mutation round trip or in-game UI, save/load, audio, and rollback tests.
- Residual Japanese can be intentional; review each finding or maintain an explicit exclusion instead of silently weakening the gate.
- DeepSeek Harness is in developer preview; pin a reviewed plugin release.
See CONTRIBUTING.md and SECURITY.md for contribution and reporting guidance.
License
Install
Install the catalog once, then DeepSeek Harness can find and install any plugin from this site automatically:
dsh plugin add dshbase-catalog Then say "install dsh-galgame-localization-reverse for me" — your agent finds it in the directory and installs it. Docs: dshbase-catalog · verified packs.
This plugin is GitHub source (not published to npm) — install it straight from the repo:
Web profile:
dsh plugin --profile web add github:julescules/dsh-galgame-localization-reverse Headless (CLI) profile:
dsh plugin --profile headless add github:julescules/dsh-galgame-localization-reverse Test report
Verified: L1 install + L2 load + L3 runtime from GitHub source on dsh 0.1.0-rc.6.