wscodec

Pure-JS toolkit for Soulmask save files. The core is a codec for actor_data property streams in Soulmask's world.db (Unreal Engine 4.27 FPropertyTag wire format).

• Zero runtime dependencies in the codec (src/).
• Round-trip byte-identical against every actor in test saves (run npm test -- "path/to/world.db").
• Decode bytes to JS objects, edit, re-encode. Or convert to/from JSON for text-based editing and diffing.
• Optional translation tables map Soulmask's internal class names and IDs to English display names.

The dev dependencies (better-sqlite3, lz4-wasm-nodejs, esbuild) are only used by the scripts and the build, not by the codec.

Install

npm install wscodec

Requires Node 20+.

For browser use, the build emits IIFE, ESM, and CJS bundles in dist/. CDN loaders can pull wscodec.global.js (window globals) or wscodec.mjs (<script type="module">).

How to use

UnrealBlob serves as your entry point.

import { UnrealBlob } from 'wscodec';

// Decode the uncompressed property stream bytes.
const blob = UnrealBlob.fromBytes(uncompressedBytes);

// Walk top-level properties, or search anywhere in the tree.
const nameProp = blob.findProperty('JianZhuDisplayName');
const deepHit  = blob.findPropertyDeep('RongQiCunQuRiZhiData');

// Edit in place. Strings, numbers, ObjectRefs, arrays, maps all map to plain
// JS values you can mutate directly.
nameProp.value.displayString = "Auroris' Cabinet";

// Re-encode. Tag sizes are recomputed from the actual encoded value bytes,
// so any structural edit is safe.
const newBytes = blob.toBytes();

fromBytes accepts a second argument: pass { strict: true } to escalate opaque-fallback warnings (unknown property type, unfamiliar FText history, delegate families, etc.) into thrown errors instead of the default console.warn + capture-as-raw-bytes behavior.

const blob = UnrealBlob.fromBytes(uncompressedBytes, { strict: true });

JSON is supported as an interchange format. Sentinel strings are used to preserve -0, Infinity, and NaN.

const json = blob.toJSONString(2);                 // 2-space indent
const back = UnrealBlob.fromJSONString(json);      // reconstruct
back.toBytes();                                    // re-encodes byte-identical

The argument to toJSONString is forwarded to JSON.stringify as its space parameter: omit it (or pass 0) for compact single-line output, pass 1–10 for that many spaces per level, or pass a string like "\t" to indent with that exact string.

From a browser without a bundler

dist/wscodec.global.js is an IIFE bundle that exposes everything on window.wscodec. Load it with a classic <script> tag — unpkg and jsDelivr serve it directly from the npm package:

<script src="https://unpkg.com/wscodec"></script>
<script>
  const blob = wscodec.UnrealBlob.fromBytes(uncompressedBytes);
  // window.wscodec re-exports the same surface as the ESM build.
</script>

Translations ship as a separate per-language bundle on window.wscodecTranslations (pick the language file you need - en, de, es, fr, ja, ko, pt, ru, zh):

<script src="https://unpkg.com/wscodec/dist/wscodec-translations.en.global.js"></script>
<script>
  wscodecTranslations.item('/Game/Blueprints/.../BP_WuQi_Dao_2_C');
</script>

For <script type="module">, point at dist/wscodec.mjs and use the same import { UnrealBlob } from '...' syntax shown above.

The actor_data envelope

UnrealBlob.fromBytes accepts the inner property stream only. Soulmask's actor_table.actor_data column wraps that in an LZ4 envelope:

[0..3]  u32 LE   outer version tag = 0x00000002
[4..]   LZ4 block   decompresses to the bytes fromBytes accepts

When loading rows from world.db, use your favorite LZ4 decompresser (I like to use lz4-wasm-nodejs):

import { UnrealBlob } from 'wscodec';
import * as lz4 from 'lz4-wasm-nodejs';

const inner = lz4.decompress(actorDataColumn.subarray(4));
const blob = UnrealBlob.fromBytes(inner);
// ...
const re = blob.toBytes();
const out = Buffer.concat([Buffer.from([2, 0, 0, 0]), Buffer.from(lz4.compress(re))]);

LZ4 has multiple valid encodings for the same payload, so the column bytes themselves are not byte-stable across an edit-free round-trip. The decompressed inner bytes are the canonical form and the level at which round-trip tests check equality.

Translations

Optional name lookups for the IDs and class names Soulmask uses. Import the language you want - en, de, es, fr, ja, ko, pt, ru, or zh; each subpath exposes the same API:

import { translate, item, npc, recipe } from 'wscodec/translations/en';

item('/Game/Blueprints/DaoJu/.../BP_WuQi_Dao_2.BP_WuQi_Dao_2_C'); // 'Beast Bone Blade'
npc('BP_DongWu_Base_C');                                          // NPC display name
recipe('WuQi_Dao_2');                                             // recipe name by id

// Generic lookup, scans every table until something matches.
translate('BP_WuQi_Dao_2_C');                                     // 'Beast Bone Blade'
translate(100011, 'gifts');                                       // disambiguate by table

There is no default language - the subpath is the language selector. To switch language at runtime, dynamically import the one you need (await import('wscodec/translations/' + lang)). To bind your own table set, createTranslations(tables) is available from the locale-neutral wscodec/translations.

Tables: items, npcs, buildings, recipes, proficiencies, mastery, attributes, fashion, tattoos, gifts, settings, categories. Display names only, no descriptions, icons, or stats.

Each language's tables are generated from the game's data export and committed as src/translations.data.<lang>.mjs. Regenerate after a game patch with npm run build-translations (see "Regenerating translations" below).

Scripts

Utilities and tests. All take a world.db path as the first argument. The ones that write files default the output path next to the input.

Script	What it does
node scripts/db-to-json.mjs <world.db> [out.json] [--pretty]	Export the entire SQLite save to a single JSON file. actor_data rows are decoded through wscodec; everything else (INTEGER, REAL, TEXT, BLOB, NULL) round-trips verbatim. Stream-writes so multi-hundred-MB saves don't blow Node's heap.
node --max-old-space-size=4096 scripts/json-to-db.mjs <in.json> [out.db]	Inverse of the above. Rebuilds a SQLite db from the JSON. The actor_data column is byte-identical at the uncompressed level (LZ4 itself may re-encode).
node scripts/diff-dbs.mjs <a.db> <b.db>	Confirm two databases round-trip equivalent. Compares schema verbatim and every cell, with actor_data compared at the uncompressed level.
node scripts/dump-actor.mjs <world.db> <actor_serial> [out.json]	Decode one actor and print its property tree as JSON. Handy for inspecting a single row.
node scripts/dump-logs.mjs <world.db> [out.log]	Merge workbench/chest access logs, NPC work logs, and clan logs into one timestamp-sorted text file shaped like the in-game log panel.
node scripts/find-string.mjs <world.db> <needle> [--limit N]	Substring search across every decoded blob (StrProperty, NameProperty, FText, etc.). Reports actor serial and property path for each hit. Useful for locating an in-game object you can recognize by a string.
node scripts/npc-log-survey.mjs <world.db> [--tz-offset -6]	Find the NPC with the most diverse set of work-log Type codes and print the most-recent entry for each. Used to cross-reference Type codes against the in-game text.
node scripts/build.mjs (npm run build)	Build the dist/ bundles via esbuild. Runs automatically on npm publish.
node scripts/build-translations.mjs (npm run build-translations)	Regenerate the per-language src/translations.data.<lang>.mjs tables from the CSV exports in ext/<lang>/ (see below).

The test-edit-*.mjs scripts in scripts/ are edit recipes against specific in-game objects (a named chest, a specific NPC's log, a wall). Useful as templates for your own edits; see the file headers for the expected before/after.

Regenerating translations

The translation tables ship pre-generated as src/translations.data.<lang>.mjs, one per language. To rebuild from a fresh game patch:

1. Generate the CSV export with SoulmaskDataMiner.
2. Drop each language export into ext/<lang>/ at the repo root (gitignored), e.g. ext/en/, ext/de/. Languages are auto-discovered from these folders.
3. Run npm run build-translations.

Adding a language also needs a one-line src/translations.<lang>.mjs wrapper and a matching ./translations/<lang> entry in package.json exports; the build prints a reminder if a language has data but no wrapper.

Tests

All test scripts take an optional path to a world.db to test against. Without an argument they look for ../world.db relative to the repo root (so a sibling save folder works without a flag).

npm test                    # runs every test:* script in package.json
npm run test:peek-fuzz      # adversarial fuzz for the FGuid-vs-tag peek heuristic
npm run test:roundtrip      # decode + re-encode every row, byte-identical check
npm run test:json-spot      # JSON round-trip on representative rows
npm run test:json-full      # JSON round-trip on every row
npm run test:dump-logs      # exercise the log dumper end-to-end

# Pass a save explicitly (note the `--` so npm forwards the arg):
npm test -- /path/to/world.db
node test/test-roundtrip.mjs /path/to/world.db

test:roundtrip is the primary regression tripwire: it decodes every actor_data row, re-encodes via toBytes(), and reports the first byte of divergence (with surrounding context) for any row that doesn't match. test:json-full does the same end-to-end through the JSON layer. Decode failures are bucketed by error pattern so identical bugs can be identified.

Exit code is non-zero if any decode failure, encode failure, or round-trip mismatch is found.

License

MIT. See LICENSE.