LostSouls.world
Team brief · For @Quik from @Marley · 9 Oct 2026

Your scripts.
One button each.
Every district faster.

How the Lost Souls build console went from "make us some tool cards" to an actual plan: v1, a launcher for your existing Unreal tools, and v2, the same launcher plus a District Kit so Brickell becomes the first district built this way, not a one-off.

Status Phase 1 built · in review Builder Codex · Review Claude + you Your scripts unchanged Game script NOT finalized
01 · The thread

How we got here

The short version of one long night: we kept building tools that described the work, until we went back to the repo and looked at where the work actually happens. Your branch.

01

Tool cards

Lost Souls Game Toolkit PDF · 25 cards

A branded deck explaining engines, 3D tools, audio, publishing and console dev programs. It looked good, but it was written as if we were starting from zero. It pitched Unity, Godot and Ren'Py, left out Tripo and Midjourney, treated Steam as first and PlayStation as "later," and described a detective game instead of an island-building one.

02

"Can we build a console?"

Control Room → Game Console V3

The idea became a creation console. Codex cloned Game Console V2 into V3 with Create & Build, Asset Needs and Tools & Trials. It's a good planning page, but it's a web page: it can't touch Unreal, Blender or any files, and notes vanish when you reload.

03

The honest take

The turning point

We asked for a console that builds the game, and we'd drifted into four layers of tools about the game. Meanwhile the real building was happening on your branch: the island-wide grid, lit windows on 47 buildings, substations, repair jobs, multiplayer tests and packaged builds.

Your AI is already your console. The scripts are the raw parts. They just don't have one place to run from.

04

Reading your scripts

Every file under tools/ and scripts/ on unreal-island-archipelago

Around a dozen of your tools could become buttons today. One correction came out of reading them: most of your tools close the editor and launch a separate Unreal process. A button panel inside Unreal would kill the editor you clicked it in. So the console has to be a small app that drives Unreal from outside.

05

v1: the Launcher

Plan written for Codex

Buttons on your existing scripts, with pass/fail results, screenshots and run history, then Tripo automation, then a V3 hookup. Details below.

06

"Can we make it scalable?"

Why v2 exists

The question: does it work like a standard you can deploy over and over, where every project makes the next one faster? v1 didn't. It was a nice launcher still locked to Brickell. 88 of the scripts are wired to Brickell, and there's no written definition of a finished asset.

07

v2: Launcher + District Kit

Current plan

The same launcher, built on shared standards: district manifests, an asset spec checked by a script, asset status tracking, batch production with a cost cap, and a second-district pilot to prove it's faster. Details below.

08

Codex builds Phase 1

Same day · two PRs into your branch

Codex built the v2 Phase 1 launcher and opened it for your review. Nothing is merged and nothing touched main. Where it stands.

02 · v1

The Launcher: buttons on what you built

A small app on the dev PC (localhost only) that runs your existing tools, shows the live log, reports a clear PASS / FAIL with your counts ("18 of 18"), shows the screenshots, and keeps a history of every run. It doesn't change your scripts or add game code.

Reference→ Tripo→ Intake→ Preview→ Swap→ Test→ Package
Assets
Clean up a Tripo modelintake_glb.py
Preview contact sheetinspect_asset.py
Swap into map · closes Unrealbrk_swap_*.py
Full Brickell rebuild · closes Unrealbrk_rebuild.ps1
World
Rebuild grid databuild_island_grid.py
Apply grid to map · closes Unrealbrk_grid_island.py
Night option on shots and testsgrid.Night 1
Shots
Brickell camera presetsbrk_shot.ps1
Island aerials, straight to The Wireisland_shot.ps1
Tests
Grid · Repair job · Night*_pie_test.py
Multiplayer suiterun_pie_multiplayer.ps1
Package + full testpackage_and_test.ps1
v1 phases
  • 1. Launcher: the buttons above.
  • 2. Tripo, one button: reference PNG → Tripo API → intake → preview → approve.
  • 3. V3 hookup: Asset Needs becomes a job queue (design only).
Where v1 fell short
  • Every button is still Brickell-only.
  • Adding a button means writing a custom parser.
  • No definition of a "finished" asset, and no status tracking.
  • Only runs on your PC.
  • Nothing proves the second district would be faster.
03 · v2

Launcher + District Kit: build it once, deploy it everywhere

The idea: standardize the design, make the process repeatable, and make every deployment teach the next one. For us, "the standard building" is a district, and the kit is what lets district three take a fraction of the time Brickell did.

88scripts in tools/ and scripts/ that reference Brickell today
34Brickell assets in the manifest, with no status field
32reference images waiting in the Tripo queue (T01–V04)
0written rules for what a "game-ready" asset is
What v2 addsWhat it means for you
District manifestsEach district is one data file: map, data paths, asset folders, camera presets, test set. Brickell's points at your files where they already are, so nothing moves.
Standard asset specOne rulebook per category: scale, pivot, triangle budget, materials, textures, collision, provenance. A validator checks it automatically. You set the budgets, from measurements of your 34 assets.
Asset lifecycleEvery asset shows where it stands. "Passes the checks" and "a human approved the look" are separate steps.
One result formatNew scripts print one LSRESULT line. Adding a button = adding a small JSON file. Your existing scripts get a legacy parser once.
Batch productionTripo batches with estimated credits shown before you submit and a monthly cap. Blender jobs run in parallel; Unreal jobs stay one at a time.
District test setOne button runs the whole standard set. Every bug you fix becomes a test that runs on every future district.
Any machinePaths move into a config file, with optional env overrides in your runners that default to today's values. Runs on Legend's PC too.
Second-district pilotThe real test: stand up a small district from the kit and time it against Brickell. Every Brickell-only assumption we find becomes the checklist for district three.
requestedreferencegeneratedcleanedvalidatedapproved · humanin maptested

Asset lifecycle. Everything is automatic except the gold step.

04 · Side by side

v1 vs v2

v1 · Launcherv2 · Launcher + District Kit
ScopeBrickell onlyAny district, via manifests (Brickell first)
Adding a buttonNew code + custom parserOne JSON file + standard result line
"Done" meansSomeone looked at itPasses the spec, then human approval
Asset trackingNot tracked8-step lifecycle, shown in launcher and later V3
TripoOne asset at a timeBatches, cost estimate, monthly cap
Running jobsOne queueParallel Blender, one-at-a-time Unreal
TestsIndividual buttonsDistrict test set in one click; fixes become regressions
MachinesYour PCAny team machine via config
Proof it scalesNoneSecond-district pilot, timed against Brickell
MetricsNoneMinutes per asset, first-pass rate, credits per asset, district bootstrap time
Your scriptsUnchangedUnchanged; any generalization is a small opt-in PR for you
05 · Where it stands · 9 Oct, afternoon

Phase 1 is built. Now it needs your machine.

Codex built the launcher on tools/launcher (PR #3) and the optional path overrides on tools/launcher-portability (PR #4), both into unreal-island-archipelago. Claude reviewed the code and re-ran the tests.

19buttons, each a JSON job file. No launcher code needed to add one.
93automated tests pass. Claude re-ran them: 93 pass, 5 Windows-only tests skipped on Linux.
1real job verified so far: grid data (165 nodes, 110 poles, 47 buildings), single and in a batch
18buttons not yet run with real Unreal or Blender. That run happens on your workstation.
AreaStatus
District manifestBrickell manifest and district test set in place, pointing at your existing files. Nothing moved.
RunnerUnreal / Blender / Python queues, file locks, dependent batches, cancel and restart, run history with district, machine and commit, and a metrics view.
SafetyLocalhost only, with Host and Origin checks; commands are fixed argument lists, never shell text; malformed or wrong-district results become ERROR, never PASS.
Your scriptsPR #3 changes none of them. PR #4 adds opt-in LS_UE_ROOT / LS_BUILDS_DIR to three PowerShell runners, with defaults unchanged.
Asset specPlaceholder only and not enforced. Budgets wait for you in Phase 2.
Known limitsReal Unreal cancellation is untested; brk_grid_island.py still has hard-coded paths; no wrapper was found for the unit suite; Preview returns four PNGs rather than one contact sheet.
Before mergingScrub the committed evidence logs: they contain a local Windows user path and machine name. Consider keeping the screenshots (about 7 MB) out of git history.
06 · The plan

Five phases. You review each.

Codex builds on a new branch off unreal-island-archipelago and opens PRs into it. It stops after every phase with evidence: run folders, test counts, screenshots, and what wasn't verified.

01

Launcher, district-aware from day one

Built · awaiting your workstation run (PR #3, #4)
  • Python standard-library server on 127.0.0.1, with one page in Lost Souls styling.
  • Data-driven job registry; Brickell district manifest pointing at existing files.
  • Separate queues by resource, batches, cancel that cleans up (including the multiplayer ini restore), and "this will close Unreal" warnings.
  • Run history: job, district, machine, commit, result, log, images.

Done when: every job has run on your machine with evidence, the parsers agree with a human reading the log, and invalid input is refused.

02

Standards

The definition of "done"
  • Asset spec drafted from your real asset measurements. You approve the numbers.
  • Validator run on all 34 Brickell assets as a report. Nothing gets "fixed" in this phase.
  • Asset lifecycle plus an Assets view with Approve / Reject.
  • Optional --district on intake and preview (default Brickell; your commands unchanged).
  • District test set, run from one button.
03

Asset factory

Reference → validated asset, in batches
  • Tripo API, built from the official docs. The key is never stored in the repo or logs.
  • Estimated credits before submitting, plus a per-batch and monthly cap.
  • A 5-asset pilot batch first, then the full 32-image T01–V04 queue.
  • Timed against the manual route.
04

Second-district pilot

The real test of scale
  • You and Legend pick a small district.
  • new_district.py creates the manifest, folders and a map from a template you approve.
  • Grid and tests read IDs from the district instead of hard-coding F1c / E090 / Worker1. These are small PRs for you, defaulting to Brickell.
  • Bootstrap time compared with Brickell's history.
05

Connect to V3

Design first, after V3 shared saving is live
  • The launcher pulls requests from V3. A human still clicks Run.
  • Results and asset status push back, so the district progress view shows real data.
  • The launcher is never exposed to the internet.
07 · For Quik

What we need from you

Questions
  1. Review PR #3 and PR #4, then run all 19 buttons on your workstation and keep the run folders, failures included.
  2. Which scripts do you run most? Those are buttons #1.
  3. Which script runs the 20/20 unit suite?
  4. Blender version, and is it on PATH?
  5. Asset status: a new field in brickell_assets.json, or a sidecar file?
  6. Triangle and texture budgets per category.
  7. Which map should be the new-district template?
  8. Is D:\Builds\ still where builds go?
  9. Anything you never want behind a button?
Ground rules
  • Codex builds on its own branch and PRs into yours. You approve.
  • Your scripts aren't changed. Any generalization is a small, opt-in PR that defaults to today's behavior.
  • Only registered scripts run, on localhost only. No arbitrary commands, and no debug probes.
  • Warns before anything closes Unreal.
  • Nothing touches main without Legend.
  • No spending past the Tripo budget Legend sets.