[bcupello]

Please read #289 first — it asks how you would like this split, and flags that one board item is not deliverable as written. Opened as requested rather than assumed to be wanted as-is.

Stacked on #292. This branch contains that commit too, because both touch the same switch in AnyGame.init(from:). Happy to rebase once it lands.

Implements most of the Steam Development Roadmap on the current engine. SteamCMD is a native macOS binary and never involves Wine, so installing, updating, verifying and enumerating all work today; Wine is only involved in launching a downloaded title, through the path that already exists.

Covers: SteamCMD installation, Functional signin + steam guard support, Fetch game list, Game installing + progress handling, Game moving, Game update checker, Custom install directory, SteamCMD autoupdate parser.

Layering

ValveDataFormat and SteamCMDOutput are dependency-free readers for Valve KeyValues and for SteamCMD output. SteamCMD drives the binary. Steam is the storefront service — the peer of Legendary. SteamGameManager conforms to StorefrontGameManager.

Things that had to be measured

Each of these produced wrong behaviour first, so each is recorded at its call site:

• licenses_print caps the app IDs printed per package and says so ((N in total)). On a real account 171 of 455 sat behind one elided line, recovered with package_info_print.
• @ShutdownOnFailedCommand defaults to 1, abandoning every remaining command once one fails. An empty app_info_print counts, so a 25-app metadata batch returned exactly one result.
• common/type casing is inconsistent — game for Terraria and Dota 2, Game for Palworld, in one batch.
• The bootstrapper localises its output mid-stream, so only the numeric [ n%] / [----] prefixes are matched, never English words.
• force_install_dir must precede login, or SteamCMD refuses outright.
• Exit code 42 is the self-update mechanism, handled by steamcmd.sh — which is why the script is invoked, not the binary.
• The mobile-authenticator flow reports its outcome on its own line (Waiting for confirmation...OK); to Steam Public... never gains a verdict in that case.
• SteamCMD writes its session relative to HOME, by default the user’s real Steam installation. HOME is scoped: 15 files written under Mythic’s directory, zero touched in the user’s.
• Bulk output is read whole, not streamed. Process.runStreamed splits whatever a single read() returned, so a line spanning two reads arrives as two fragments — an enumeration that received 188 licence lines parsed 7. Streaming stays for downloads and sign-in, which tolerate fragments.
• Concurrent runs share one session directory and corrupt each other rather than failing: two overlapping enumerations returned 4 and 18 app IDs where one returns 455.

Not attempted

The Steam client's own UI does not run on the engine — its Chromium 126 helper crash-loops on wine-7.7 — so the client can be started but never signs in, and anything needing a Steam auth ticket fails.

That is narrower than "cannot launch", which is what I first wrote here and have corrected in #289. The game I tested does launch and run; what fails is online authentication, and it says so itself (ConnectLoginNoEAS() failed due to the platform OSS giving an empty auth token). A title that never asks for a ticket may well be fine. Launches that do fail now explain themselves instead of silently doing nothing.

Verification

Exercised end to end against a real account on Engine 2.6.1, in the app rather than in a harness:

Step	Result
SteamCMD installed from the Accounts pane	15 files under Mythic's own directory, zero touched in the user's real Steam install
Signed in	Steam Mobile App approval; Waiting for confirmation...OK, then [Logged On] with a real SteamID
Library enumerated	259 owned app IDs from 113 licences → 103 games, DLC and soundtracks filtered by common/type
Title installed	41.3 GB requested, 38 GB on disk, Success! App '1623730' fully installed.
Update state	read from appmanifest_<appid>.acf: StateFlags 4, buildid == TargetBuildID
Launch	starts, reaches RHI init, then exits — it needs a live signed-in client (measured; see the issue comment)

xcodebuild clean and swiftlint clean for every file touched — the three warnings it still reports all predate this branch.

The parsing layers additionally carry 67 assertions over verbatim SteamCMD captures, compiled standalone since there is no test target. Not included here (it trips force_try under your .swiftlint.yml); offered in #289.

Screenshots

Hosted on a branch of the fork rather than committed here, so they add nothing to this diff.

Library, after enumeration — 103 titles, artwork from Steam's CDN, app ID on each badge:

Import by app ID — the tab that was hidden and .hidden() upstream. Empty state explains itself rather than asking the CDN for artwork it cannot have:

A launch that fails now says so. Previously the operation completed without error, so the UI showed a brief spinner and nothing else:

Interface language is the tester's; the strings this PR adds are English source entries in the catalogue.