DonovanMods · dyoung522 · Aug 2, 2026 · Jul 31, 2026 · Jul 31, 2026 · Jul 31, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -7,6 +7,32 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.28.0] - 2026-08-02
+
+### Added
+
+- CLI output is now colorized by default when stdout is a terminal, extending the existing `colorGreen`/`colorRed`/`colorYellow` accent mechanism (previously only used by `deploy`/`verify`) with a full 4-color palette (green/yellow/red/cyan, plus bold/dim) across `list`, `status`, `search`, `update`, `conflicts`, and `mod show`. Table headers are bold+cyan. `lmm list` tints the whole row identically with or without `-v` (the row-tint decision is a single shared helper keyed on the mod's actual state, not the display flag): green for the common enabled+deployed case, yellow for enabled-but-undeployed, dim for disabled. `search` tints an installed mod's whole row green; `update`'s POLICY column colors per row. `status`/`mod show` color their values, not just the odd count: `lmm status -g <game>`'s active profile and per-profile "(active)" marker are green, mod/profile counts are cyan, Link Method is cyan, Last Deploy is green (or dim when never deployed); `mod show`'s Version fields are cyan and its Update policy is colored per state (green for auto, yellow for pinned); `conflicts`' stale winner suffix is yellow; and the existing `✓`/`✗` success/failure markers extend to `update` and `mod`'s confirmation lines. Detection is TTY-aware (piped/redirected output stays plain) and layers on top of the existing `--no-color` flag and `NO_COLOR` env var (presence-only per no-color.org), which continue to work unchanged; `--json` output is never colored. Table color is applied only to already-tabwriter-padded text (accented headers, whole-row tints, or a table's genuinely last column) — never to interior cell values before they reach `text/tabwriter`, which pads columns by raw byte length and would misalign them (#112, #193)
+- **Icarus built-in mod source** (`internal/source/icarus`): a public, unauthenticated Firestore-backed catalog (Project Daedalus) — `lmm search`/`install`/`update` work against it like NexusMods/CurseForge. A `.exmodz` mod file is validated and its row-level table diffs retained via a new, game-agnostic `internal/unrealpak` PAK reader/writer and the new `deploy_mode: compile` game setting (see the merged-pak bullet below for how it actually deploys); a plain `.pak` file from the same catalog is unaffected and deploys through the existing extract/copy pipeline unchanged. Base data tables are read directly from the installed game's own `data.pak`, so a merge always matches the installed game version and works entirely offline; `internal/unrealpak` reads both the stored and the Zlib-compressed entries that pak contains, using only the standard library (#136, #175)
+- `lmm game detect` now recognizes Icarus (Steam App ID `1149460`) and generates a complete `games.yaml` entry for it (`deploy_mode: compile`, `sources: {icarus: icarus}`) — no more hand-editing `games.yaml` to get started. The known-games schema (`steam-games.yaml`, built-in or your own override) gained two optional fields, `deploy_mode` and `sources`, generalizing detection beyond NexusMods-only games; every existing entry is unaffected (#177)
+- Custom `api` sources' `search` endpoint gains `{category}`/`{tags}` path placeholders, fed from `SearchQuery.Category`/`.Tags` (URL-escaped; multiple tags comma-joined) — previously these were silently dropped with no way for a declarative source to express category/tag filtering. A definition whose `search` path omits the new placeholders is unaffected: the values are computed but never substituted in, matching today's behavior exactly (#120)
+- Compiled mods (`deploy_mode: compile`, e.g. Icarus) with more than one enabled `.exmodz` mod now compose correctly instead of silently shadowing each other: every enabled mod's table-row diffs are applied sequentially, in profile load order, into ONE merged `zzz_LMM_Merged_P.pak` per profile (named to mount last, so it always wins over a plain prebuilt `.pak`'s own table override) — two mods patching different fields of the same row, or entirely different rows of the same table, both survive; only a genuine same-field conflict is last-wins, and a bundled-asset path collision (which can't compose) is last-wins with a loud warning. This also fixes "the Friday problem" (a weekly base-pak refresh silently reverting a mod's patched tables, with nothing to notice): the merge regenerates whenever the enabled-mod set, load order, a mod's version, or the base pak itself changes. `lmm update` (CLI and TUI) reports a "recompile needed" row for the profile's merged pak (additive `--json` field `recompile_needed`/`reason`) alongside normal version updates; applying it regenerates and redeploys — pinned mods' diffs recompile normally, and a LOCKED mod's diff still participates in every merge (a lock pins that mod's own version, not the profile's merged pak). Installing/importing a `.exmodz` now only validates and retains it (a per-mod compiled pak is no longer generated or deployed); a plain prebuilt `.pak` mod, and every non-`deploy_mode: compile` game, is completely unaffected. `lmm verify` gains a matching "RECOMPILE NEEDED" warning row (`stale_compile`) for the profile's merged pak (#136, #175, #196, #197)
+
+### Changed
+
+- `lmm list` now displays mods in the profile's load order (the same order `lmm profile reorder` sets and the TUI's mod list already showed) instead of DB install order (`installed_at`) — the visible order is now the order that actually decides merge precedence for a `deploy_mode: compile` game. A mod installed but missing from the load order is still shown, never silently dropped, placed first (lowest priority). README and `docs/configuration.md` gain a "Merge precedence" paragraph explaining that later-in-load-order mods win conflicting table-row _fields_ (a per-field upsert; untouched fields from earlier mods survive) while bundled assets are whole-file last-wins with a warning, and that `lmm profile reorder` regenerates the merged pak immediately (#201)
+- An unrecognized, non-empty `link_method` (`games.yaml`, profile files, imported profiles) or `deploy_mode` (`games.yaml`; also `lmm game detect`'s `steam-games.yaml`) is now a load-time error naming the field, the offending value, the owning game/profile, and the valid options — instead of silently falling back to the default (`symlink`/`extract`). **Breaking for configs that were already silently misbehaving:** a typo like `deploy_mode: compil` previously ran as `extract` with no warning; it now refuses to load until fixed. An empty/absent value is unaffected and keeps today's default exactly (#172)
+
+### Fixed
+
+- `lmm uninstall` left behind the now-empty per-mod cache directory (`<source>-<modID>/`) after removing a mod's last cached version — cosmetic litter that accumulated indefinitely. `Cache.Delete` now removes the container too, but only when it's actually empty (a mod with other cached versions, or `--keep-cache`, is unaffected) (#190)
+- `lmm game detect` printed every stale-library warning twice on Linux installs where `~/.steam/steam` is a symlink to `~/.local/share/Steam` (or the reverse): both existed as candidate Steam roots, so the whole library scan — and every warning it produced — ran twice against the identical real directory. `FindSteamRoots` now dedups candidates by their resolved (symlink-followed) real path, keeping only the first; a genuinely separate second root is unaffected (#190)
+- `internal/unrealpak`'s unsupported-compression-method refusal named a blank `CompressionMethods` table slot as `compression method "" (index N)` — an in-range but never-named slot, distinct from a genuinely out-of-range index. It now names the slot itself (`unnamed method N`), so the refusal is actionable instead of pointing at an empty string (#190)
+- `lmm install` (and the TUI's equivalent progress line) announced a `.exmodz` compile step as "Extracting to cache..." — actively misleading, since compiling never extracts anything. It now prints "Compiling `<source>` → `<output>`..." instead, naming the actual archive and the compiled `_P.pak` output; a plain extract/copy install is unaffected and keeps today's exact "Extracting to cache..." text (#190)
+- `Service.GetEffectiveLinkMethod` — the profile > game > global resolution behind every deploy/install/import/status/verify operation — no longer silently swallows an invalid profile `link_method`: since #172, `config.LoadProfile` fails loud on an unrecognized value, but `GetEffectiveLinkMethod` treated ANY profile-load error, including that one, as "no explicit override" and fell back to the game/global default with nothing surfaced — meaning a hand-edited profile with a typo'd `link_method` could deploy with the wrong method, no error, no warning. It now distinguishes that case (`errors.Is(err, domain.ErrInvalidLinkMethod)`) from a missing/unreadable profile file — which still degrades silently by design, since profiles are optional — and returns the validation error instead, propagated through every call site (`DeployProfile`, `EnableMod`/`DisableMod`, `ApplyInstall`/`ApplyUpdate`/`ApplyRollback`/`ApplyImport`/`ApplyProfileSwitch`, `lmm deploy`/`import`/`install`/`profile apply`/`verify --fix`). Narrow in practice — both save and load paths validate now, so only a profile hand-edited after the fact can trigger it — but it closes the one place #172's fail-loud contract didn't reach (#189)
+- `lmm import` of a local `.exmodz` file for a `deploy_mode: compile` game (Icarus) now routes through the same compile step as a download: it resolves the game's mapped `source.Compiler`-capable source from the registry and compiles the archive against the installed base pak, caching the resulting `_P.pak` the same way `DownloadModToCache` does. Previously the import path extracted/copied `.exmodz` files as-is, landing an uncompiled archive in the cache instead of a deployable pak. A missing compiler-capable source or missing base pak now fails loud with an actionable error rather than silently caching the uncompiled file; non-`.exmodz` imports are unaffected (#173)
+- `lmm mod disable` undeployed a mod's files and cleared `enabled`, but never cleared `deployed` — `lmm list -v` kept showing DEPLOYED yes after disable. The disable flow now clears `deployed` unconditionally after the undeploy attempt, even when the undeploy itself only partially succeeds (already a non-fatal, Note-reported condition), so the flag always reflects disable-intent rather than lagging behind a best-effort file cleanup. The symmetric enable path had the same gap — enabling a disabled mod re-deployed its files without ever setting `deployed` back to true — and is fixed the same way. Both `SetModDeployed` calls follow the same non-fatal Note convention already used by `DeployProfile`/`PurgeProfile` for this same setter: a failure to record the flag doesn't block the primary enable/disable outcome (#183)
+- `lmm install`'s multi-select path (`batchInstallMods`, used whenever more than one search result is selected) never synced the profile's merged pak for a `deploy_mode: compile` game (Icarus): installing `.exmodz` mods this way validated and cached them but left the merged pak undeployed, silently, until a later `lmm update` self-healed it. A seam audit of every mutation entry point that can change a profile's enabled-exmodz set/order/versions found and closed 4 more gaps with the same shape — `lmm profile apply`/`profile sync`, `lmm mod edit --version`, and `lmm verify --fix` also never reached the merged-pak sync. All five now call the same shared `Service.SyncMergedPak` entry point the rest of the flows already used, instead of bypassing it. Separately, several flows that _did_ call the sync already (including single-mod `lmm install`) only recorded a failure in a diagnostic field their own CLI caller never read back — a sync failure could be completely silent, not even `--verbose`-gated. Sync failures now surface unconditionally on stderr everywhere the merged pak is synced (CLI and TUI). Also corrected three misleading messages this bug produced along the way: `lmm install`/`import` say "Installed (merged pak updated)" instead of a false "(0 files)" for a compile-mode mod (deploying zero files of its own is correct, not a failure); `lmm mod files` explains that such a mod participates in the profile's merged pak instead of suggesting it "may need to be redeployed"; and `lmm verify`'s "RECOMPILE NEEDED" row now names the real cause (a changed base pak vs. an artifact simply missing from disk) and points at `lmm update --all`, since this row is never auto-applied by a bare `lmm update` (#197)
+
 ## [1.27.1] - 2026-07-30
 
 ### Fixed
@@ -1156,7 +1182,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Comprehensive test coverage for core components
 - MIT License
 
-[Unreleased]: https://github.com/DonovanMods/linux-mod-manager/compare/v1.27.1...HEAD
+[Unreleased]: https://github.com/DonovanMods/linux-mod-manager/compare/v1.28.0...HEAD
+[1.28.0]: https://github.com/DonovanMods/linux-mod-manager/compare/v1.27.1...v1.28.0
 [1.27.1]: https://github.com/DonovanMods/linux-mod-manager/compare/v1.27.0...v1.27.1
 [1.27.0]: https://github.com/DonovanMods/linux-mod-manager/compare/v1.26.0...v1.27.0
 [1.26.0]: https://github.com/DonovanMods/linux-mod-manager/compare/v1.25.0...v1.26.0

diff --git a/README.md b/README.md
@@ -421,8 +421,20 @@ games:
       nexusmods: "starfield"
     link_method: copy # This game requires file copies instead of symlinks
     cache_path: /mnt/fast-ssd/starfield-mods # Store this game's mods on fast storage
+
+  icarus:
+    name: "Icarus"
+    install_path: "/path/to/Steam/steamapps/common/Icarus"
+    mod_path: "/path/to/Steam/steamapps/common/Icarus/Icarus/Content/Paks/mods"
+    sources:
+      icarus: "icarus"
+    deploy_mode: compile
 ```
 
+Steam auto-detection (`lmm game detect`) knows about Icarus (App ID `1149460`) and generates an equivalent entry for you, `install_path`/`mod_path` filled in from your actual Steam library — the YAML above is kept here as reference for what gets written, not something you need to type by hand.
+
+**Merge precedence**: with more than one `compile`-mode mod installed (currently Icarus only), the profile's load order — the same order `lmm list` displays and `lmm profile reorder` changes — decides how conflicting changes resolve. Mods are merged in load order, so a mod later in the list is applied later and wins conflicting _fields_ on a shared data-table row; it's a per-field upsert, not a whole-row overwrite, so untouched fields from earlier mods still survive. Bundled asset files can't compose that way — a same-path collision between two mods is whole-file last-wins, and installing or updating a colliding mod prints a warning naming both. Either way, the bottom of the load order has final say, and `lmm profile reorder` regenerates the merged pak immediately, so a reorder's effect on precedence is visible right away rather than at the next deploy.
+
 ### Deployment Methods
 
 Mods can be deployed using three methods:
@@ -655,6 +667,8 @@ api:
 | `{page}`      | The internal 0-based page number, plus `page_start` (default `1`)                                                                        | `search`                                         |
 | `{page_size}` | The requested page size (defaults to 20 when unspecified or ≤ 0)                                                                         | `search`                                         |
 | `{offset}`    | The internal 0-based page × `page_size` — independent of `page_start`, for offset-paginated APIs                                         | `search`                                         |
+| `{category}`  | The search query's category filter (source-specific ID or name), empty when unset                                                        | `search`                                         |
+| `{tags}`      | The search query's tag filters, comma-joined, empty when unset                                                                           | `search`                                         |
 | `{mod_id}`    | The mod ID                                                                                                                               | `get_mod`, `mod_files`, `download_url`           |
 | `{file_id}`   | The file ID                                                                                                                              | `download_url`                                   |
 
@@ -859,6 +873,8 @@ A `directory` source now shows up with real capabilities in `lmm source list` (`
 | `--no-hooks` |       | Disable all hooks at runtime                                                                                            |
 | `--no-color` |       | Disable colored output (respects NO_COLOR env)                                                                          |
 
+Output is colorized by default whenever stdout is a terminal (headers, status accents like enabled/disabled/pinned, success/warning/error markers); piped or redirected output stays plain automatically, and `--json` output is never colored. Disable explicitly with `--no-color` or the `NO_COLOR` environment variable.
+
 ### Commands
 
 | Command                                            | Description                                                                                                                       |
@@ -1085,7 +1101,7 @@ The mod cache location can be customized via `cache_path` in `config.yaml`. Sett
 - [x] Automatic dependency installation (opt out with `--no-deps`)
 - [x] Interactive TUI (Bubble Tea) - see the Terminal UI section above
 - [x] CurseForge integration
-- [ ] Additional first-party built-in sources beyond NexusMods/CurseForge
+- [x] Additional first-party built-in sources beyond NexusMods/CurseForge (Icarus)
 - [ ] Game auto-detection beyond Steam (Lutris, Heroic, Flatpak)
 - [ ] Backup and restore