chore: migrate repo to TaraxioT org, free-tier CI, and deterministic re-probe test

[0xgeorgemathew]

Problem

The repo moved from 0xgeorgemathew/t3trade to the TaraxioT org, but 16+ tracked references still pointed at the old location, the install doc pointed users at upstream releases, release.yml hid GitHub-hosted macOS/Windows matrix entries that would burn non-free runner minutes if dispatched, and the ProviderRegistry codex re-probe test flaked on 2-vCPU CI runners (2 of 4 runs today).

What this does

• 47ee6d903 — point all repo links (README, marketing constants, installer, server package metadata, upstream docs) at TaraxioT/t3trade; upstream pingdotgg/t3code attribution untouched.
• 0a754c0fc — comment out release.yml's native-OS build matrix entries so a dispatch can't run paid GitHub-hosted macOS/Windows jobs. All 30 active runs-on across 12 workflows are now blacksmith-2vcpu-ubuntu-2404 (free tier); no cron schedules remain.
• 348c5d923 — replace the re-probe test's budgeted poll with a Deferred awaited on the real spawn's exit; verified 15x locally plus CPU-pressure runs and green twice on the 2-vCPU runner.
• 3f116f6ce — install doc: fork releases link + clarify winget/brew/AUR deliver upstream T3 Code.

Note: base branch review/base-pre-migration is pinned at the pre-migration commit a6cad93a2 so this PR shows exactly the migration diff; merge target in practice is keeping main as-is.

Work performed by GLM-5.3 via the ZCode harness.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is enabled.}

Note

Add full pull request system, mobile theming overhaul, and migrate repo to TaraxioT org

• Introduces end-to-end pull request support: contracts in pullRequest.ts, provider implementations for GitHub/GitLab/Bitbucket/Azure DevOps, PullRequestService with caching and rate-limit handling, HTTP diff endpoint, client-runtime atom families, and a full web PR route with list/detail/timeline/summary tabs, review threads, reactions, and reviewer management
• Overhauls mobile theming: replaces useColorScheme with useAppearancePreferences throughout, adds MobileThemeId selection with per-appearance light/dark themes, native diff/editor theming via createNativeReviewDiffTheme, and showcase capture theme coordination
• Migrates repo identity from t3code to TaraxioT/t3trade: renames Electron protocol schemes, Linux WM class/desktop entries, SSH askpass dir, WSL markers, installer URLs, and shared T3_HOME_DIR_NAME constant
• Adds desktop hold-to-quit (QuitHold), favicon capture pipeline, remote editor deep-link support (buildRemoteOpenUrl), and preview tab zoom/appearance defaults
• Migrates relay infrastructure from Planetscale to self-hosted Postgres via Cloudflare Access tunnel with deploy-time migrations
• Risk: sidebar-filter-button.tsx and sidebar-header-actions.tsx reference useThemeColor without importing it, causing runtime errors; trading-contracts/src/observation.ts renames funding fields (mean→meanPer8h, latestRate→latestRatePer8h) which is a breaking schema change for consumers; CloudCliOAuthConfig shape changes (authorizationEndpoint removed, loopbackPort added)

^{Macroscope summarized bdf852f.}

[macroscopeapp]

Reviewed the web UI surface of this sync against the shared component system and Tailwind/theme ownership rules.

What checks out:

• MissionLivePanel's log rows and order legs now use Popover + PopoverTrigger openOnHover with the named tooltipStyle popup variant instead of a native title on a non-interactive element, so the detail is reachable by keyboard and touch — this resolves the finding from the previous run.
• The remaining native title attributes in apps/web/src/components/trading/** carry explicit oxlint-disable-next-line t3code/no-native-title-tooltip directives with a stated reason, which is a documented deferral rather than a silent evasion.
• BrowserDeviceToolbar moves its aspect-ratio hint from title to Tooltip/TooltipPopup while keeping the Button variant="ghost" size="icon-xs" primitive and aria-pressed.
• The index.css restructure keeps the mask/fade contracts intact: topbar-scroll-fade and virtualized-scroll-fade both preserve the opaque scrollbar lane, and every retired selector (settings-page-scroll-fade, workspace-titlebar-controls, sidebar-brand) has its replacement consumer (data-settings-page-scroll, data-workspace-titlebar-controls, SidebarChrome/sidebar.tsx) in this diff.
• The new _chat.pull-requests.tsx route composes ui/* primitives throughout — no raw button/input reconstructions.

One finding below, on the new fork-owned stylesheet.

^{Posted via Macroscope — UI Consistency}

[macroscopeapp]

Reviewed the changed web UI surfaces against the shared component system and Tailwind ownership rules.

The index.css → owner/utility split, trading.css (now on @variant dark/light), the new virtualized-scroll-fade contract, the source-level MenuPopup width defaulting, and the new compact/micro/icon-micro/ghost-muted/glass primitive variants all look consistent, and the previously native-title tooltips are now real Tooltips (with t3code/no-native-title-tooltip enforced).

Two findings in the new pull-request UI, where controls are hand-rolled instead of going through Button. Both matter beyond style: Tailwind v4's preflight gives a bare <button> cursor: default, and buttonVariants is what restores cursor-pointer, the pointer-coarse 44px hit target and disabled:opacity-64 — so these controls behave differently from every other control in the app.

^{Posted via Macroscope — UI Consistency}

[macroscopeapp]

Reviewed the changed web UI surfaces against the shared component system and Tailwind ownership rules.

The index.css → owner/utility split, trading.css (now on @variant dark/light), the new virtualized-scroll-fade contract, the source-level MenuPopup width defaulting, and the new compact/micro/icon-micro/ghost-muted/glass primitive variants all look consistent, and the previously native-title tooltips are now real Tooltips (with t3code/no-native-title-tooltip enforced).

Two findings in the new pull-request UI, where controls are hand-rolled instead of going through Button. Both matter beyond style: Tailwind v4's preflight gives a bare <button> cursor: default, and buttonVariants is what restores cursor-pointer, the pointer-coarse 44px hit target and disabled:opacity-64 — so these controls behave differently from every other control in the app.

^{Posted via Macroscope — UI Consistency}

[macroscopeapp]

🟠 High plans/Executables.md:68

Every G1–G6 execution fails at mandatory step 0 because artifacts/investigations/G-RULES.md is required but is absent from the reviewed tree; workers therefore have no Protection rules before they are instructed to start isolated trading missions. Add and track the shared rules file (or provide it through a guaranteed bootstrap) before these prompts can be executed safely.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @artifacts/plans/Executables.md around line 68:

Every G1–G6 execution fails at mandatory step 0 because `artifacts/investigations/G-RULES.md` is required but is absent from the reviewed tree; workers therefore have no Protection rules before they are instructed to start isolated trading missions. Add and track the shared rules file (or provide it through a guaranteed bootstrap) before these prompts can be executed safely.

Evidence trail:
artifacts/plans/Executables.md:68-70, 97-100, 134-140, 186-190, 223-227, 270-274, 309-313, 353-357 @ a9ffeb09343e899d0f8f5738c65f8352bec436ff; repository lookup: artifacts/investigations/G-RULES.md at REVIEWED_COMMIT (not found); git grep "G-RULES|bootstrap" REVIEWED_COMMIT

[macroscopeapp]

🟡 Medium plans/plan-39-orders-in-positions.md:171

The planned open row cannot display truthful live P&L per opening leg: the available live position only provides aggregate P&L, so scaled entries at different prices have no basis for allocation (for example, $100 and $200 buys at a $150 mark are +$50 and -$50 while the aggregate is $0). Define a lot-allocation rule such as FIFO and expose the required per-lot entry data, or change the row requirement to show aggregate P&L.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @artifacts/plans/plan-39-orders-in-positions.md around line 171:

The planned `open` row cannot display truthful live P&L per opening leg: the available live position only provides aggregate P&L, so scaled entries at different prices have no basis for allocation (for example, $100 and $200 buys at a $150 mark are +$50 and -$50 while the aggregate is $0). Define a lot-allocation rule such as FIFO and expose the required per-lot entry data, or change the row requirement to show aggregate P&L.

Evidence trail:
artifacts/plans/plan-39-orders-in-positions.md:160-172 @ a9ffeb09343e899d0f8f5738c65f8352bec436ff
apps/web/src/components/trading/tradingPresentation.ts:2645-2665, 2680-2715, 2750-2776 @ a9ffeb09343e899d0f8f5738c65f8352bec436ff

[macroscopeapp]

🟠 High plans/Executables.md:229

G3 can block forever before the audit loop, server shutdown, and report generation when its discretionary mission stands aside without publishing a plan. Add a bounded wait with a failure/NOT-EXERCISED path that still performs cleanup and writes the required report.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @artifacts/plans/Executables.md around line 229:

G3 can block forever before the audit loop, server shutdown, and report generation when its discretionary mission stands aside without publishing a plan. Add a bounded wait with a failure/`NOT-EXERCISED` path that still performs cleanup and writes the required report.

Evidence trail:
artifacts/plans/Executables.md:145-155, 217-231 at c2d2c9f698; git blame REVIEWED_COMMIT -- artifacts/plans/Executables.md -L 217,231

[macroscopeapp]

🟡 Medium plans/Executables.md:155

When no harness run appears within 10 minutes, the executor jumps from step 4 to step 8, so it never stops the dev server or copies the isolated state.sqlite; failed runs therefore leak the server/process and omit the reproduction evidence. Route this failure path through explicit cleanup and database capture before reporting the failure.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @artifacts/plans/Executables.md around line 155:

When no harness run appears within 10 minutes, the executor jumps from step 4 to step 8, so it never stops the dev server or copies the isolated `state.sqlite`; failed runs therefore leak the server/process and omit the reproduction evidence. Route this failure path through explicit cleanup and database capture before reporting the failure.

Evidence trail:
artifacts/plans/Executables.md:152-173, c2d2c9f698

[macroscopeapp]

🟡 Medium plans/Executables.md:425

Following this checklist keeps the ~/.t3 archiver running while potentially stopping T3 Trade's canonical ~/.t3trade archiver, so market data collection stops for this fork and an unrelated upstream process is retained. Update the preserved archiver path to ~/.t3trade.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @artifacts/plans/Executables.md around line 425:

Following this checklist keeps the `~/.t3` archiver running while potentially stopping T3 Trade's canonical `~/.t3trade` archiver, so market data collection stops for this fork and an unrelated upstream process is retained. Update the preserved archiver path to `~/.t3trade`.

Evidence trail:
artifacts/plans/Executables.md:414-427 (REVIEWED_COMMIT)
apps/server/src/trading/archive/config.ts:105-113 (REVIEWED_COMMIT)
packages/shared/src/forkPaths.ts:15-20 (REVIEWED_COMMIT)
README.md:137-164 (REVIEWED_COMMIT)

[macroscopeapp]

🟡 Medium plans/Executables.md:134

The G1–G4/G6 prompts point executors to the nonexistent .claude/skills/test-t3-app/SKILL.md, so their mandatory first step fails and they never receive the isolation or browser-handling rules required by later steps. Update each reference to the versioned .agents/skills/test-t3-app/SKILL.md path.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @artifacts/plans/Executables.md around line 134:

The G1–G4/G6 prompts point executors to the nonexistent `.claude/skills/test-t3-app/SKILL.md`, so their mandatory first step fails and they never receive the isolation or browser-handling rules required by later steps. Update each reference to the versioned `.agents/skills/test-t3-app/SKILL.md` path.

Evidence trail:
artifacts/plans/Executables.md:134-135, 186-187, 223-224, 270-271, 353-354 at 885fad79a08c2a8ad96e900538e599d8da375e7a; repository glob at REVIEWED_COMMIT: `.agents/skills/test-t3-app/SKILL.md` exists and no `.claude/skills/test-t3-app/SKILL.md` match exists.

[macroscopeapp]

Effect service conventions review, scoped to the Effect service code this PR adds under apps/**, packages/**, infra/**.

The new server-side pull-request stack (apps/server/src/pullRequest/*) and apps/desktop/src/wsl/DesktopWslServerTree.ts follow the canonical layout (subpath namespace imports, inline Context.Service interface, exported make, export const layer), so no findings there. Two deviations in the newly added client-runtime pull-request service module are noted inline.

^{Posted via Macroscope — Effect Service Conventions}

[macroscopeapp]

pullRequestDiffLoaderLayer erases the module namespace for what is otherwise a canonical single-service module. Suggest exporting the construction as make and the layer as layer, so consumers read PullRequestDiffHttp.PullRequestDiffLoader / PullRequestDiffHttp.layer.

This pairs with the individual re-export in state/pullRequests.ts (export { PullRequestDiffLoader, pullRequestDiffLoaderLayer } ...), which could become export * as PullRequestDiffHttp from "./pullRequestDiffHttp.ts" to keep the module shape intact.

^{Posted via Macroscope — Effect Service Conventions}

[macroscopeapp]

This new service failure stores an unstructured message as its only data. Consider declaring it with Schema.TaggedErrorClass and deriving message — the tag alone identifies this failure, so it needs no fields.

-export class EnvironmentHttpConnectionNotReadyError extends Data.TaggedError(
-  "EnvironmentHttpConnectionNotReadyError",
-)<{ readonly message: string }> {}
+export class EnvironmentHttpConnectionNotReadyError extends Schema.TaggedErrorClass<EnvironmentHttpConnectionNotReadyError>()(
+  "EnvironmentHttpConnectionNotReadyError",
+  {},
+) {
+  override get message(): string {
+    return "The environment HTTP connection is not ready.";
+  }
+}

That also needs import * as Schema from "effect/Schema", and the construction in the diff atom becomes new EnvironmentHttpConnectionNotReadyError({}).

^{Posted via Macroscope — Effect Service Conventions}

[macroscopeapp]

🟡 Medium marketing/DESIGN-CONTRACT.md:63

The claim that public/t3trade-screenshot.webp and docs/media/t3trade-mission.png have no remaining references is false: scripts/build-t3trade-themed-assets.py still generates both files. Running the asset rebuild therefore recreates the explicitly prohibited fake images, so remove or update those generator targets instead of only deleting the outputs.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @apps/marketing/DESIGN-CONTRACT.md around line 63:

The claim that `public/t3trade-screenshot.webp` and `docs/media/t3trade-mission.png` have no remaining references is false: `scripts/build-t3trade-themed-assets.py` still generates both files. Running the asset rebuild therefore recreates the explicitly prohibited fake images, so remove or update those generator targets instead of only deleting the outputs.

Evidence trail:
apps/marketing/DESIGN-CONTRACT.md:62-65 at REVIEWED_COMMIT; scripts/build-t3trade-themed-assets.py:7, 141-212 at REVIEWED_COMMIT; .gitignore:58-61 at REVIEWED_COMMIT; git diff MERGE_BASE..REVIEWED_COMMIT -- apps/marketing/DESIGN-CONTRACT.md

[0xgeorgemathew]

Closing: this was a review-only PR (main → pinned review/base-pre-migration). All four migration commits (47ee6d903, 0a754c0fc, 348c5d923, 3f116f6ce) are already on main, so merging would only push main onto the pinned review base and destroy the diff window.

The free-tier-CI portion of the description is also stale — .github/workflows/ was removed entirely in bdf852fa2 and aa8c4780f after this PR was opened.

Outstanding review findings are on files outside the migration commits and will be handled separately.