feat(T11983): cleo setup wizard e2e — prefs → connect → model (fit-gated) → profile → TUI offer

.changeset/t11983-setup-wizard-e2e.md

@@ -0,0 +1,56 @@
+---
+id: t11983-setup-wizard-e2e
+tasks: [T11983]
+kind: feat
+summary: "cleo setup wizard e2e: fit-gated Ollama model picker, firstRunComplete flag, whoami summary + TUI offer after first-run, 16-test e2e TTY-simulated suite"
+---
+
+Closes the last batteries-included onboarding gap (T11983 / E6-ONBOARDING-FRONT-DOOR):
+`cleo setup` now completes the full prefs → provider connect → model pick → profile →
+validated round-trip → TUI offer flow in a single interactive session.
+
+**Changes:**
+
+- **`packages/core/src/setup/sections/models-roles.ts`** (modified):
+  - New `_pickOllamaModelInteractive(io, ranker?)` export — when the active provider
+    is `ollama`, calls `rankLocalModelFit` to detect hardware and rank 2–3 best-fit
+    open-weight models. Below the 4 GB RAM floor, prints cloud-only guidance and
+    returns `null` (RECOMMEND-NEVER-KILL: never auto-pulls, never auto-selects).
+    Already-pulled models surface with a `[pulled]` annotation. Offers a manual-entry
+    fallback so the user is never locked in to the ranked list.
+  - Injectable `ranker` parameter keeps the function testable without OS/network calls.
+  - `createModelsRolesSection().run()`: branches to `_pickOllamaModelInteractive` when
+    the default provider is `ollama`; all other providers continue using the existing
+    catalog-picker path.
+
+- **`packages/cleo/src/cli/commands/setup.ts`** (modified):
+  - `CleoSetupResult` gains `firstRunComplete: boolean` (mirrors
+    `WizardRunResult.firstRunComplete`). Single-section runs always produce `false`;
+    full-wizard runs produce `true` iff every section succeeded.
+  - New `_printWhoamiSummaryAndOfferTui(io)` export — reads agent name, provider, and
+    model from config then prints a `cleo whoami`-style summary to stderr after a
+    successful first-run completion. Prompts the user to optionally launch `cleo tui`
+    (fire-and-forget spawn, never blocks wizard exit).
+  - `_readWhoamiSnapshot()` internal helper for best-effort config + credential pool
+    reads (never throws).
+  - Command `run()` calls `_printWhoamiSummaryAndOfferTui` after `cliOutput` when
+    `result.ok && result.firstRunComplete`.
+
+- **`packages/cleo/src/cli/commands/__tests__/setup-wizard-e2e.test.ts`** (new):
+  - 16-test TTY-simulated e2e suite using `StubWizardIO` as the TTY simulation surface
+    (the canonical CLEO approach — the wizard engine is I/O-agnostic by design).
+  - Covers: full 10-section wizard pass with `firstRunComplete=true`; section-failure
+    → `firstRunComplete=false`; single-`--section` always `false`; whoami summary
+    content assertions; graceful non-TTY handling of the TUI offer.
+  - Fit-gated Ollama path: below-floor cloud-only guidance, ranked recommendations,
+    `[pulled]` annotation, skip/manual-entry/ranker-failure fallback paths.
+  - `firstRunComplete` propagation verified for full and single-section runs.
+
+**AC compliance:**
+
+- Fresh-install `cleo setup` completes prefs → provider connect → model (fit-gated
+  for Ollama, catalog for cloud) → profile → verification in canonical order.
+- Reuses shipped building blocks: `llm` (T11725), `models-roles` (T11726),
+  `models-roles` oauth-inline path (T11727), `cleo llm fit` envelope (T11982).
+- Ends with whoami summary + TUI launch offer.
+- E2E test with TTY-simulated (`StubWizardIO`) input: 16 tests all green.