feat(tui): onboarding teaches the model + empty-state teachers

[pedrofuentes]

Summary

Onboarding now opens with a teaching intro step before the model picker: shows the Council mental model (experts → panels → convene → debate → conclusion) with a static ASCII pipeline. Enter advances to the existing model picker; Esc skips to home directly from intro. All existing model-picker logic (inFlight guard, captured-input effect, ↑↓ cursor, confirmation flow) is preserved unchanged.

List-screen empty states became next-step teachers using each screen's actual keybindings:

• Panels: No panels yet — [c] auto-compose from a topic, or [n] build from experts
• Experts: No experts yet — [n] create one. Experts are the members of your panels.
• Sessions: No debates yet — convene a panel to watch them deliberate

HomeScreen is untouched (launchpad already covers the first-move prompt).

Phase 10 TUI overhaul — PR-14

TDD

• RED commit: test(tui): add failing tests for onboarding intro + empty-state teachers (13 failing tests)
• GREEN commit: feat(tui): onboarding teaches the model + empty-state teachers (impl only, 4 source files)
• Full TUI suite: 864/864 passing · lint zero · typecheck clean

[pedrofuentes]

Status: APPROVED

Sentinel Review Report

Ref: feature/tui-onboarding-teach → main
Report ID: sentinel-1746-c2e0668
Reviewed SHA: c2e0668
Sentinel ruleset: v1
Reviewed at: 2026-06-26T08:51:04Z
Mode: standard
Review depth: Tier 2 (full)
Required action: MERGE

Phase 1 — TDD / Test Evidence

• Tests exist & meaningful: ✅ New tests assert intro copy/pipeline, Enter/Esc transitions, preserved model confirmation, and list empty-state teacher text. Evidence: tests/unit/tui/onboarding-screen.test.tsx:75-78 contains expect(frame).toContain("Council assembles AI experts into panels"); and expect(frame).not.toContain("claude-sonnet-4.5");; tests/unit/tui/panels-screen.test.tsx:56 contains expect(lastFrame()).toContain("[c] auto-compose from a topic, or [n] build from experts");.
• Test-first history verified: ✅ git --no-pager log --oneline origin/main..HEAD shows cdb7de9 test(tui): add failing tests for onboarding intro + empty-state teachers before c2e0668 feat(tui): onboarding teaches the model + empty-state teachers; git show --name-status c2e0668 lists only packages/cli/src/tui/screens/{ExpertsScreen,OnboardingScreen,PanelsScreen,SessionsScreen}.tsx; git show --name-status cdb7de9 lists only TUI test files.
• RED evidence: ✅ On cdb7de9, targeted tests failed as expected: Test Files 4 failed (4) and Tests 13 failed | 25 passed (38); representative failure: expected 'Welcome to Council!\nChoose the defau…' to contain 'Council assembles AI experts into pan…'.
• Full suite green on SHA: ✅ cd packages/cli && npx --yes pnpm@10.30.1 exec vitest run tests/unit/tui --reporter=dot → Test Files 93 passed (93) and Tests 864 passed (864).
• Lint: ✅ npx --yes pnpm@10.30.1 lint → eslint packages/cli/src packages/cli/tests exited 0.
• Typecheck: ✅ npx --yes pnpm@10.30.1 typecheck → tsc --noEmit exited 0.
• Coverage: N/A (threshold unset/not required for this review).

Phase 1.5 — Fast-path Evaluation

🔴 count: 0 | LOC: 149 non-test source changed (≤150: Y) | Security paths: Y (TUI terminal-rendering surface) | New deps: N | Commit types qualify: N (feat)
→ Fast-path eligible: NO → Phase 2

Phase 2 — Execution Log

Dim	Tool Call	Agent ID / Ref	Status
A1	task(agent_type="general-purpose", name="sentinel-dim-a1", model="claude-sonnet-4.6")	N/A (sync task tool returned no separate ID)	✅ No findings
A2	task(agent_type="general-purpose", name="sentinel-dim-a2", model="claude-sonnet-4.6")	N/A (sync task tool returned no separate ID)	✅ No findings
B	task(agent_type="general-purpose", name="sentinel-dim-b", model="claude-sonnet-4.6")	N/A (sync task tool returned no separate ID)	✅ No findings
C	task(agent_type="general-purpose", name="sentinel-dim-c", model="claude-sonnet-4.6")	N/A (sync task tool returned no separate ID)	✅ No findings
D	task(agent_type="general-purpose", name="sentinel-dim-d", model="claude-sonnet-4.6")	N/A (sync task tool returned no separate ID)	✅ No blocking/material findings
E	N/A	N/A	✅ Auto-skipped: no dependency/supply-chain surface changed
F	task(agent_type="general-purpose", name="sentinel-dim-f", model="claude-sonnet-4.6")	N/A (sync task tool returned no separate ID)	✅ No material findings after Phase 3 calibration
Persist	gh pr comment 1746 --body-file -; gh pr comment 1746 --edit-last --body-file -	#1746 (comment)	✅ Report persisted

Findings

• 🔴 CRITICAL: 0
• 🟡 IMPORTANT: 0 new / 0 known
• 🟢 MINOR: 0

Details (ordered by severity)

No findings. Dimension D's real-timer observation and Dimension F's documentation/i18n observations were calibrated below the materiality floor: they either follow an existing test convention, lack a concrete trigger→mechanism→consequence risk path, or match existing hardcoded-string practice.

Follow-ups & Actions

• APPROVED → MERGE: no new Sentinel follow-up issues required by this report.

Decision rationale

• Reviewed SHA is exactly c2e06683a049eb4bcadd58f116a1c974f0122524.
• TDD choreography is compliant: RED test commit precedes GREEN feat commit, with tests/source separated by commit.
• RED tests fail without implementation and the reviewed HEAD passes the targeted TUI suite, lint, and typecheck.
• Correctness checks found the intro/model state machine preserves the model-confirm flow, in-flight guard, input capture, and isActive gating; intro Esc navigates home; listed action keys match screen bindings.
• Phase 2 dimension review found no blocking or material follow-up findings.