feat(tui): compose auto-convenes the new panel

[pedrofuentes]

Summary

The auto-compose flow now finishes the job — after the user confirms the composed panel it navigates straight to the debate run with the topic prefilled, completing the type-a-topic→watch-a-debate wizard. Phase 10 TUI overhaul — PR-21.

Changes

PanelComposeScreen.tsx

• confirmPreview now navigates to /convene/<panelName>/run with state: { topic, panelName } instead of /panels/:name
• The typed topic is captured from topicRef.current before the async persist call so it is always available at navigation time
• Preview hint updated from y save · n/e edit · Esc cancel → y Save & convene · n/e edit · Esc cancel
• All other behavior (compose/preview/persist async, inFlight guard, n/e reset, Esc cancel, captured-input effect) is unchanged

panel-compose-screen.test.tsx

• Added ConveneRunProbe component that reads useParams + useLocation.state to expose both the panel name and prefilled topic
• Added /convene/:panel/run route to renderScreen helper
• Updated the navigation test to assert CONVENE_RUN safe-panel + topic=pricing strategy (was DETAIL safe-panel)
• Added "y Save & convene" hint assertion test

TDD verification

Commit	Type	Tests
fa70851	test(tui)	2 FAIL (navigation + hint)
1f2c7c6	feat(tui)	11/11 PASS; full TUI suite 865/865

[pedrofuentes]

Status: APPROVED

Sentinel Review Report

Ref: feature/tui-compose-wizard → main
Report ID: sentinel-1747-1f2c7c6
Reviewed SHA: 1f2c7c6
Sentinel ruleset: v1
Reviewed at: 2026-06-26T09:20:48-07:00
Mode: standard
Review depth: Tier 2 (full)
Required action: MERGE
Persisted artifact: #1747 (comment)

Phase 1 — TDD / Test Evidence

• Exact SHA bound: ✅ git rev-parse HEAD → 1f2c7c6d225c676f6c6d21b6e968d430c8c4f3db.
• Tests exist & meaningful: ✅ panel-compose-screen.test.tsx asserts the new route and state: expect(lastFrame()).toContain("CONVENE_RUN safe-panel"); and expect(lastFrame()).toContain("topic=pricing strategy");; hint test asserts expect(lastFrame()).toContain("y Save & convene");.
• Test-first history verified: ✅ git --no-pager log --oneline origin/main..HEAD shows fa70851 test(tui): add failing tests for compose→convene wizard before 1f2c7c6 feat(tui): compose auto-convenes the new panel.
• Commit separation verified: ✅ git show --name-status fa70851 changed only M packages/cli/tests/unit/tui/panel-compose-screen.test.tsx; git show --name-status 1f2c7c6 changed only M packages/cli/src/tui/screens/PanelComposeScreen.tsx.
• RED verified: ✅ running the new focused tests at fa70851 failed as expected: AssertionError: expected 'DETAIL safe-panel' to contain 'CONVENE_RUN safe-panel' and expected 'Auto-compose panel\nSafe Panel\nSafe …' to contain 'y Save & convene'.
• GREEN verified: ✅ TUI suite on reviewed SHA: Test Files 93 passed (93) and Tests 865 passed (865).
• Full suite green on SHA: ✅ npx --yes pnpm@10.30.1 exec vitest run --reporter=dot exited 0 with Test Files 356 passed | 1 skipped (357) and Tests 4483 passed | 1 skipped (4484).
• Coverage: ✅ full coverage summary: statements: 91.97%, branches: 82.16%, functions: 92.6%, lines: 93.06% (meets configured thresholds 90/75/90/90).

Phase 1.5 — Fast-path Evaluation

🔴 count: 0 | LOC: 9 non-test changed lines (≤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
Quick scan	task(agent_type="explore", name="sentinel-quick-scan")	N/A (sync task tool returned no ID)	✅ Quick scan 🔴 count: 0
A1	task(agent_type="general-purpose", name="sentinel-dim-a1")	N/A (sync task tool returned no ID)	✅ No findings
A2	task(agent_type="general-purpose", name="sentinel-dim-a2")	N/A (sync task tool returned no ID)	✅ No findings
B	task(agent_type="general-purpose", name="sentinel-dim-b")	N/A (sync task tool returned no ID)	✅ No findings
C	task(agent_type="general-purpose", name="sentinel-dim-c")	N/A (sync task tool returned no ID)	✅ No adopted findings
D	task(agent_type="general-purpose", name="sentinel-dim-d")	N/A (sync task tool returned no ID)	✅ No adopted findings
E	N/A	N/A	N/A (no dependency/supply-chain surface changed)
F	task(agent_type="general-purpose", name="sentinel-dim-f")	N/A (sync task tool returned no ID)	✅ No findings
Gates	npx --yes pnpm@10.30.1 exec vitest run tests/unit/tui; npx --yes pnpm@10.30.1 lint; npx --yes pnpm@10.30.1 typecheck	N/A	✅ TUI tests, lint, typecheck passed
Persistence	gh pr comment 1747 --body-file .sentinel-report-1747.md	PR comment	✅ persisted

Dispatch verification: A1, A2, B, C, D, and F were dispatched as distinct sub-agent tool calls in a single parallel invocation; E was skipped under the Dim E auto-skip rule because no dependency surface changed. This platform's sync task tool did not expose per-agent IDs, so each ID is recorded as N/A with that limitation.

Findings

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

Details (ordered by severity)

• None.

Follow-ups & Actions

• APPROVED → MERGE: no new Sentinel findings to file.

Decision rationale

• Reviewed HEAD exactly matches 1f2c7c6d225c676f6c6d21b6e968d430c8c4f3db and the worktree was clean after verification.
• TDD evidence is complete: failing test-only commit preceded the implementation-only feat commit, and the same focused tests pass on HEAD.
• Implementation captures the topic before async persist, uses encodeURIComponent(result.panelName), keeps the inFlight/Esc/reset/error paths intact, and renders untrusted terminal strings through existing toSingleLineDisplay sinks.
• /convene/:panel/run is registered in AppRouter, and DebateStreamScreen reads location.state.topic before starting streamDebate.
• Sub-agent notes about bypassing the convene prompt and raw topic trimming were not adopted as findings: direct auto-compose→run navigation and topicRef capture are the explicit user-approved PR purpose/spec, and no normal-use correctness, security, or operability failure was demonstrated.