3.3 KiB
id, title, status, assignee, created_date, updated_date, labels, dependencies, priority, ordinal
| id | title | status | assignee | created_date | updated_date | labels | dependencies | priority | ordinal |
|---|---|---|---|---|---|---|---|---|---|
| TASK-10 | Modernize savings chat agent UI with AI Elements phases | In Progress | 2026-06-16 08:38 | 2026-06-16 08:57 | high | 10000 |
Description
Implement the planned AI Elements-inspired savings chat agent UI in three sequential phases: conversation/message/prompt/tool trace primitives, reasoning/work-progress disclosure, and sources/citation support.
Acceptance Criteria
- #1 Phase 1 replaces the basic chat surface with reusable conversation, message, prompt input, and tool trace UI primitives
- #2 Phase 2 adds a safe reasoning/work-progress disclosure derived from existing tool traces, without exposing hidden chain-of-thought
- #3 Phase 3 adds structured source/citation support through stored assistant metadata and visible UI affordances
- #4 Each phase is covered by failing-first tests, verified after implementation, and committed separately
Implementation Plan
- Phase 1 TDD: add tested chat UI view-model helpers, reusable conversation/message/prompt/tool trace components, and integrate them into SavingsChatPage.
- Commit Phase 1.
- Phase 2 TDD: derive safe reasoning/work-progress summaries from tool traces and render them as collapsible disclosure.
- Commit Phase 2.
- Phase 3 TDD: extend chat message metadata with structured sources/citations, persist it through Convex history, and render sources/inline citations.
- Commit Phase 3.
- Run focused tests, lint, build, and record verification notes without closing the task until user confirmation.
Implementation Notes
Phase 1 complete: added AgentChat primitives for conversation, message rendering, prompt input, and tool trace disclosure; integrated SavingsChatPage. Verification: npx vitest src/components/chat/AgentChat.test.tsx --run, npx eslint targeted chat/page files, npm run build (Vite chunk-size warning only).
Phase 2 complete locally: added safe work-progress/reasoning disclosure derived from toolTrace result summaries, removed raw inputSummary display from the disclosure, and added active progress state while a response is pending. Verification: npx vitest src/components/chat/AgentChat.test.tsx --run (9 tests), targeted eslint, npm run build (Vite chunk-size warning only). Spec subagent review approved.
Phase 3 complete locally: added optional sources/citations metadata to chat messages, preserved it through import/list/append history flows, generated private finance sources from executed tool traces in savings chat responses, and rendered source lists plus inline citation markers in AgentChat. Verification: npx vitest src/components/chat/AgentChat.test.tsx convex/savingsChat.test.ts --run (37 tests), targeted eslint, npm run build (Vite chunk-size warning only).
Phase 3 review follow-up: addressed spec blocker by generating citation metadata from tool-derived sources and appending visible citation markers to live ask/sendMessage answers. Re-verified: npx vitest src/components/chat/AgentChat.test.tsx convex/savingsChat.test.ts --run (37 tests), targeted eslint, npm run build. Spec re-review approved.