test(agent): pin the conversation message contract - ConversationMessage

[dmitrii-f-t27]

Summary

Hold the one export of trios/agent-server/apps/agent/entrypoints/app/agent-command/ConversationMessage.tsx — ConversationMessage — against its contract (#1569), so the next change to that file has something to fail against.

The message is a render plan over the turn, and what is pinned is what it answers today:

• the live turn (streaming, a running tool in the batch): the user text in its bubble; the done thinking part behind the closed reasoning trigger with its Thought for a few seconds label; the task entry with the Working… (3 actions) title, the wrench trigger, and every tool row — status icon (lucide-loader-circle / lucide-circle-check / lucide-circle-x), label, · subject, and the measured duration (1.5s); the produced-files strip with the file name and size; and the assistant text draws nothing — the shared streamdown renderer's default stream mode mounts empty, the same measured reality the ChatMessages suite pins — with the bouncing-dots placeholder away while parts exist;
• the streaming empty turn: the bot avatar with exactly three bouncing dots, and no strip;
• the strip-only mount: the strip alone — the user text and the task stay away;
• the done turn: the singular Agent activity (1 action) title with the task rows collapsed away (the task's defaultOpen wiring).

No module is mocked here at all: the strip's turn-files fallback rides a real query client the way the outputs-rail suite mounts it, and every kit stays real — so this suite adds no global stand-ins for any other suite to race.

Verification

• bun test entrypoints/app/agent-command/ConversationMessage.test.ts — 4 pass, 0 fail.
• Broken-copy proof: swapping the task-title ternary (allDone → !allDone) turns exactly the first and fourth tests red (2 fail); the copy restores clean via git checkout.
• Mixed-order run with the whole entrypoints/app/agent-command/ group: 47 pass + the one pre-existing ConversationInput local-order failure that fails identically without this suite (the marker-stub readdir artifact, not touched by this change — this suite registers no mocks).
• biome ci on the suite file: clean. tsgo --noEmit: 0 mentions of the suite file.

Closes #1569

Dmitrii Fedorov (@dmitrii-f-t27)

[tri-claw]

Summary by t27-bees

This PR adds a contract test for the ConversationMessage component, pinning its current render output across four turn states (live, streaming-empty, strip-only, done) so future changes have a baseline to fail against. The test uses real dependencies (no mocks) and renders to static markup for string assertions.

Worth a look

• ConversationMessage.test.ts:27 — A single QueryClient instance is shared across all tests; React Query's cache persists between test cases, so later tests may see stale data from earlier ones (e.g., the strip's turn-files query could resolve from a prior test's cache).
• ConversationMessage.test.ts:30–36 — renderToStaticMarkup produces SSR HTML only; it does not run client-side effects, layout measurements, or the streamdown renderer's streaming logic, so assertions like "assistant text draws nothing" only verify the initial server render, not the mounted behavior.
• ConversationMessage.test.ts:13–14 — The dynamic import with ?conversation-message-suite-real relies on a bundler-specific query parameter to force a fresh module graph; this ties the test to Vite (or compatible) and may break in other environments or if the bundler configuration changes.
• ConversationMessage.test.ts:58–95 — Assertions check for exact icon class names (lucide-loader-circle, lucide-wrench, animate-bounce) and hardcoded label strings (Thought for a few seconds, Working… (3 actions)); these are implementation details of the UI kit and copy that will churn independently of the component's contract, making the test brittle.
• ConversationMessage.test.ts:78 — The animate-bounce count assertion (split('animate-bounce').length - 1) assumes the class appears exactly once per dot and nowhere else; any wrapper or utility adding the same class will cause a false failure.

---

_{Read: 1 files, +193 -0, the whole diff. Head 5c117c3. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}

[github-actions]

✅ Tests passed — 3880/3950

Suite	Passed	Failed	Skipped
✅ agent	423/423	0	0
✅ build	9/9	0	0
✅ cdp-protocol	5/5	0	0
✅ eval	93/93	0	0
✅ server-agent	280/280	0	0
✅ server-api	2244/2313	0	69
✅ server-browser	6/6	0	0
✅ server-integration	10/11	0	1
✅ server-lib	279/279	0	0
✅ server-pglive	166/166	0	0
✅ server-root	68/68	0	0
✅ server-sim	8/8	0	0
✅ server-skills	31/31	0	0
✅ server-tools	244/244	0	0
✅ shared	14/14	0	0

View workflow run