fix(ui): manually created tasks render as "Unnamed task" in the sidebar

[jromualdez-scale]

Summary

Every task created by typing into the prompt box renders as "Unnamed task" in the sidebar. The message and reply are fine — only the label is wrong. Scheduled tasks are unaffected.

#377 made the list-tasks response a lean TaskSummary that omits params (correctly — params is the arbitrary caller-supplied create-time payload and can carry sensitive material, so it must not be exposed in bulk). The sidebar's label resolver was repointed accordingly, from params.description to task_metadata.display_name → task.name. But the writer was never updated to match: the prompt box still created tasks with only params.description, which the sidebar can no longer see, so the label fell all the way through to the default.

Fix

Frontend-only — the backend already accepts and persists task_metadata on task/create and returns it in both the create and list responses:

• prompt-input.tsx now writes the label the resolver reads first: task_metadata: { display_name: deriveTaskDisplayName(prompt) }.
• deriveTaskDisplayName (in task-utils.ts, next to the reader) trims, collapses whitespace/newlines, and truncates to 80 chars.
• useCreateTask now plumbs task_metadata through to the RPC body; previously anything besides agentName/params was silently dropped.

The label renders immediately on create (the optimistic cache insert uses the create response, which carries task_metadata) and correctly after reload (the list response includes task_metadata).

Deliberately not used: task.name. It looks like the obvious fix given the resolver's fallback, but name is a globally-unique get-or-create idempotency key: task/create with an existing name returns that task (and can overwrite its params). Writing prompts into it would silently merge unrelated conversations — a data-correctness bug worse than the cosmetic one being fixed.

Trade-off, stated explicitly: this intentionally publishes a normalized 80-char prefix of the prompt into task_metadata, which the broadly-scoped list response includes. That is strictly less exposure than pre-#377 behavior (full params, including the full prompt, in every list item), and it is the same mechanism the scheduler already uses to title tasks — but display_name should be treated as a public-ish display field, and nothing sensitive should ever be auto-derived into it beyond this.

Tests

• task-utils.test.ts: unit coverage for deriveTaskDisplayName, plus a writer/reader contract test — a task shaped like the prompt-box writer's output must not resolve to "Unnamed task". The absence of exactly this test is how the writer and reader drifted apart unnoticed.
• use-create-task.test.tsx (new): asserts the mutation forwards task_metadata in the task/create request body, and sends null when omitted.

Not addressed here

• Tasks created after fix(tasks): omit params from the list-tasks response #377 with null task_metadata keep their default label (no backfill; cosmetic).
• The stock notebook template in the Python SDK repo sets name (an idempotency key) as its only human-visible identifier, which now surfaces in the sidebar as <uuid>-task; fixing that to also set display_name is a follow-up in that repo.

Verification

• npm run typecheck, npm run lint, npm test — all green (82 tests, 7 new).

🤖 Generated with Claude Code

Greptile Summary

The PR aligns manually created task labels with the sidebar’s metadata-based reader.

• Derives a normalized display name from the submitted prompt.
• Forwards task metadata through the task-creation hook.
• Adds coverage for metadata forwarding and task-name normalization.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
agentex-ui/components/primary-content/prompt-input.tsx	Adds a normalized prompt-derived display name to newly created task metadata.
agentex-ui/hooks/use-create-task.ts	Extends task creation to forward optional task metadata to the RPC.
agentex-ui/lib/task-utils.ts	Adds the display-name normalization and truncation helper.
agentex-ui/hooks/use-create-task.test.tsx	Verifies task metadata is forwarded and omitted metadata is sent as null.
agentex-ui/lib/task-utils.test.ts	Covers normalization, code-point truncation, and the writer/reader naming contract.

Sequence Diagram

sequenceDiagram
  participant User
  participant Input as PromptInput
  participant Hook as useCreateTask
  participant API as task/create
  participant Sidebar
  User->>Input: Submit prompt
  Input->>Input: Derive display_name
  Input->>Hook: params + task_metadata
  Hook->>API: Create task
  API-->>Hook: Task with task_metadata
  Hook-->>Sidebar: Seed task cache
  Sidebar->>Sidebar: Resolve display_name

Loading

_{Reviews (2): Last reviewed commit: "fix(ui): truncate display_name by code p..." | Re-trigger Greptile}

Context used:

• Knowledge Base — Frontend task experience
• Knowledge Base — Agentex shared frontend UI