Fix conversation runtime regressions

[actus7]

What changed

• normalize conversation titles at the API boundary so stale clients cannot fail long prompts with Invalid conversation payload
• keep Harness runs active until every explicitly requested iterative section is present, while accepting qualified headings such as Entrega Final
• avoid retrying DuckAI when the configured local Chrome runtime is deterministically unavailable
• load the persisted accent through an external beforeInteractive script to remove React/Next development console errors
• add regression coverage for long titles, iterative completion, and DuckAI fallback behavior

Root cause

The local client used the full first prompt as a conversation title while the conversations API rejected titles over 200 characters. In addition, the Harness trusted a provider stop finish reason even when the requested multi-round format was incomplete, and an exact heading comparison could keep a completed run alive when the model wrote Entrega Final.

Impact

Long prompts now create conversations successfully even from stale bundles. Explicit multi-round prompts continue until their requested sections are delivered, deterministic provider failures fall through faster in Auto mode, and the chat loads without the accent hydration/console regression.

Validation

• exact reported prompt exercised in authenticated ModelHub Auto + Harness mode on localhost
• conversation creation returned 201 and Harness continuation calls returned 200
• all three rounds and the final FapiBird artifact were rendered without Invalid conversation payload
• pnpm test — 467 passed, 1 skipped
• pnpm typecheck — passed
• pnpm lint — 0 errors (1 unrelated pre-existing warning under .remember/tmp)
• pnpm build — passed

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
modelhub	Ready	Preview	Aug 19, 2026 8:17pm