Make Harness run lifecycle resilient

[actus7]

Summary

• reconcile orphaned root runs after an expired worker lease before accepting a new turn
• heartbeat active workers, reconnect interrupted streams, replay durable output, and cancel runs server-side
• prevent stale UI generations from racing with newer sends and show actionable structured conflicts
• contain Mermaid parse failures inside the Canvas error panel instead of leaking raw SVG errors into the page

Root cause

The chat treated a prematurely closed Harness stream with status running as a successful completion. The database run remained active, so the next turn received Conversation already has active run. Stop/navigation only aborted the browser request and could leave the server worker orphaned.

Validation

• pnpm exec vitest run — 477 passed, 1 skipped
• pnpm typecheck
• pnpm lint — 0 errors (one pre-existing warning under .remember/tmp)
• pnpm build
• local browser test against the affected conversation: new run completed, immediate stop produced database status cancelled with no remaining lease, no console errors
• invalid Mermaid Canvas: scoped fallback remained visible and no body-level parser output leaked into the layout

[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 20, 2026 2:59pm