fix(tui): reconcile pending prompts with the live server

[sahiljagtap08]

Issue for this PR

Closes #36585

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

Pending questions and permissions live in a process local map on the server, so a managed service restart destroys them. The TUI only learns about pending prompts through live events, so after a restart it keeps rendering a form the replacement server has never heard of. Submitting fails with a not found error that was silently ignored, leaving a form that can never be answered.

Two changes:

1. Bootstrap now hydrates pending questions and permissions from the live server (GET /question and GET /permission) and replaces the event fed state. Prompts that survived appear, prompts the server no longer knows are dropped.
2. When a question reply or reject fails, the TUI rechecks the live pending list. If the request is gone from the server, the form is dismissed instead of rendering forever. If the recheck fails too (for example the server is briefly unreachable), the form stays, so a transient network error cannot eat a real question.

The durable fix for keeping prompts alive across restarts is tracked in #36347. This is the client side escape hatch so users are never stuck looking at an unanswerable form.

How did you verify your code works?

Added TUI tests: one mounts with pending prompts served by the mock and asserts they hydrate, then simulates a restart by emptying the server lists and re-running bootstrap and asserts the stale entries are dropped. Another covers dismissal of an event fed question. bun test test/cli/cmd/tui/ in packages/tui: 25 pass. Typecheck and oxlint are clean.

Screenshots / recordings

The change removes a stuck form after a service restart. No visual change in the normal flow.

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

Sound design — treating the server's pending-prompt list as the source of truth at bootstrap (with a comment explaining why: process-local state orphaned by restarts) plus a verify-then-dismiss path on settlement failure covers both halves of the stale-prompt problem. I verified the risky bits hold together: question.asked inserts via the same binary-search helper the new dismiss() uses (packages/tui/src/context/sync.tsx:411-444), so the id-sorted invariant required by search() survives mixing event-fed and hydrated entries; and reply() resolving as an {error} object rather than throwing matches this SDK's generated-client style, so settle() won't silently miss failures.

Two suggestions:

1. Same treatment for permissions. permission.reply/reject failing against an orphaned request leaves the identical unanswerable dialog; the hydration half now covers permissions (packages/tui/src/context/sync.tsx:~545-551) but there's no runtime dismiss-on-failure equivalent for permission.tsx. Either add it for symmetry or note why questions are special.

2. Inconsistent scoping between the two list calls. Bootstrap queries list({ workspace }) (sync.tsx:~546) while settle() queries list({ directory: props.directory }) (packages/tui/src/routes/session/question.tsx:~54). If a workspace-level list ever spans directories the TUI isn't showing, the membership check could disagree with what bootstrap hydrated. Worth aligning both on the same scope parameter (or documenting why they intentionally differ).

Test coverage is genuinely good here — hydrate/drop-on-rebootstrap and double-dismiss are exactly the cases worth pinning (packages/tui/test/cli/cmd/tui/sync-pending-prompts.test.tsx:171-218). Nice-to-have: one case asserting settle does not dismiss when the server still lists the request (transient reply failure).

[github-actions]

Automated PR Cleanup

Thank you for contributing to opencode.

Due to the high volume of PRs from users and AI agents, we periodically close older PRs using automated criteria so maintainers can focus review time on the most active and community-supported contributions.

This PR was closed because it matched the following cleanup criteria:

• The PR was created more than 1 month ago
• The PR had fewer than 2 positive reactions
• Positive reactions are counted as thumbs-up, heart, celebration, or rocket reactions on the PR

PRs created within the last month are not affected by this cleanup.

If you believe this PR was closed incorrectly, or if you are still actively working on it, please leave a comment explaining why it should be reopened. A maintainer can review and reopen it if appropriate.

Thanks again for taking the time to contribute.