[M4n5ter]

Summary

External agents already display their assistant replies in the conversation, but the session catalog only indexed native model output and user messages. The sidebar could stay unread after the external reply had been viewed because the conversation and the catalog disagreed on the visible tail.

Index visible external assistant text with the same message ID used by history and live delivery. Backfill that index from existing external-run facts on startup, preserving existing read receipts.

Verification

• cargo test --locked -p maka-agent-state passed (7 tests).
• cargo test --locked -p maka-cli --test external_jobs passed (5 tests). The Antigravity restart case verifies that the final reply becomes the catalog tail and acknowledging it clears unread status.
• The production Workbench bundle ran headlessly against an isolated copy of an existing Antigravity conversation. hasUnread started as true; opening the conversation and exposing the final reply produced an automatic read receipt. After switching to a different conversation, hasUnread remained false.
• cargo fmt --all -- --check, Biome for the ACP fixture, and git diff --check passed.

AI use

• Generative tooling made a substantive contribution

OpenAI Codex implemented the catalog and migration changes, updated the regression test, and ran local verification.

Checklist

• Tests cover the change
• Format and the affected suites pass locally

Does this PR entail a change in behavior?

• Yes, described under Summary above