[KKeygen]

The import picker usually labels Codex conversations with their first user message, or a raw thread ID, even when Codex has a saved title. The reader never populated header.title, and the importer never consulted the local title metadata.

Read the local title metadata once per catalog/import request, keyed by exact thread ID. The read-only SQLite name/title takes precedence over session_index.jsonl, followed by rollout metadata and the existing first-prompt/ID fallbacks. This also picks up subsequent Codex renames on the next catalog refresh. Preserve full saved titles while keeping the 80-character limit for prompt-derived labels.

The desktop Node runtime uses native SQLite; older Node can use the optional sqlite3 CLI. Unavailable/corrupt metadata retains the existing fallbacks. No Codex data is written, and re-import does not rename existing DSH sessions.

Validation:

• npm test: 64 passed, 0 failed, 0 skipped.
• Three new regression tests passed on both Node 20 (SQLite CLI) and the actual desktop host, Electron 44 / Node 24.18.1 (native SQLite).
• Coverage includes title precedence, malformed index lines, subsequent renames, older DB schemas, unavailable/corrupt metadata, full Unicode titles, catalog/import agreement, and unchanged existing DSH titles.
• git diff --check and npm pack --dry-run passed.

The separate compressed-rollout fix is PR #5; this title fix is independently based on main. Only synthetic title fixtures are included; no private title index or conversation data is committed.