appendContentBlock in src/lib/formatters/conversation.ts calls truncate(content, 600) when rendering the human-readable transcript, silently cutting off user and assistant message content at 600 characters. LLM responses routinely exceed this length, so sentry conversation view cannot show complete transcripts in its default output despite being advertised as a full-transcript viewer. (JSON output likely retains the full content, which limits impact.)

In extractTurns (src/lib/formatters/conversation.ts), tool spans are assigned to a turn only when t >= span["precise.start_ts"] of that turn's ai_client span. Any tool span whose precise.start_ts precedes the first ai_client span fails this predicate for every turn and is omitted from the rendered transcript with no warning. While uncommon (tools usually run after an LLM call), agent setup or pre-LLM tool invocations would be invisible in sentry conversation view.

The parameter is declared optional: true in the positional spec but typed as non-optional string in the function signature — TypeScript won't catch undefined access in future code. Compare with src/commands/org/view.ts which correctly uses orgSlug?: string.