Streaming transcript & tool cards

The transcript layer folds the session event stream into items and renders them. This page describes what you see; the event-to-item rules are identical for live streams and snapshot replay.

Message items

• User messages — ❯ gutter bubbles (roleUser color); images render inline where the terminal supports it (12-line cap), otherwise as [image] placeholders.
• Assistant messages — rendered as Markdown while streaming (headings, links, inline code, code blocks, quotes, and lists each have dedicated tokens), separated by blank lines, first line bulleted ●, continuation indented two columns.
• Thinking blocks — reasoning streams as its own block above the body: a braille spinner + thinking... label with a rolling tail window while live; after close, folded to a two-line italic preview with ... (N more lines, ctrl+o to expand).

Tool cards

Tool calls render as the generic card by default: a status dot — ○ running (primary) or ● settled (success/error colored) — plus an indented ⎿ one-line summary; Ctrl-O toggles between summary and full output.

Two dedicated cards register through render intents and take over when a tool's presentation view declares diff or terminal:

• Diff card — per-file unified diff with LCS line coloring (added/removed/strong/gutter tokens); 12 lines per file collapsed, 200 expanded.
• Terminal card — $ command header (shellMode color) + cwd + exit badge (error/warning colored); output rows 10 collapsed, 120 expanded; a finished run with no output shows (no output).

Unknown tools, or tools without presenters, always fall back to the generic card — intent resolution never throws.

Step folding and the long-session window

• step-summary — within a turn, when the next step starts, the previous step's tool items fold into a single … step N · Tool ×M line; the turn's last step stays expanded (a visible tail of tool cards per turn).
• Ctrl-O scope — applies to the most recent 3 turns of tool cards and thinking blocks.
• Sliding window — only the latest 15 completed turns stay mounted; older turns are silently evicted from the render tree (event data stays complete in the session record). A 200-turn session keeps ~90 mounted components — scrolling stays smooth.

Special cases

• todo_write calls and results never appear in the transcript — the todo pane is their only surface (see Bottom panes).
• Synthetic injected messages (workspace instructions, context snapshots) render as nothing — the content still reaches the model, the stream stays clean (see the FAQ).