Block-based command output

Each command and its output is rendered as a visual "block" — a left-edge accent bar in the terminal gutter plus a hover-only action toolbar.

Why blocks

When you scroll back through 200 lines of tail -f you don't want to find the command boundary by eye. Blocks make it obvious: every command starts a new shaded gutter region, every output flows underneath it.

Actions

Hover over any block to surface the toolbar (top-left of the block):

• copy — Copies the entire block text (command + output, ANSI-stripped) to

the system clipboard.

• ask AI — Pre-fills the AI sidebar composer with "Help me understand…"

followed by the command and output. One click to start a conversation about what just happened.

• rerun — Sends the command text back through the PTY (followed by CR).

Useful for retries after a fix.

The header also shows the command (truncated to 60 chars) and total duration.

How prompt detection works

The parser runs client-side after every PTY chunk. The default prompt regex is:

/^(?:[\w][\w.~/:_-]*[@:~][\w.~/:_-]*\s*)?[$#❯➜>]\s+\S?/u

It matches:

• A typical bash/zsh prompt: user@host:/path$ cmd
• A starship/oh-my-zsh prompt: ❯ cmd
• A custom prompt with @, :, or ~ in the prefix

Lines that don't match are appended to the open block as output.

Limitations

• Exit codes aren't shown — would require shell-side instrumentation (Phase 2)
• The "command" is extracted heuristically (text after the prompt sigil) — exotic

PS1s may need a custom regex (planned: user-overridable in settings)

• OSC 133 (semantic prompt escapes) support coming in Phase 2 — will replace the

regex for shells that emit it