Skip to main content

Output styles

Every command routes its output through a single style, controllable via the --output-style flag:

bb pr list --output-style normal    # default, colored table
bb pr list --output-style json      # raw JSON for jq
bb pr list --output-style ai        # minimal plain text for LLMs

The three styles​

normal — for humans​

Formatted, colored tables via cli-table3 + chalk. The default.

┌────┬──────────────────┬──────────┬────────────────────┬───────┬────────────┐
│ ID │ Title            │ Author   │ Branches           │ State │ Created    │
├────┼──────────────────┼──────────┼────────────────────┼───────┼────────────┤
│ 42 │ Add login page   │ Jane Doe │ feature/login → main│ open  │ 2026-03-20 │
└────┴──────────────────┴──────────┴────────────────────┴───────┴────────────┘

json — for scripts​

Raw JSON to stdout. Every command emits whatever the command's domain type serializes to, with no wrapping envelope:

$ bb pr list --output-style json | jq '.[] | .title'
"Add login page"
"Refactor auth"

ai — for LLMs and awk/cut​

Minimal, token-efficient plain text. No box borders, no padding, no color codes — every character carries meaning. Tab-separated, one row per line.

$ bb pr list --output-style ai
42	Add login page
37	Refactor auth

Pipe it straight into an LLM:

bb pr view 42 --output-style ai | llm "summarise this PR"

Or parse it with shell tools:

bb branch list --output-style ai | awk '{print $1}'   # branch names

Set a persistent default​

If you always want JSON or AI mode, save it:

bb option --output-style ai

This writes output_style: "ai" into ~/.bb-cli-config.json. Every subsequent command defaults to that style unless you pass --output-style explicitly.

Check the current setting:

bb option show

Resolution order​

The active style for any command invocation is resolved in this order (first match wins):

  1. --output-style flag — per-invocation override
  2. output_style field in ~/.bb-cli-config.json — saved default
  3. normal — built-in fallback

Why ai exists​

Feeding bb pr view 42 --output-style ai into an LLM takes roughly 5× fewer tokens than the human-readable normal output (no box-drawing, no padding, compact key=value pairs). If you use LLMs in your dev loop, this matters.