Context Health

MacHatter1Token Usage & Limits76
Get it for macOS

See what contributes to your thread’s context usage. Search recorded text, inspect observed skill reads and browse available skills. Optional Codex session detail reads logs on the thread host; no separate account or paid service is required.

Overview

View source

See what contributes to context usage

Context Health gives each thread a side panel for exploring recorded messages, tool output and skills. Use the category bars to find large contributions, then search, sort and expand individual entries.

What you get

  • A context usage meter when the provider exposes usage through BB.
  • Recorded-text estimates grouped by category, with searchable previews.
  • A Loaded skills view for skill reads observed in tool results.
  • A separate Available skills inventory with descriptions and plugin ownership.
  • Optional Codex session-log detail for additional instructions, catalogue entries and tool definitions.

The panel refreshes every 30 seconds while visible. Open it from a thread’s right panel using New tab → Context Health.

Understand the coverage

The shared view uses BB’s normalized event data across Codex, Claude Code, Pi and ACP providers. Detail depends on what the provider exposes. Text estimates do not reconstruct the full model context or add up to the usage meter.

Loaded skills means a qualifying skill read was observed; it does not guarantee the skill remains in context after compaction. Available skills is inventory, not loading evidence. Coverage notices explain missing data and inspection limits.

Requirements and privacy

Requires BB 0.42 or later and a compatible Plugin SDK runtime of 0.4.48 or later. Optional Codex detail requires access to the matching session log on the thread host.

No separate account or paid service is required. The plugin does not modify session files or send context to an external service. Real previews can contain private project information; review them before sharing screenshots.

More from MacHatter1

Agent TVSee what every running agent is doing, and which one needs you, from a live wall at the foot of the sidebar. Each tile shows the tool call in flight, the files touched in the last minute, a 60-second activity trace, and how much of its context window the thread has used. Threads blocked on a question or an approval are marked and sorted to the top. Pop the wall out as a monitor you can drag anywhere in BB. The bb agent-tv status command gives terminals and agents the same feed, and nothing leaves your BB server.NewAnimationPlay technical diagrams as a sequence of beats, then export a looping HTML page. Open a .scene.json file to get a stage and a timeline: drag a step boundary to retime it, or click a part to quote its state into chat. Agents write the same JSON with ordinary file edits. GIF and MP4 export are not included; HTML is the shareable output. Requires BB 0.43 or later.NewRecapGenerates concise, display-only summaries for BB threads on demand or after they become idle.137FollowupsTurn a finished answer into your next message: pick one of its suggested follow-ups and get a full draft in the composer. When a thread finishes, a model reads its last answer and your last three messages, then shows two to five next steps above the message box, each with the reason it is worth doing. Click one, or press 1–9, and it is written out as a first-person message in your composer; nothing is sent for you, and anything you typed stays. A switch in each thread's header turns follow-ups on or off for that thread, and a footer shows how long each run took and how many tokens it used. Each suggestion and draft is a short run on a model you choose from your bb providers, in a hidden worker thread that uses the provider's least privileged permission mode and is deleted when it finishes.New

More in Token Usage & Limits 25

Understand or control token, context-window, and provider-quota use.

View all