What the bar shows
Read the AutoMode status bar in Claude Code: effort and model per prompt, context gauge, prompt cache countdown, the route reason, and theme options.
One quiet line above the prompt, in your theme's colors. It reports what actually ran.
| On the bar | Means |
|---|---|
| High · Opus | The effort and the model for this prompt. Deciding… while the judge thinks. |
| ◔ 38% | How full the chat's context is. |
| cache 58:12 | Time until the prompt cache goes cold. After that, the next message pays full price to re-read the chat. |
| saved ~$0.38 | The estimated saving against the model the session started on. Right after a switch it reads ↺ $0.07 to pay back. |
| a refactor across modules | Why the route was chosen. |
| Auto | Switches routing on or off for this session. |
| ⚙ | Settings. |
Change how it looks
Run /plugin configure automode@automode:
| Setting | Options | Default |
|---|---|---|
| Layout | expanded: the bar above the prompt. minimal: one short line. | expanded |
| Theme | native follows your Claude Code and editor theme. automode: blue routing, violet effort, teal guardian. mono: neutral accents. contrast: stronger text and edges. | native |
| Appearance | auto follows your Claude Code theme. dark and light fix it. | auto |
| Bar metrics | Any of pending, context, cache, cost, saved, guardian, stage. The mode, model and effort always show. | all seven |
Live savings
The bar shows what AutoMode saved in this session once a prompt runs on another model than the session's own. It is honest about a switch: a new model must first write the whole chat into its own prompt cache, so the bar reads ↺ $0.07 to pay back until cheaper requests pay that off. Then it reads saved ~$0.38.
Price the real tokens twice
Each main-session request is priced as it ran, cache writes included. Then the same tokens are priced on the session's model, the one your /model or Claude Code's default picked.
Keep the baseline's cache warm
The baseline keeps its own prompt cache while requests come within the cache lifetime. So it reads what it already cached and writes only what is new, as it really would.
Subtract AutoMode's own calls
The judges, extracts and compaction calls count against the saving.
It is an estimate, so it shows a ~. A different model would write different replies and might need a different number of turns, and nobody can measure that inside one session. Effort savings on the same model and subagents are left out. On the five benchmark sessions, it estimated a 56% saving against Opus, where the measured Opus runs showed 52%, within $0.18 per session. /automode details shows every part of the sum.
Without the bar
AUTOMODE_UI=offhides the bar. Routing goes on./automodeprints the same facts as text./automode detailsadds the workflow, the guardian and usage.- The VS Code extension does not draw the band above the prompt. Use
/automodethere. See Where AutoMode runs.
Last updated on
How a route is decided
How AutoMode picks a model and effort for each Claude Code prompt: two Haiku judges, quality floors for risky work, and a cache check before any switch.
The workflow
AutoMode's plan, build, verify and review workflow for open-ended Claude Code work: risk bands S0 to S3, stage models, the S3 rule, and how to start it.