Breaking

Claude Code Cost Controls Draw Developer Attention Around Advisor, Skill-Doctor and Usage Commands

October 4, 2026 at 06:15 EDT

Software Dev & Coding
AI Agents

Developer attention is shifting toward three existing Claude Code controls that attack token spend from different angles: route difficult decisions to a stronger advisor, remove unused skill context, and inspect where usage is accumulating. The commands are /advisor opus, /skill-doctor, and /usage, all documented in Anthropic's Claude Code materials.

Oct. 2026 - Claude Code

Three commands turn token spend into a workflow

Use a stronger advisor only at hard decisions, audit unused skill context, and inspect where each session's usage is going.

2.7 pp

SWE-bench lift with Sonnet plus Opus advisor

-11.9%

Cost per agentic task in the same evaluation

400-700

Typical advisor text tokens for a short plan

$4/$20

Opus 5.5 input and output per million tokens

BrowseComp: Haiku alone versus Haiku with Opus advisor

19.7%

Haiku alone

41.2%

Haiku + Opus advisor

The advisor pattern concentrates stronger reasoning at decision points instead of running the larger model for the whole task.

Cost-control loop

/advisor opus

Escalate hard decisions

→

/skill-doctor

Find unused context

→

/usage

Check attribution

Why developers like it

The commands make cost visible where work happens: routing models, trimming unused skills, and attributing session usage.

What still needs judgment

Advisor calls can add tokens, reports are estimates, and pruning skills still requires a user decision.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.