Claude Code uses tokens to process the context it receives, so keeping irrelevant history, instructions, and tool output out of that context can help manage usage. Anthropic puts it plainly: “Token costs scale with context size: the more context Claude processes, the more tokens you use.” The practical goal is not a promised percentage reduction; it is to give Claude the context needed for the task without carrying along material that is no longer useful.
1. Clear context when you switch to unrelated work
When a session moves from one task to an unrelated one, use /clear to avoid carrying stale conversation history into the new task. Anthropic says old context can waste tokens on later messages. Check current usage with /usage; Anthropic’s documentation describes it as a way to inspect usage and context/cache behavior, with usage attribution available on supported plans and versions. See Anthropic’s cost-management guidance and the CLI reference.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
As an Amazon Associate I earn from qualifying purchases.
Do not clear automatically just because a conversation is long. If the work is continuing and earlier decisions or investigation matter, clearing discards potentially useful context. If you may return to the old session, rename it first and resume it later; use /clear for the new, unrelated task.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →2. Compact a continuing task with a deliberate summary
For related work that needs to continue, use /compact rather than treating the conversation as a fresh start. Give it guidance about what to retain, such as /compact Focus on code samples and API usage. Anthropic also documents project-level compact instructions in CLAUDE.md. See Manage costs effectively.
#1 Best Overall
Compaction summarizes; it does not guarantee that every detail will survive. Name the state that the next step depends on—such as relevant code, decisions, test results, or API details—and inspect the result before relying on it for a detail-sensitive task.
3. Keep startup instructions short and load specialized guidance only when needed
Claude Code reads CLAUDE.md files at session start. Keep these files focused on durable project information, such as conventions and build commands, rather than placing every procedure and reference in the context of every session. Anthropic recommends scoping repository-specific guidance and using skills for repeatable procedures. Its documentation explains how project memory and path-scoped rules work, and how to extend Claude with skills.
Rank #2
Put guidance where it belongs
CLAUDE.md: concise facts and conventions that are broadly useful across project work.- Path-scoped rules: guidance for particular files or directories, stored under
.claude/rules/with path patterns so matching rules apply when Claude works on those files. - Skills: reusable procedures or domain knowledge that need not be part of every task’s starting context. Skill bodies load when invoked or when Claude determines they are relevant.
This division keeps always-loaded instructions lean without losing access to specialized guidance when a task calls for it.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match4. Filter noisy output and delegate only when it helps
Large logs, command output, and research results can crowd a session with material Claude does not need. Anthropic recommends hooks that preprocess noisy input before it reaches Claude; its cost guidance illustrates filtering a 10,000-line log to show only errors. That is an example of filtering, not a measured claim about tokens saved. A relevant skill can also provide domain knowledge that would otherwise require repeated exploration. See Anthropic’s cost-management guidance.
Use subagents as context isolation, not free work
A subagent can handle focused research or exploration in its own context and return a concise summary instead of filling the main conversation with raw results. But its requests still count toward usage limits, so delegation does not guarantee lower total token use. Keep the assignment narrow and ask for only the findings the main task needs. Anthropic documents custom subagents and notes that focused subagents can use a faster, cheaper model such as Haiku where appropriate.
Other controls that can help manage usage
Match the model to the work
Anthropic recommends Sonnet for most coding tasks and reserving Opus for complex architectural decisions or multi-step reasoning; simple subagent tasks may suit Haiku. Model availability and relative costs can change, so check the current options in Anthropic’s cost documentation rather than assuming a fixed price or lineup.
Check context and trim inactive MCP tools
Use /context to inspect context use. Anthropic says MCP tool definitions are deferred by default and recommends preferring CLI tools when they can do the job, as well as disabling MCP servers that are not active for the task. These choices can limit unnecessary context or tool overhead; they do not establish a fixed amount of savings. Details are in Manage costs effectively.
Choose the adjustment that fits the task
| Approach | Best fit | What it changes | Setup effort |
|---|---|---|---|
/clear |
Starting unrelated work | Removes the prior session’s context from the new task. | Quick session command. |
/compact with instructions |
Continuing related work | Summarizes history around the information the next step needs. | Quick session command; compact instructions can also be set in CLAUDE.md. |
| Path-scoped rules and skills | Project guidance that applies only to certain files or procedures | Moves specialized instructions out of the always-loaded project guidance. | Requires configuration. |
| Hooks and focused subagents | Noisy input or side exploration | Can keep irrelevant output out of the main conversation; subagent requests still count toward usage limits. | Requires configuration and a focused task. |
| Model and MCP choices | Matching capability and tools to the task | Changes which model or tool definitions the workflow uses. | Review current options and configuration. |
These practices manage context and workflow, not a guaranteed token-saving rate. Their value depends on the task: clearing helps when work is unrelated, while compaction and scoped instructions preserve continuity when it matters.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
