How to Use the Claude Code /btw Command to Save Tokens and Ask Questions Mid-Task
The Claude Code /btw command lets you ask questions or add context without interrupting your main thread — cutting token costs by up to 50%.
Why Mid-Task Questions Are Costing You More Than You Think
Every Claude Code user hits this pattern eventually. You’re an hour into a complex session — deep in a refactor, tracing a subtle bug, or scaffolding a feature across multiple files. Claude has your full project context loaded. Things are moving. Then you need to ask something quick.
Maybe it’s a library compatibility question. Maybe you forgot to mention a deployment constraint. Maybe you need to confirm your understanding of a concept before proceeding. So you type it, Claude answers, and you move on.
But something happened in the background that you probably didn’t think about: that exchange just became part of your main conversation thread. The question, the answer, all of it — appended to your session history, adding tokens that will now be re-read on every subsequent request for the rest of the session.
Claude Code token costs don’t work the way most people intuitively expect. You’re not just paying for the words you type right now. You’re paying for the entire conversation context that Claude reads every time it processes a new message. As that context grows, every new exchange gets proportionally more expensive — because the base keeps expanding.
This is where the Claude Code /btw command comes in. It’s a mechanism for handling questions, clarifications, and context additions without those exchanges being added to your primary conversation thread. The result is a leaner context window, lower per-session token costs, and sessions that stay efficient even as they get long.
According to users tracking their Claude Code token spend across extended sessions, consistent use of the /btw command can reduce total token consumption by up to 50% in sessions with frequent mid-task questions. That’s not a marginal improvement — it’s material enough to affect how teams budget for AI-assisted development work.
This guide covers exactly how the /btw command works, how to use it correctly, when it helps most, and how to build it into a workflow that actually reduces your costs.
What the /btw Command Is and How It Works
The basic concept
/btw stands for “by the way.” The naming is deliberate and useful — it maps directly to the kind of message this command is designed for.
When you prefix a message with /btw in Claude Code, that exchange is handled in a separate context from your main conversation. Claude processes your message, generates a response, and the whole exchange completes — but it doesn’t get appended to your primary conversation history the way a normal message would.
This is a significant departure from how regular Claude Code messages work. Every normal message you send is recorded in your session history. Every response Claude generates gets added to that history. On the next message, Claude reads all of it. And the one after that. The context window just keeps accumulating.
A /btw message sidesteps that accumulation for the specific exchange. The conversation thread you’ve been building stays exactly as it was.
What “keeping it out of the main thread” actually means
It helps to understand how Claude Code’s context window works before appreciating why this matters.
In any conversation with a language model, the model doesn’t just see your current message — it sees the entire message history as a single long sequence. That’s how it maintains context and continuity. The full conversation, from your first message to your most recent one, gets sent as the input for every new response.
This means that if your conversation history is 20,000 tokens long, every new message you send costs at least 20,000 input tokens before you even count the tokens in your new message. A 100-token question on top of a 20,000-token history costs roughly the same as a 100-token question on top of a 5,000-token history — except the former is four times more expensive.
The /btw command creates a branch rather than an extension. Think of your main conversation as a document being written progressively. A /btw exchange is like a margin note — it’s readable, it informs the author, but it doesn’t add to the document’s word count. The next time someone reads the document, they don’t re-read the margin notes as part of the main text.
After a /btw exchange, your main conversation thread is exactly as long as it was before. Claude absorbed the context from the aside, but your running token count for the main thread didn’t change.
Does Claude remember /btw context?
Yes — within the session. This is an important clarification.
Context provided via /btw is available to Claude throughout the current session. If you tell Claude via /btw that you’re targeting a specific Python version, or that your infrastructure runs on AWS Lambda with memory constraints, that information will factor into Claude’s main-thread responses. It’s not isolated from the main task.
What it doesn’t do is pad your conversation history with the back-and-forth of that exchange. You get the informational benefit without the token cost of having that exchange permanently baked into your context.
This is also why /btw is different from just opening a second Claude conversation to ask your question. A second conversation loses all the context from your first one. /btw keeps your full task context intact while still handling the aside efficiently.
How /btw differs from other context management commands
Claude Code has several commands for managing session context:
/compact — Compresses your existing conversation history into a summarized version. It reduces the token count of what’s already in your main thread by replacing detailed history with a compact summary. Useful when your thread has already grown long.
/clear — Wipes your conversation history entirely. You start fresh with no prior context. Useful when you’re genuinely starting a new task and don’t need continuity.
/btw — Handles specific exchanges outside the main thread in real time. It’s preventive rather than corrective — it stops thread bloat from happening one exchange at a time.
These commands solve different problems. /btw is most effective when used consistently throughout a session to prevent the thread from growing unnecessarily. /compact is the right tool when the thread has already grown and needs to be trimmed. Neither replaces the other.
How to Use /btw in Claude Code: A Practical Guide
Basic syntax
The command format is simple:
/btw [your message]
That’s it. Everything after /btw is your message. It can be a question, a context addition, a preference, or a quick confirmation request.
Examples:
/btw does Python's asyncio.gather() catch exceptions from individual tasks or does one failure cancel all?
/btw we're targeting Python 3.11 minimum for this codebase — just wanted to flag that
/btw quick sanity check: in REST semantics, PUT replaces the whole resource and PATCH does a partial update, right?
Each of these gets processed, answered, and handled without becoming part of your main conversation history.
Prerequisites
To use /btw, you need:
- Claude Code installed and authenticated (follow Anthropic’s Claude Code setup guide if you haven’t already)
- An active Claude Code session in a project directory
- A running conversation (you can use
/btwfrom the very start of a session, or introduce it mid-way)
Step-by-step walkthrough: /btw in an active coding session
Here’s a realistic example of how /btw fits into a typical session:
Step 1: Start your primary task
Open Claude Code in your project directory and send your main task prompt:
I need you to help me refactor the authentication module in /src/auth/. We're transitioning from JWT tokens to session-based auth, and we need to maintain backward compatibility for 90 days during the transition period. Start by reviewing the current implementation and outlining an approach.
Claude reads the relevant files, understands the task, and begins its analysis. You now have a meaningful conversation thread started.
Step 2: Ask a quick question using /btw
While Claude is working through the refactor, you realize you need to confirm something about a library you’re planning to use:
/btw does Flask-Session support Redis clusters, or only single-node Redis?
Claude answers your question. Your main auth refactor thread is unchanged — it still shows only your initial prompt and Claude’s response to that.
Step 3: Add context you forgot to mention
You realize there’s a constraint you should have mentioned upfront:
/btw also, we have a load balancer in front of our app servers, so session storage needs to be external — in-memory sessions won't work across instances
Claude takes this on board for the session. Your main thread still reflects only your original prompt and response. The context is absorbed but the thread isn’t padded.
Step 4: Continue with the main task
Your next main-thread message continues the primary task:
That approach looks good. Can you now generate the updated session management class based on the outline you just gave me?
Claude has full context from both the main thread and your /btw asides. It produces the updated class incorporating the Redis cluster compatibility and load balancer constraint without you having to repeat them.
Step 5: Use /btw for mid-implementation questions
As Claude generates code:
/btw should we use redis-py or aioredis for async Redis operations here?
Another quick aside handled without entering the thread. You get the answer you need, Claude continues, and the token count of your main conversation stays lean.
What types of messages work best as /btw
The /btw command performs best for:
- Single-question factual lookups: “What’s the max key size for Redis?” “Does Docker support multi-stage builds in Compose?” “Is JWT considered stateless?”
- Context injections: Constraints, preferences, team conventions, infrastructure details — information Claude should have that you forgot to include initially
- Quick confirmations: Checking your own understanding before proceeding, validating assumptions
- Style and behavior preferences: How you want output formatted for the session, level of verbosity, code style preferences
- Terminology clarifications: Making sure you and Claude are using terms consistently
The Token Savings: Understanding the Math
Why token costs in Claude Code compound over time
To understand why /btw saves tokens, you need to understand how context works in practice.
When Claude Code processes a message, it reads the entire conversation history as a single input. Every prior exchange — your messages and Claude’s responses — gets re-read. This isn’t a bug in the system; it’s fundamental to how transformer-based language models maintain context.
The practical implication: your token costs for any given message are not just the tokens in that message. They’re the tokens in that message plus every token in every prior exchange in the session.
Here’s a simplified numeric illustration:
Session A: No /btw usage
| Exchange | New tokens | Cumulative context read |
|---|---|---|
| Initial task prompt | 500 | 500 |
| Claude’s response | 2,000 | 2,500 |
| Your question 1 | 150 | 2,650 |
| Claude’s answer 1 | 400 | 3,050 |
| Your question 2 | 120 | 3,170 |
| Claude’s answer 2 | 350 | 3,520 |
| Your question 3 | 80 | 3,600 |
| Claude’s answer 3 | 200 | 3,800 |
| Main task continuation | 100 | 3,900 |
| Claude’s main response | 3,000 | 6,900 |
Total input tokens across the session: roughly 27,270 (every exchange reads the cumulative context up to that point)
Session B: Questions 1–3 as /btw messages
| Exchange | New tokens | Context read for main thread |
|---|---|---|
| Initial task prompt | 500 | 500 |
| Claude’s response | 2,000 | 2,500 |
| /btw question 1 | 150 | Separate — doesn’t extend main thread |
| /btw question 2 | 120 | Separate |
| /btw question 3 | 80 | Separate |
| Main task continuation | 100 | 2,600 (main thread only) |
| Claude’s main response | 3,000 | 5,600 |
Total input tokens for main thread: roughly 11,200. The /btw exchanges add their own small costs, but those are processed against smaller, scoped contexts.
The difference compounds throughout longer sessions. By exchange 20 or 30 in a long session, the gap between a thread-bloated context and a lean one can be substantial.
Where the 50% figure comes from
The “up to 50% savings” estimate applies to specific session profiles — long sessions with frequent mid-task questions where those questions would otherwise be adding significantly to the running context.
The savings potential scales with three main factors:
Session length: Longer sessions have more room for the compounding effect to play out. A 10-exchange session won’t see as dramatic a difference as a 40-exchange session.
Proportion of asides: If roughly 40-50% of your exchanges are questions and context additions that don’t need to be in the main thread, and you route those through /btw, the savings are meaningful. If only 10% of your exchanges are asides, the impact will be smaller.
Model cost tier: The absolute token savings are the same regardless of model, but more expensive models (Sonnet, Opus) mean larger absolute dollar savings per session.
For teams running Claude Code at scale — multiple developers running multi-hour sessions regularly — even a 25-30% reduction in per-session token costs represents a meaningful reduction in total AI spend over a month.
Estimating your own potential savings
A useful exercise: look at your last three or four Claude Code sessions and audit the exchanges.
For each exchange, ask: was this advancing the main task, or was it a question or context addition that I could have handled as an aside?
If more than 20-25% of your exchanges fall into the “aside” category, you have real savings potential with /btw. Start using it systematically for a week and compare your token counts.
Claude Code’s /cost command shows you running token usage in a session, which makes it easy to track the difference between sessions where you actively use /btw and sessions where you don’t.
When to Use /btw and When to Use a Regular Message
The core decision rule
The clearest way to decide: does this exchange need to be part of your task’s permanent record?
If the answer is no — if this is a quick question, a tangential lookup, or context that’s useful right now but not central to the task — use /btw.
If the answer is yes — if this exchange represents a decision, a pivot, or something you’ll want Claude to reference later in the main task — use a regular message.
Strong use cases for /btw
Factual lookups you’d normally Google
When you’re mid-task and need a quick fact:
/btw what's the default timeout for fetch() requests in modern browsers?
/btw does TypeScript's Record<K, V> utility type preserve optional properties?
/btw is bcrypt's salt round count stored in the hash output, or do we need to store it separately?
These are questions that inform your work but don’t advance it. They don’t need to be in your conversation history.
Constraint additions
When you remember a project constraint you forgot to mention:
/btw important: we have a 10MB bundle size budget, so keep dependencies minimal
/btw the app needs to work offline — we're targeting a PWA with service worker caching
/btw the client has strict GDPR requirements — no data should be stored on servers without explicit consent
Claude needs to know these things. They don’t need to be documented as main-thread exchanges.
Code style and output preferences
When you want to calibrate Claude’s behavior for the session:
/btw for this session, please keep code comments minimal — the codebase uses self-documenting code style
/btw show me complete files rather than diffs — I find it easier to review
/btw when you give me options, keep it to two choices maximum and make a clear recommendation
These preferences apply session-wide without being part of the task thread.
Quick confirmations and sanity checks
When you want to verify your mental model before proceeding:
/btw just confirming: when we say "idempotent" in API design, we mean the same request can be made multiple times with the same result, right?
/btw a Docker volume is different from a bind mount — volumes are managed by Docker, bind mounts are tied to the host filesystem path. Did I get that right?
Confirmation that you understand something doesn’t need to be a main-thread record.
When to use regular messages
When the answer will change your direction
If asking the question might cause you to take a materially different approach to the main task, that exchange should be in the main thread. Claude needs to be able to see that decision-making conversation when it continues the task.
For example: “Should we use a message queue or direct API calls for this integration?” — this is a consequential decision. The reasoning and the answer should be visible in the main thread.
When the exchange is part of the debugging process
Debugging conversations are core task work. Keeping iterative debugging steps in the main thread gives Claude the right visibility into what’s been tried, what the results were, and why certain approaches were ruled out.
When you need the exchange as a reference point
If you know you’ll want to say “as we discussed earlier” or need Claude to trace back to a specific decision, keep it in the main thread. /btw context is available for the session but isn’t positioned as a reference point in the same way.
When the response will directly feed the next step
If Claude’s answer is the immediate input to the next task step, it should be a main-thread exchange so the flow is clear and traceable.
Combining /btw with Other Claude Code Features
/btw and CLAUDE.md
CLAUDE.md is Claude Code’s persistent context mechanism — a markdown file in your project root that Claude reads at the start of every session. It’s the right home for information that should always be available without repetition.
/btw and CLAUDE.md work as complementary layers:
- CLAUDE.md: Project-level constants — team conventions, infrastructure facts, coding standards, architectural decisions. Things that apply to every session.
/btw: Session-specific context — things that are relevant today but not necessarily project-wide constants.
A practical pattern: use /btw during active sessions to inject context. When a piece of that context proves repeatedly useful, migrate it to CLAUDE.md so you don’t have to inject it next time.
For example, if you find yourself regularly using /btw to remind Claude that your app needs to work with a specific legacy API, that’s information that belongs in CLAUDE.md.
/btw and /compact
These two commands work well together as a two-layer efficiency system:
- Use
/btwthroughout the session to keep asides out of your main thread — this slows the rate at which your main thread grows - When your main thread still grows lengthy over a long session, use
/compactto compress what’s already there - Continue working with a compressed main thread and ongoing
/btwfor new asides
This combination keeps your context window manageable regardless of session length. /btw prevents unnecessary growth. /compact handles the growth that does happen.
/btw and the /cost command
The /cost command in Claude Code shows your running token usage for the current session. Use it in combination with /btw to track the impact of your context management practices.
A useful habit: at the start of a new project, run two similar sessions — one with systematic /btw usage and one without — and compare the /cost output at the end. This gives you real data on the savings from your specific workflow rather than relying on general estimates.
/btw in multi-step workflows
If you’re running Claude Code as part of a larger automated pipeline — where one agent’s output feeds another agent’s input — /btw can be used to inject operational context between pipeline steps without that context bloating each step’s working context.
For instance, routing instructions, status metadata, or step-level constraints can be passed via /btw without becoming part of the accumulated conversation history that gets handed forward.
Advanced /btw Patterns
Batching multiple asides into one /btw message
Instead of sending several /btw messages separately, consolidate them:
/btw few things before we continue: (1) PostgreSQL 14 is our database, (2) the app is containerized with Docker Compose, (3) we have a 512MB memory limit per container, (4) we're using Black for Python formatting with default settings, (5) all new code needs type hints
One processed exchange for five pieces of useful context. More efficient than five separate messages, and still keeps all of it out of the main thread.
Front-loading /btw at session start
You don’t have to wait until mid-task to use /btw. Using it at session start before your main task prompt is a smart way to inject project context without that context becoming the opening of your conversation thread:
/btw context for this session: this is a legacy Python 2.7 codebase we're gradually migrating to 3.11, the team prefers explicit over implicit, and we're not using type hints yet but planning to add them during this migration
Your main task prompt then stays focused on the actual task rather than doubling as a context dump.
Using /btw to calibrate output format
A frequent source of unnecessary thread exchanges: asking Claude to change how it’s formatting output. These adjustments often turn into small back-and-forth exchanges that don’t add value to the main task thread.
Route them through /btw instead:
/btw the code you're generating is showing up with explanatory comments on each line — for this project, please remove inline comments and let the code speak for itself
/btw I'd prefer you give me each function in a separate code block rather than one big file — easier for me to review incrementally
Format adjustments handled without extending the main thread.
Documenting important /btw answers selectively
Sometimes a /btw answer turns out to be more important than you expected — important enough that you want it in the main thread as a reference point.
In those cases, you can explicitly bring it into the thread:
/btw does asyncio.create_task() work the same way inside and outside a running event loop?
If the answer reveals something important, follow up in the main thread:
Worth capturing: Claude confirmed that create_task() requires a running event loop, while asyncio.run() creates its own. We'll use run() as our entry point to avoid the event loop dependency issue.
This gives you selective control. The initial question and exploration happened via /btw, but the conclusion you want to reference later gets anchored in the main thread with your deliberate framing.
/btw for iterative model calibration
During complex sessions, you might find Claude’s responses drifting in ways you want to correct — too verbose, too terse, using an approach you don’t prefer. Address these via /btw:
/btw you've been giving me a lot of explanatory text before the code — for the rest of this session, lead with the code and put explanations after, or skip them if the code is self-explanatory
/btw the error handling pattern you're using is wrapping every function in try/except — we prefer to let exceptions propagate and handle them at the service boundary
These are behavioral corrections that don’t need to live in the main thread.
Common Mistakes When Using /btw
Using /btw for consequential decisions
This is the most common and impactful mistake. If a question might result in a significant change to your approach, it needs to be in the main thread.
The test: would you want to refer back to this exchange later? Would future-you want Claude to be able to trace the reasoning? If yes, it’s a main-thread exchange.
Routing a decision through /btw means that decision isn’t clearly anchored in the main conversation. If you change direction based on a /btw answer, Claude may not have clear visibility into why the direction changed when it needs to reason about it later.
Forgetting /btw exists
More common than misuse is non-use. After a few days of work, many developers stop thinking about /btw as an option and default back to normal messages for everything.
A useful habit to build: before you type any message in Claude Code, take a brief moment to classify it. Is this advancing the primary task, or is it a question/aside? If it’s an aside, type /btw first.
After a week of conscious practice, the classification becomes automatic and fast. The workflow cost of using /btw is essentially zero once it’s habitual.
Writing multi-paragraph /btw messages
/btw is optimized for lightweight exchanges. If you find yourself writing a /btw message that spans multiple paragraphs or needs a detailed, multi-part answer, reconsider.
Complex, lengthy exchanges often involve reasoning and nuance that benefits from being in the main thread where Claude can refer back to it. The exchange is probably substantive enough to warrant main-thread treatment.
A rough heuristic: if your /btw message is more than three to four sentences, think about whether it should actually be a main-thread exchange.
Assuming /btw context carries to the next session
/btw context, like all Claude Code session context, ends with the session. It doesn’t persist when you start a new conversation.
If something injected via /btw is important enough to matter in future sessions, add it to your CLAUDE.md file. Don’t rely on /btw as a memory system across sessions.
Treating /btw as the only token management strategy needed
Using /btw consistently helps — but it’s not a complete strategy on its own. Your main thread can still grow unwieldy if your primary task involves many iterative steps, detailed responses, or extensive back-and-forth.
Use /btw as one layer of a broader context management approach:
- Keep your main task prompts focused and specific
- Use
/compactwhen the main thread grows long - Put project-level context in CLAUDE.md rather than repeating it each session
- Be intentional about which exchanges need main-thread treatment
/btw handles one specific problem — preventing asides from bloating your context — very well. The rest of your context hygiene still matters.
Building Efficient AI Workflows at Scale with MindStudio
The token efficiency principle behind /btw — keeping different types of work in appropriately scoped contexts rather than letting everything accumulate in one growing thread — applies beyond Claude Code sessions. It’s a useful design principle for any AI-powered workflow that runs across multiple steps or agents.
MindStudio is a no-code platform for building and deploying AI agents, and it handles this architecture challenge natively. When you build multi-step workflows in MindStudio, each step operates in its own scoped context rather than a single growing monolith. A data-fetching step doesn’t carry the full token weight of a reasoning step that came before it. A formatting step doesn’t re-read everything that happened during the research phase.
This is particularly relevant for teams extending Claude Code into automated pipelines. If you’re building workflows where Claude Code agents hand off to other agents, or where AI-generated outputs flow into further processing steps, context management becomes an architectural concern — not just a per-session habit.
MindStudio’s Agent Skills Plugin provides an npm SDK (@mindstudio-ai/agent) that lets agents — including Claude Code agents — call typed capabilities as simple method calls. Things like agent.searchGoogle(), agent.runWorkflow(), agent.sendEmail(). The infrastructure layer — rate limiting, retries, authentication — is handled by the SDK, so the agent focuses on reasoning rather than plumbing.
For developers who want to apply the same “right context for the right task” logic at the system level that /btw applies at the session level, MindStudio offers a practical way to do that without building the infrastructure from scratch.
You can try MindStudio free at mindstudio.ai.
Frequently Asked Questions
What does /btw stand for in Claude Code?
/btw stands for “by the way.” The name reflects the type of message it’s designed for: quick questions, tangential context, and asides that are useful in the moment but don’t need to be part of the main conversation thread. It’s the kind of thing you’d naturally say as a parenthetical in conversation — relevant, but not central.
Does the /btw command actually save tokens, or is it mainly organizational?
It does both, and the token savings are real. Because /btw exchanges don’t get appended to your main conversation history, they don’t increase the context size that Claude reads on every subsequent request. Since Claude reads the full conversation history on every exchange, a leaner history means meaningfully lower token costs on every future message. The organizational benefit — keeping your conversation thread focused and readable — is a side effect, not the primary goal.
Can I use /btw to give Claude instructions about how it should behave?
Yes. Style preferences, output format instructions, verbosity calibration, code conventions — these all work well as /btw messages. You’re updating Claude’s working approach for the session without those instructions becoming part of your task thread. The only thing to keep in mind is that like all /btw context, these instructions are session-scoped. They don’t persist to your next Claude Code session.
What happens if I send a very long message via /btw?
Very long /btw messages are processed, but they reduce the command’s token efficiency advantage. The /btw mechanism is designed for lightweight, scoped exchanges. When you send a lengthy /btw message, it still processes in a separate context from the main thread, but the volume of work involved suggests the exchange might be substantial enough to warrant main-thread treatment — especially if it will inform a major part of the task.
Is /btw session-specific, or does it persist across sessions?
Session-specific only. Context provided via /btw — facts, preferences, constraints — is available to Claude for the current session. When you start a new Claude Code session, that context is gone. If a piece of /btw context proves repeatedly useful across sessions, move it to your CLAUDE.md file, where it will be loaded at the start of every session automatically.
Does /btw work with all Claude Code models?
The /btw functionality works within Claude Code’s interface regardless of which underlying model you’ve selected. The absolute dollar savings from using /btw will be higher with more expensive model tiers — Opus and Sonnet cost more per token than Haiku — but the relative savings percentage stays roughly similar. Teams using more capable models for their Claude Code work have proportionally more to gain from consistent /btw usage.
How do I know when to use /btw versus a regular message?
The clearest heuristic: would you want to reference this exchange later in the session, or does the decision you’re making here need to be traceable in your conversation history? If yes — main thread. If no — /btw. A secondary heuristic: is this advancing the primary task, or is it a question or context addition that supports the task without being part of it? Supporting context goes to /btw. Task advancement goes to the main thread.
Can /btw be used in Claude Code’s headless or API mode?
The /btw command is a Claude Code feature for interactive sessions. In headless or API usage patterns, context management works differently and the /btw mechanism may not apply in the same way. For automated pipelines using the Claude API directly, you’d typically manage context through message role scoping and custom system prompts rather than slash commands designed for interactive use.
Are there any tasks /btw is specifically not suited for?
Yes. /btw is not well-suited for: consequential decisions that will change your task direction, debugging exchanges where history and iteration matter, any exchange you’ll want Claude to explicitly reference later, or multi-step reasoning where the back-and-forth is part of the work. For anything where the process of arriving at an answer matters — not just the answer itself — use the main thread.
Key Takeaways
The Claude Code /btw command is a targeted solution to a specific problem: the token cost of mid-task questions and context additions that don’t need to be part of your main conversation history.
Here’s what matters:
- Token costs compound in Claude Code because Claude reads the full conversation history on every exchange. A lean main thread means every future message is cheaper.
/btwkeeps asides in a separate context, preventing those exchanges from adding to your running thread length.- The savings are most significant in long sessions with frequent mid-task questions — up to 50% in token costs for sessions that fit that profile.
- Best use cases are lightweight: quick factual lookups, context injections, style preferences, quick confirmations. Not consequential decisions or core task steps.
- Combine it with other tools:
/compactfor main-thread compression, CLAUDE.md for persistent cross-session context,/costfor tracking actual savings. - The habit is worth building early: once using
/btwfor asides becomes automatic, the workflow overhead is essentially zero.
If you’re building AI workflows beyond individual Claude Code sessions — multi-step pipelines, autonomous agents, production systems — MindStudio applies the same “right context for the right task” architecture at the system level. It’s worth exploring if your AI work is growing beyond what individual sessions can handle efficiently.