---
title: "Discussion on \"Looking Through Claude's Context Window\" | Hashnode"
description: "Discussion on \"Looking Through Claude's Context Window\". What actually fills a Claude Code context window before you type anything, how memory, tools, skills and MCP differ, and why sub-agents are important"
url: "https://hashnode.com/posts/looking-through-claude-s-context-window/6a833d8ec3b7bb972b00d568"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Looking Through Claude's Context Window](https://downtherabithole.dev/looking-through-claude-s-context-window?utm_source=hashnode&utm_medium=feed)

What actually fills a Claude Code context window before you type anything, how memory, tools, skills and MCP differ, and why sub-agents are important

Aug 17 · 9 min read · downtherabithole.dev  
Updated Aug 27  
Read the article

-   [#claude](https://hashnode.com/tag/claude)
-   [#claude-code](https://hashnode.com/tag/claude-code)
-   [#ai](https://hashnode.com/tag/ai)
-   [#llm](https://hashnode.com/tag/llm)
-   [#skills](https://hashnode.com/tag/skills)
-   [#agentic-ai](https://hashnode.com/tag/agentic-ai)
-   [#mcp](https://hashnode.com/tag/mcp)
-   [#context-engineering](https://hashnode.com/tag/context-engineering)
-   [#context-window](https://hashnode.com/tag/context-window)
-   [#compaction](https://hashnode.com/tag/compaction)

## Written by

[George Sims](https://hashnode.com/@george-sims)

@george-sims · Barcelona, Spain

  
[View profile](https://hashnode.com/@george-sims)

Senior DevOps Engineer @ Bynder

Writing about systems, engineering and developer experience.

## Responses 4

Markdown supported · ⌘↵ to post

[Skillselion](https://hashnode.com/@skillselion)

[Aug 18](https://hashnode.com/posts/looking-through-claude-s-context-window/6a833d8ec3b7bb972b00d568/comment/6a84c99099a2ee3b0e855477)

"Sending an agent to read it and report back costs you a paragraph" is the sentence I would put on a poster. It also has a second-order benefit the post implies but does not spell out: the sub-agent shields the main thread from tool-result bloat, not just file contents. A failed grep, a 500-line test log, a noisy build output, all of it stays in the child window and only the conclusion crosses over. In long sessions that is where most of my context actually went when I measured it with /context, not the files I deliberately read. Small addition on skills: the name-plus-description roster is a rounding error per skill, but it scales with how many you install, so a hoarder setup pays a real tax before invoking anything. disable-model-invocation is also handy for skills you only ever want by name. Curious what the next post says about context rot thresholds, since "more recall" degrading into "worse recall" matches what I see around the compaction boundary.

[George Sims](https://hashnode.com/@george-sims)

[Aug 19](https://hashnode.com/posts/looking-through-claude-s-context-window/6a833d8ec3b7bb972b00d568/comment/6a84c99099a2ee3b0e855477#reply-6a85eb5c443b90a9b39f09e1)

Hi Skillselion, using subagents is a small superpower in of itself when used correctly. Good observation too, as you pointed out there's a lot more being saved by the agent than suggested by this post. Also a fair point about skill hoarders, there's certainly a threshold of being detrimental rather than useful at scale. I think you will enjoy the next one, I will get deeper into the rabbit hole about context rot.. the research is quite dense so it's taking a bit longer, but stay tuned!

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Aug 18](https://hashnode.com/posts/looking-through-claude-s-context-window/6a833d8ec3b7bb972b00d568/comment/6a84a1b24786a4a92afe84b6)

Getting concrete about what actually sits in the context window is underrated, because most "the model ignored my instruction" bugs I chase turn out to be a context problem, not a reasoning one. Once I started printing the exact assembled prompt, half of those mysteries resolved themselves. Good walkthrough of looking at what the model is really seeing.

[George Sims](https://hashnode.com/@george-sims)

[Aug 19](https://hashnode.com/posts/looking-through-claude-s-context-window/6a833d8ec3b7bb972b00d568/comment/6a84a1b24786a4a92afe84b6#reply-6a85e9834516adf5578e673f)

Thanks for your feedback Kartik, much appreciated. Indeed, as the input token count grows the model accuracy deteriorates quickly depending on how vague your prompt is. Hoping to show that in some more detail in my next post, stay tuned!