--- title: A Claude Code skill was eating 200,000 tokens before... description: Anthropic's built-in /claude-api skill for Claude Code was found to consume over 200,000 tokens before answering a single question, because it inlined all of... canonical: https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: A Claude Code skill was eating 200,000 tokens before answering a single question | daily.dev og:description: Anthropic's built-in /claude-api skill for Claude Code was found to consume over 200,000 tokens before answering a single question, because it inlined all of... og:url: https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy og:image: https://api.daily.dev/og/posts/b8XOyeGPY.png og:image:alt: A Claude Code skill was eating 200,000 tokens before answering a single question og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # A Claude Code skill was eating 200,000 tokens before answering a single question **[The New Stack](https://daily.dev/sources/newstack)** · 4 min read · 0 upvotes · 0 comments ## Summary Anthropic's built-in /claude-api skill for Claude Code was found to consume over 200,000 tokens before answering a single question, because it inlined all of its reference documentation — including language-specific docs for eight languages and 26 shared Markdown files — regardless of relevance. Developers flagged the issue in GitHub reports in July and August, with one report measuring 812,650 bytes loaded when only a 32,954-byte file was needed. With Claude Code v2.1.234, Anthropic switched to on-demand loading of reference material, cutting the skill's footprint to roughly 25,000 tokens, an 85.7% reduction, freeing up context window space for actual repository work. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why was Claude Code's /claude-api skill using 200,000 tokens before answering anything? The skill inlined all of its reference documentation directly into the skill body instead of loading it on demand. When it could not detect a project's language, it loaded documentation for C#, cURL, Go, Java, PHP, Python, Ruby and TypeScript plus 26 shared Markdown files, totaling 812,650 bytes, even though only a single 32,954-byte file was needed for the task. _daily.dev surfaces fixes like this for developers tracking token costs in agentic coding workflows._ ### How much did Claude Code v2.1.234 reduce the /claude-api skill's token usage? Claude Code v2.1.234 cut the skill's token usage from over 200,000 tokens down to roughly 25,000 tokens, an at least 85.7% reduction. Anthropic achieved this by loading the skill's reference documentation on demand rather than bundling everything upfront, freeing more context window for the actual repository and task. _Developers optimizing agent context budgets can follow updates like this on daily.dev._ ### How should I structure a Claude Code skill's SKILL.md to avoid loading too much context? SKILL.md should contain only the core instructions and links, while API specifications, examples, and other detailed material should live in supporting files that Claude opens only when needed. Loading happens in three stages: always-present metadata, the SKILL.md body when the skill triggers, and bundled resources pulled in during the task as needed. _Teams designing Claude Code skills can track best practices like this via daily.dev._ ## Similar posts on daily.dev - [Claude Code cache confusion as Anthropic tweaks defaults, but quotas still drain](https://daily.dev/posts/claude-code-cache-confusion-as-anthropic-tweaks-defaults-but-quotas-still-drain-uvn7zes3k) · DEVCLASS · 5 upvotes · 1 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#anthropic](https://daily.dev/tags/anthropic), [#claude-code](https://daily.dev/tags/claude-code), [#context-engineering](https://daily.dev/tags/context-engineering) [View this post on daily.dev](https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"A Claude Code skill was eating 200,000 tokens before answering a single question","url":"https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy"},"datePublished":"2026-08-18T17:50:17.960Z","dateModified":"2026-08-18T18:41:35.474Z","description":"Anthropic's built-in /claude-api skill for Claude Code was found to consume over 200,000 tokens before answering a single question, because it inlined all of...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/af9146376b3953e96ba0ec0e1b67756f?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/af9146376b3953e96ba0ec0e1b67756f?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,anthropic,claude-code,context-engineering","timeRequired":"PT4M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"A Claude Code skill was eating 200,000 tokens before answering a single question"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/a-claude-code-skill-was-eating-200-000-tokens-before-answering-a-single-question-b8xoyegpy#faq","mainEntity":[{"@type":"Question","name":"Why was Claude Code's /claude-api skill using 200,000 tokens before answering anything?","acceptedAnswer":{"@type":"Answer","text":"The skill inlined all of its reference documentation directly into the skill body instead of loading it on demand. When it could not detect a project's language, it loaded documentation for C#, cURL, Go, Java, PHP, Python, Ruby and TypeScript plus 26 shared Markdown files, totaling 812,650 bytes, even though only a single 32,954-byte file was needed for the task. daily.dev surfaces fixes like this for developers tracking token costs in agentic coding workflows."}},{"@type":"Question","name":"How much did Claude Code v2.1.234 reduce the /claude-api skill's token usage?","acceptedAnswer":{"@type":"Answer","text":"Claude Code v2.1.234 cut the skill's token usage from over 200,000 tokens down to roughly 25,000 tokens, an at least 85.7% reduction. Anthropic achieved this by loading the skill's reference documentation on demand rather than bundling everything upfront, freeing more context window for the actual repository and task. Developers optimizing agent context budgets can follow updates like this on daily.dev."}},{"@type":"Question","name":"How should I structure a Claude Code skill's SKILL.md to avoid loading too much context?","acceptedAnswer":{"@type":"Answer","text":"SKILL.md should contain only the core instructions and links, while API specifications, examples, and other detailed material should live in supporting files that Claude opens only when needed. Loading happens in three stages: always-present metadata, the SKILL.md body when the skill triggers, and bundled resources pulled in during the task as needed. Teams designing Claude Code skills can track best practices like this via daily.dev."}}]} ```