--- title: How Claude 5 changed the rules for context engineering description: Anthropic removed ~80% of the Claude Code system prompt for Claude 5, signaling a shift in how developers should approach prompt engineering. Elaborate system... canonical: https://daily.dev/posts/how-claude-5-changed-the-rules-for-context-engineering-poazhn1ad twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How Claude 5 changed the rules for context engineering | daily.dev og:description: Anthropic removed ~80% of the Claude Code system prompt for Claude 5, signaling a shift in how developers should approach prompt engineering. Elaborate system... og:url: https://daily.dev/posts/how-claude-5-changed-the-rules-for-context-engineering-poazhn1ad og:image: https://api.daily.dev/og/posts/pOAzHN1ad.png og:image:alt: How Claude 5 changed the rules for context engineering 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. # How Claude 5 changed the rules for context engineering **[Collections](https://daily.dev/sources/collections)** · 4 min read · 2 upvotes · 0 comments ## Summary Anthropic removed ~80% of the Claude Code system prompt for Claude 5, signaling a shift in how developers should approach prompt engineering. Elaborate system prompts and complex instructions that were necessary for earlier models now often hinder rather than help. The key insight is that the leverage has moved from optimizing how you prompt to finding novel applications for the tool itself. ## Content ## What changed The team behind Claude Code recently removed roughly 80% of its system prompt for Claude 5 models and saw no measurable drop in coding evaluations. That's a striking result, and it's worth understanding why it happened. The old prompt was long for a reason. Early models needed explicit rules to avoid bad behavior — don't delete files, don't write unprofessional comments, don't add multi-line comment blocks. The team wrote those rules knowing they'd sometimes be wrong, but the tradeoff made sense when models couldn't read context well enough to figure things out themselves. Newer models can. And that changes everything about how you should write system prompts. ## The problems with long prompts When you pile up rules written for less capable models, a few things go wrong. First, rules start conflicting. A skill might say "don't add comments" while the user asks about documentation. Claude can usually figure out what you mean, but it burns thinking on reconciling instructions instead of solving the actual problem. Second, rigid rules prevent good judgment. A rule like "don't write multi-line comment blocks" is a proxy for something — probably "don't clutter the code." But the model can now read the surrounding code and match its style directly. The proxy rule just gets in the way. Third, prompt bloat compounds like technical debt. Every guardrail you built for an older model is still a cost on a newer one, even if the model no longer needs it. ## What the new approach looks like Six concrete shifts came out of this rethink: **Describe goals, not rules.** Instead of "don't write multi-line comment blocks," the new prompt says "write code that matches the surrounding code." The model figures out the rest. **Design expressive tool interfaces instead of stuffing prompts with examples.** A status field limited to `pending`, `in_progress`, and `completed` shows proper usage without a single example. When you provide examples, models tend to copy them literally rather than reason about intent. **Load context progressively.** Skills activate only when needed. Some tools stay hidden until the agent discovers them — they don't consume any context until they're actually relevant. This replaces front-loading everything upfront. **Say things once.** Instructions live in tool descriptions, not repeated across the system prompt and multiple skill files. **Let memory handle itself.** Claude.md no longer needs to be a running notebook. Auto-memory saves context automatically. Keep Claude.md short — just genuine gotchas that aren't obvious from the repo. **Use rich artifacts instead of prose specs.** A test suite, an HTML mockup, or a rubric conveys more information than a paragraph of description. These give the model something concrete to work against. ## The broader point Boris Cherny, who created Claude Code, frames this as "unhobbling" — the idea that over-specified prompts actively prevent models from using capabilities they already have. Opus 5 can apparently run autonomously for days without scaffolding. Elaborate prompt engineering doesn't unlock that; it constrains it. The practical implication is that the tricks people developed to coax older models into behaving well are now mostly friction. The bigger opportunity is finding novel ways to apply these models to your work, not optimizing the harness. One caveat worth noting: this advice applies most cleanly to frontier models. Smaller or older models may still benefit from more explicit guardrails. And rebuilding your system prompt empirically after each major model release — rather than carrying forward old assumptions — is probably the right habit regardless. Context engineering has shifted from writing rules to designing how well the model can read a situation. That's a different skill, and most existing setups haven't caught up yet. --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-coding](https://daily.dev/tags/ai-coding), [#claude](https://daily.dev/tags/claude), [#prompt-engineering](https://daily.dev/tags/prompt-engineering), [#context-engineering](https://daily.dev/tags/context-engineering) [View this post on daily.dev](https://daily.dev/posts/how-claude-5-changed-the-rules-for-context-engineering-poazhn1ad) ```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":"How Claude 5 changed the rules for context engineering","url":"https://daily.dev/posts/how-claude-5-changed-the-rules-for-context-engineering-poazhn1ad","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-claude-5-changed-the-rules-for-context-engineering-poazhn1ad"},"datePublished":"2026-07-24T21:03:47.932Z","dateModified":"2026-07-28T21:56:17.514Z","description":"Anthropic removed ~80% of the Claude Code system prompt for Claude 5, signaling a shift in how developers should approach prompt engineering. Elaborate system...","isAccessibleForFree":true,"articleSection":"Collections","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":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/how-claude-5-changed-the-rules-for-context-engineering-poazhn1ad","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-coding,claude,prompt-engineering,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":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"How Claude 5 changed the rules for context engineering"}]} ```