Rewrite your AI prompts using the Claude Fable 5.2 system prompt as a template - in 10 minutes
Take Anthropic's officially published system prompt, feed it to any free AI chatbot, and get an improved version of your prompt built on the same techniques Anthropic uses to tune their best model
- 1.Open the officially published Claude Fable 5.2 system prompt page in Anthropic's documentation. Scroll through it and note the structure: blocks covering model identity, behavior rules, knowledge boundaries, and response formatting.
This is Anthropic's official document, updated November 9, 2026. A 426-thousand-character file circulating on social media from an anonymous account is unverified - do not use it as a source: 94% of its text cannot be cross-checked.
Anthropic Docs: System Prompts ↗About this tool - 2.Open Anthropic's prompt engineering guide for the same model and copy the «Finishing work» block - the text that keeps an agent from stopping halfway. You will need it in step 4.
According to Anthropic's tests, with this text roughly half as many agent sessions end with unfinished work. The guide is fully open and free.
Anthropic Docs: Prompting Claude Fable 5.2 ↗About this tool - 3.Open any free chatbot with a large context window: Claude (claude.ai), Kimi (kimi.com), DeepSeek (chat.deepseek.com), or GLM. Start a NEW conversation.
All listed services have a free tier and do not require a card. If one is unavailable in your region - just pick another from the list.
Claude / Kimi / DeepSeek ↗About this tool - 4.Paste the meta-prompt below into the chat. It asks the model to analyze the techniques from the Fable 5.2 system prompt and rewrite your prompt using the same rules. Replace [YOUR TASK IN ONE SENTENCE] and [YOUR CURRENT PROMPT] with your own.
Do not paste personal or confidential data into the prompt: the text goes to a third-party server. Your current prompt is what you normally write to an AI - for example, «write me an email to a client».
Claude / Kimi / DeepSeek ↗About this toolPromptBelow is the official system prompt of Claude Fable 5.2, published by Anthropic. Study how it is structured: explicit role definition, clear scope boundaries, formatting rules, tone instructions, and handling of edge cases. --- SYSTEM PROMPT --- [PASTE THE PROMPT TEXT FROM THE ANTHROPIC DOCS PAGE HERE] --- END SYSTEM PROMPT --- Now apply the same techniques to improve MY prompt. My task: [YOUR TASK IN ONE SENTENCE] My current prompt: [YOUR CURRENT PROMPT HERE] Rewrite my prompt so it: 1) defines an explicit role for the model, 2) states scope and boundaries, 3) specifies output format and length, 4) includes a rule to persist and finish the work without stopping to ask unnecessary questions. Output ONLY the improved prompt, ready to copy-paste. After it, add 3 short bullet points explaining what you changed and why.
What this prompt doesThe prompt is in English because that is the language of the published system prompt and AI models follow technical instructions more accurately this way. It says: «Below is the official Claude Fable 5.2 system prompt; study its structure (role, boundaries, format, tone) and rewrite MY prompt using the same techniques». Replace three placeholders: instead of [PASTE THE PROMPT TEXT...] insert the prompt text copied from the Anthropic docs page, instead of [YOUR TASK IN ONE SENTENCE] put your task in one sentence, instead of [YOUR CURRENT PROMPT HERE] put your current prompt. At the end, the model should output only the ready improved prompt and three - 5.Copy the improved prompt from the response. If you are working with an agent that abandons tasks halfway, add the «Finishing work» block copied in step 2 to the end of your prompt.
The key phrase from this block: «Please avoid stopping while you still owe work the user asked for. Be persistent and do not stop unless needed» - the model should work around errors and timeouts on its own, without asking permission for every step. But keep confirmation for genuinely dangerous actions.
Anthropic Docs: Prompting Claude Fable 5.2 ↗About this tool - 6.Test the improved prompt on your real task in a new conversation and compare the result with the old prompt. If the answer is still not ideal - go back to step 4 and add one line to the meta-prompt: what exactly was unsatisfactory in the first version.
Usually one or two iterations are enough. If the service is unavailable in your region, the entire recipe works in any other chatbot from the step 3 list - the techniques from the system prompt are universal.
Claude / Kimi / DeepSeek ↗About this tool
- freeEntire recipe workflowAnthropic documentation pages and all steps are free; the Claude, Kimi, DeepSeek, and GLM chatbots have free tiers with daily message limits and no card required - check the exact limits on each service's pricing page.
Why today
AI services change fast - interfaces and free limits may differ from what's described.