3 Rules to Steal From the Leaked Claude Fable 5.1 System Prompt
TL;DR
The full Claude Fable 5.1 system prompt leaked, and it runs over 270,000 characters. You don't need all of it. Three of its rules make Claude sound better right away: ban filler words, answer before asking, and never end on a bare "Done." Paste the block below into your CLAUDE.md and Claude Code follows them every session.
👉 https://www.skool.com/claudecodeclub
See what leaked in the Fable 5.1 system prompt
A system prompt is the rulebook Claude reads before every chat. It sets the tone, the format and what Claude will and won't do. Right after Claude Fable 5.1 launched, the hacker known as Pliny the Liberator posted the full claude.ai version. He called it "a WHOPPING 270,000+ characters."
Here is the part most people miss: Anthropic publishes the core of this prompt on its own docs site. The leak adds the rest: the memory rules, the tool rules and thousands of example lines. So the three rules below are real, and you can check each one on Anthropic's page.
Rule 1: Ban the filler words
Fable is told to never say "genuinely", "honestly" or "straightforward". Anthropic's reason: those words try to convince you instead of just making the point, and they come off as fake.
What this does for you: shorter answers that get to the point. Add your own pet peeves to the list too.
Rule 2: Answer first, then ask
Fable doesn't always ask questions, but when it does, it tries to answer even a vague request first. You get something useful right away instead of a list of questions back.
What this does for you: fewer back-and-forth turns. Claude takes its best shot, then asks one question if it really needs to.
Rule 3: Never end on "Done."
After its last step, Fable has to state the answer you asked for in one or two sentences. The prompt says it plainly: a sign-off alone, like "Done.", is not a reply.
What this does for you: in Claude Code, long jobs end with a clear result you can read in five seconds, instead of "Done." and a wall of tool output.
Paste all 3 rules into your CLAUDE.md
CLAUDE.md is the file Claude Code reads at the start of every session. Rules you put there apply to every chat in that project. Copy this block and add it to the file:
markdown## How to answer me
- Never say "honestly", "genuinely" or "straightforward". Say the point directly.
- If my request is vague, give your best answer first, then ask one question if you still need to.
- When you finish a task, tell me the result in one or two sentences. "Done." alone is not an answer.
- Every sentence should add something new. Cut filler and repeats.Prompts to try:
promptAdd this section to my CLAUDE.md file, keep everything else the same, then show me the final file:
## How to answer me
- Never say "honestly", "genuinely" or "straightforward". Say the point directly.
- If my request is vague, give your best answer first, then ask one question if you still need to.
- When you finish a task, tell me the result in one or two sentences. "Done." alone is not an answer.
- Every sentence should add something new. Cut filler and repeats.promptRead my CLAUDE.md and the Claude Fable 5.1 system prompt at https://platform.claude.com/docs/en/release-notes/system-prompts/claude-fable-5-1. Find 3 more rules from the prompt that fit how I work on [your kind of project], and suggest the exact lines to add.Read the full Fable 5.1 prompt
- Anthropic's official core prompt: https://platform.claude.com/docs/en/release-notes/system-prompts/claude-fable-5-1
- The full leaked version (over 270,000 characters): https://github.com/elder-plinius/CL4R1T4S/blob/main/ANTHROPIC/Claude-Fable-5.1.md
- A longer copy with the memory and tool sections (8,742 lines): https://github.com/asgeirtj/system_prompts_leaks/blob/main/Anthropic/claude-fable-5.1.md
👉 https://www.skool.com/claudecodeclub
Common questions
Is the Claude Fable 5.1 system prompt leak real?
The core part matches the prompt Anthropic publishes on its own docs site, dated September 1, 2026. The leaked copies add the memory, tool and example sections that Anthropic doesn't publish.
How long is the Fable 5.1 system prompt?
The leaked claude.ai version is over 270,000 characters. A longer copy that also has the memory sections runs 8,742 lines.
Do these rules work in Claude Code?
Yes. Put them in your CLAUDE.md file and Claude Code reads them at the start of every session.
Should I copy the whole system prompt?
No. Most of it covers claude.ai tools and safety cases you don't need. Copy only the rules that change how Claude answers you.
Want every rule, skill and prompt done for you?
Get 650+ plug-and-play skills, MCPs & prompts, plus 8,000+ members - $9/mo, cancel anytime.
Join the Club