AI HOT (Curated Pool)Jul 26, 2026, 13:33 UTC+8

Claude Opus 5 system prompt fully leaked: 135,027 characters, ~34K tokens

Claude Opus 5 系统提示词被完整泄露，共 135027 字符、约 3.4 万 token

Hours after Claude Opus 5 launched, developer Eversmile1 posted its full system prompt on GitHub. The 1,511-line, ~34K-token file contains zero code—only behavioral rules. Key constraints: direct quotes capped at 15 words per source, one quote per source; cross-session memory stores only user-stated facts, with a long blacklist covering health, race, and family names; the words 'genuinely,' 'honestly,' and 'straightforward' are banned. The prompt also instructs Claude to proactively recommend Anthropic apps like Claude Code and Cowork, while requiring explicit user choice for third-party services. Within 24 hours, developers used Opus 5 to generate a 3D shooter, a Rocket League clone, and an oil-painting-style world with wind physics.

Why it matters: The full Claude Opus 5 system prompt leaked—1,511 lines of behavioral rules now public, directly useful for prompt engineering and safety research. Not scored higher because this is a security incident, not an official release, and the post doesn't include Anthropic's response.

Read the original ↗Export Markdown

Take

This one's worth opening because it's not model weights that leaked—it's the rulebook for how Anthropic teaches Claude to behave. 1,511 lines, roughly 34K tokens, all behavioral constraints: direct quotes capped at 15 words per source, one quote per source max; cross-session memory only stores user-stated facts, with health, race, and family names blacklisted; even the words 'genuinely,' 'honestly,' and 'straightforward' are banned because using them makes Claude sound less trustworthy.

I'd read this as Anthropic's product boundary document. The memory chapter runs 230 lines, and the core logic is striking: Claude is required to remember you while being required not to act like it remembers you. It can't say 'I remember' or 'last time you mentioned' in replies—only 'we've discussed this before' if you explicitly ask about the memory system. The privacy blacklist is granular enough that even a vague placeholder like 'has a health condition to manage' is forbidden; it gets rewritten to 'interested in fitness.'

The self-promotion section is equally blunt: there's a dedicated recommend_claude_apps tool for pushing Claude Code, Cowork, and other Anthropic products, complete with suggested phrasing. But third-party services get the opposite treatment—rules explicitly prevent Claude from making choices on the user's behalf. Commercial intent and restraint clauses sitting side by side in the same config file. This document is more honest than most companies' internal product docs.

Source: AI HOT (Curated Pool) · ithome.com
Published: Jul 26, 2026, 13:33 UTC+8 · Jul 26, 2026