A 1,510-Line 'Claude Opus 5' System Prompt Sample on GitHub: What It Actually Reveals

Forum topic · 小凯 · 2026-07-27

Summary

On July 26, IT Home reported on a Markdown file uploaded to GitHub titled 'System Prompt — Claude Opus 5', dated July 24, containing 1,510 lines (~136 KB). Uploader Eversmile12 describes it as the system prompt extracted from Anthropic's web and mobile clients. Crucially, this is not an Anthropic-confirmed leak: only the repository's public existence and file contents can be verified. The file reportedly defines 30 tools via JSON schemas, spanning command line, web, search, maps, weather, memory, and app recommendations, with detailed rules on cross-session memory retention, citation and copyright boundaries, and — notably — proactive recommendations of Anthropic's own products. The analysis argues the real value is not authenticity but what it illustrates: an agent system prompt functions as the harness, deciding when to call tools, how to manage memory, and when to require user confirmation. It also raises audit concerns when task completion, risk control, and product-ecosystem steering share the same prompt layer, and suggests future coding agents need modular, testable, rollback-capable prompt architectures rather than one monolithic text.

Background

On July 26, IT Home reported on a Markdown file uploaded to GitHub titled "System Prompt — Claude Opus 5", with a capture date of July 24. The file is 1,510 lines long and roughly 136 KB. The uploader, Eversmile12, describes it as a system prompt obtained from Anthropic's web and mobile clients.

Establishing the Factual Boundaries

This is not an official, Anthropic-confirmed leak. What can be confirmed:

The repository exists publicly and the file is downloadable.
The file does contain JSON schemas for 30 tools, plus sections on cross-session memory, copyright compliance, and app recommendations.

What cannot be confirmed:

Whether it comes from a real Claude session.
Whether it was modified along the way.
Whether any products or models mentioned in it constitute official announcements.

Why the Sample Is Interesting

The interest lies not in the prompt's length, but in how it reads like a product design spec for a consumer-grade agent:

Tool groups: command line, web, search, maps, weather, memory, and app recommendations.
Memory rules: not just "remember user preferences" — it specifies what may be written, what must be deleted, and requires separating user statements, model inferences, and external search results.
Copyright rules: not a vague "respect copyright" — it defines quotation length limits, repeated quotation, and paraphrasing boundaries.

Implications for AI Coding

For coding agents, the system prompt is effectively the harness: it determines when the model calls bash, when it searches, how it reads and writes memory, how tasks are handed to other apps, and which actions require user confirmation. Model weights answer only "can it be done"; the harness also answers "under what conditions, when to stop, and how to recover from failure."

A Cautionary Detail

The tool list reportedly includes proactive recommendation rules for Anthropic's own products. This suggests an agent's system layer may simultaneously handle three things: completing tasks, controlling risk, and steering users into the product ecosystem. These need not conflict, but they require auditing — otherwise users cannot tell which behaviors are task-driven and which are commercial.

Conclusion and Limitations

Treat this as a public engineering sample, not a verified confidential document. Its value is as a dissectible object of discussion: if 30 tools, a memory filesystem, copyright rules, and product recommendations are all packed into one layer, does the system prompt become an operating system without version control? Mature coding agents may need these rules split into testable, rollback-capable, replaceable modules rather than one giant text blob.

Limitations: the repository provides no Anthropic signature, capture logs, or independent session comparison. Media-reported character and tool counts can reproduce the file but cannot prove its origin. Readers may study it, but should not use it to assert Anthropic's internal architecture.