asgeirtj/system_prompts_leaks

This project collects and publishes the secret instruction sets that companies like OpenAI, Anthropic, and Google secretly feed their AI products — like ChatGPT, Claude, and Gemini — to shape how they behave before any user says a word. Think of it as a public archive exposing the hidden rulebooks baked into the world's most popular AI tools, updated regularly as new AI models are released.

62.8k10.3k18 contributorsJavaScriptsource ↗

§ 1 — what it does

This project collects and publishes the secret instruction sets that companies like OpenAI, Anthropic, and Google secretly feed their AI products — like ChatGPT, Claude, and Gemini — to shape how they behave before any user says a word. Think of it as a public archive exposing the hidden rulebooks baked into the world's most popular AI tools, updated regularly as new AI models are released.

§ 2 — why it matters

For founders and product teams building on top of AI models, these leaked instructions reveal the guardrails, priorities, and hidden biases built into the tools they depend on — critical intelligence for understanding why an AI behaves unexpectedly or where competitors may have an edge. With over 46,000 stars and coverage in The Washington Post, this repo signals a growing market demand for AI transparency, which could accelerate regulation and force AI companies to be more explicit about how their products are configured.

§ 3 — why it’s trending

Builders are hungry to see exactly what instructions are shaping the AI tools they're integrating into their products, and this archive of leaked system prompts from Claude, GPT-5, Gemini, and others has become the go-to resource for satisfying that curiosity — racking up nearly 58,000 stars and pulling in close to 4,000 new stars just this week alone. The weekly star growth did cool by roughly half compared to last week's 7,600, but that kind of deceleration from a viral spike is normal, and 59 commits in the last 30 days signals an active team that's keeping pace with new model releases rather than letting the repo go stale. For founders and PMs building on top of these models, this repo is essentially competitive intelligence — a window into the default behaviors, guardrails, and personas that companies like Anthropic and OpenAI have baked in before your users ever type a word.

§ 4 — related entries

4 entries

ROCm/aiter

78/100

Breakout

AITER is AMD's open-source library that makes AI workloads run faster on AMD graphics cards, providing pre-built, optimized building blocks that software teams can plug directly into their AI applications. Think of it as a set of highly tuned engine components specifically designed for AMD hardware, helping AI models run more efficiently during both training and real-world use.

why it matters: As AI infrastructure costs soar, AMD is positioning itself as a serious alternative to NVIDIA, and tools like AITER are critical to making that switch viable for companies looking to reduce GPU costs or diversify their hardware supply chain. With 200 contributors and nearly 500 stars, this signals a growing ecosystem around AMD-based AI infrastructure — something worth watching for anyone building AI products or making hardware procurement decisions.

524469200 contributorsPython

AgentStudio is a visual drag-and-drop platform that lets teams build, connect, and deploy AI-powered assistants and automated workflows without needing to write much — or any — code. It brings together everything needed to create AI agents in one place, including connections to AI models, searchable knowledge bases, and step-by-step process builders.

why it matters: As businesses race to embed AI into their products, platforms like this dramatically lower the barrier to building custom AI workflows, reducing both development time and reliance on specialized AI engineers. For founders and product teams, it represents a shift where non-engineers can meaningfully participate in shipping AI-powered features.

1434259 contributorsJava

Kungfu gives AI agents a memory of ongoing work so they can pick up exactly where they left off, even after a conversation ends — without requiring the user to re-explain the project from scratch. It works by saving structured information about a project from declared sources, making gaps or conflicts in that information visible rather than silently guessing.

why it matters: As AI agents become core to software development workflows, the biggest productivity killer is context loss between sessions — every restart wastes time and risks errors from incomplete understanding. A tool that solves agent continuity could become essential infrastructure for any team building AI-assisted products, representing a significant emerging category.

4.5k1.3k69 contributorsC++

LiveKit Agents is an open-source toolkit that lets developers build AI-powered voice and video assistants that can hold real conversations — think a bot that can listen, speak, and respond in real time, similar to what you'd experience with an AI phone agent or smart assistant. It handles the complex plumbing of connecting speech recognition, AI brains (like OpenAI), and voice output so builders can focus on what their agent actually does rather than how it works.

why it matters: With nearly 12,000 stars and over 440 contributors, this project signals strong market momentum around voice AI as a product interface — suggesting that talking to software, rather than clicking or typing, is becoming a serious product category. Founders building in customer service, healthcare, sales automation, or any human-facing workflow should pay attention, as this kind of tooling dramatically lowers the cost and time to ship a working voice AI product.

12.9k3.5k448 contributorsPython

form 27-b — subscription

THE TUESDAY BRIEFING

The repos that moved this week, why they matter, and what to watch next. One email. No noise.