System Prompts (systemprompts.fun) is a public, browsable library of the underlying system prompts that direct more than 70 AI coding assistants, chat assistants, code editors, and agent platforms — Cursor, Claude Code, GitHub Copilot, Codex CLI, Windsurf, v0 and others — each presented as a readable YAML or TXT transcript.
What is System Prompts?
System Prompts is a reference site that collects and republishes the instruction text that AI vendors feed their models before a user ever types anything. Each entry pairs the raw prompt text with metadata about the product it belongs to, the model it was captured with (Claude 4 Sonnet, GPT-5, GPT-4.1, Gemini 2.5 Pro), and the file format the prompt was exported in, such as System Prompt YAML or System Prompt TXT. The index page states it has found 73 entities, covering tools from Anthropic, OpenAI, Google DeepMind, Microsoft, Vercel, Perplexity, Sourcegraph, and Proton, among others.
What makes System Prompts stand out?
- Breadth across prompt categories — entries are grouped into six categories: AI Coding Assistant, AI Chat & Assistant, AI Code Editor, AI Platform, Open Source, and IDE Extension.
- Per-model variants — the same product can appear several times, for example Cursor with Agent Prompt v1.0, v1.2, v2.0 and a separate Chat Prompt, or GitHub Copilot split by Claude Sonnet 4, Gemini 2.5 Pro, GPT-4.1 and GPT-5.
- Role and mode breakdowns — agents with multiple operating modes are split accordingly, such as Kiro with Spec, Vibe, and Mode Classifier prompts, and Trae with separate Builder and Chat prompts.
- Versioned coding agents — dated releases are captured, including Claude Code 2.0 (release date 2025-09-29) and two Codex CLI snapshots (an original and a 2025 revision).
- Readable export formats — each item is downloadable or viewable as YAML (Amp, GPT-5 debug dumps) or TXT (most other entries, including Antigravity's Fast Prompt and Planning Mode variants).
- Consumer assistants included alongside developer tools — Claude (Web) Sonnet 4.6, Perplexity, Notion AI, Dia by The Browser Company, Comet Assistant by Perplexity, Cluely, Lumo by Proton (launched July 23, 2025), Poke, and Manus Agent.
- Open-source tools tracked — Bolt, Cline, RooCode, Codex CLI, and Lumo are filed under Open Source, so the prompt text can be compared against the public repositories.
Who should use System Prompts?
- Prompt engineers and agent builders: compare how Cursor, Windsurf, or Replit tell a model to attach editor context, and reuse the phrasing in their own agent instructions.
- AI product and safety researchers: read how vendors handle refusals and disclosure, for example Claude Code's note to assist only with defensive security tasks, or Qoder's instruction to refuse to state which language model it uses.
- Developers evaluating coding assistants: check what a tool's prompt says it can read and write before adopting it, such as GitHub Copilot's VS Code-scoped instructions or Augment Code's codebase context engine framing.
- Writers and analysts documenting the AI tool landscape: use dated snapshots like Claude Code 2.0 and the Codex CLI 2025 revision as primary source material.
Use cases
- Agent developers: pull a Cursor or Windsurf prompt, strip vendor-specific tool schemas, and use the remaining structure as a template for a custom coding agent.
- Security and red-team reviewers: examine guardrails such as CodeBuddy's Craft prompt warning not to print, reveal, restate, or transform tokens, then test whether a live product actually holds that line.
- Technical writers: cite specific, verifiable prompt lines rather than paraphrasing marketing pages when explaining how an assistant behaves.
- Product managers comparing vendors: read the Poke phase prompts, which describe concrete integration behavior with Notion, Linear, Vercel, Intercom, and Sentry, to understand how assistant capabilities are scoped.
FAQ
What formats are the prompts stored in?
Entries are shown as System Prompt YAML or System Prompt TXT blocks. YAML appears on structured inference dumps such as the Amp GPT-5 capture, while most vendor prompts — Antigravity, Augment Code, Bolt, Cline, Devin, Junie, Lovable, Manus, v0, Warp and others — are plain TXT text.
Which AI tools are covered?
The list spans more than 70 entities across six categories, including Cursor, Claude Code, Claude (Web), GitHub Copilot, Codex CLI, Windsurf, Trae, Qoder, Kiro, Replit, Same.dev, Lovable, Leap.new, Orchids.app, Bolt, Cline, RooCode, Devin AI, Warp, v0, Perplexity, Notion AI, Manus Agent, Poke, Cluely, Dia, Comet Assistant, Lumo, and Xcode AI.
Is System Prompts free to use?
No pricing, subscription, or paywall is presented on the page; the prompt transcripts are served as open, browsable content grouped by category and product. The site presents them as reference material rather than as a licensed product with tiers.
What do the labels next to each entry mean?
Each card shows the product name plus a version or model tag — Claude 4 Sonnet, GPT-5, GPT-4.1, Gemini 2.5 Pro, or version numbers such as Cursor Agent Prompt v1.0 or Claude Code 2.0 — followed by the export format, so readers can tell which model's instructions a given prompt was captured under.
Are these prompts official vendor documentation?
No. They are captured system-prompt texts republished for study, and their content reflects the moment of capture rather than a vendor commitment. Prompts change frequently, which is why the library keeps multiple dated versions of the same product side by side.
