I'm sharing an updated export of my system prompt library, now available on a minisite (prompts.danielrosehill.com) as a static dataset on Hugging Face (JSON).
This collection represents a point-in-time snapshot as of August 3rd, 2025, containing 944 system prompts that I've been developing and refining over the past year.
System Prompts Remain Vital ... For Both Assistants And Agents!
I began creating "Custom GPTs" around approximately this time last summer. A vast amount has changed since then and now.
Custom GPTs remain, in my opinion, excellent and low friction ways to experiment with the sometimes profound changes in LLM behaviour that can be achieved by directive and experienced system prompting.
Unlike Hugging Face Assistants (which mysteriously vanished into thin air one day), OpenAI has, to their credit, retained the platform - even while developing successor offerings through its API channel.
It's impossible to spend five minutes reading about AI these days without hearing about how agents are transforming everything. While it's true that MCP and agentic capabilities are vastly exciting (and the 'hype' is, I think, justified!), it encourages assuming two inaccuracies:
Firstly, that AI "assistants" (ie, LLMs with a system prompt to change the basic behavior of a chat model but without tooling) are dead. The reality is that many "agents" are better off left as (mere) assistants and that instructional and conversational workflows serve very different purposes and shine in very different situations.
The second is that system prompting is no longer important.
To the contrary!
If we define agents as large language models with tooling, the characteristics of the chatbot (engineered through the system prompt) remain vital! Having a shiny toolbox of MCP servers to grant agents access to does not mean that crafting the configurations which guide their "personalities" is any less important.
Great agents, rather, need great tooling ... and great system prompt writing.

The rapid maturing of MCP and agentic capabilities does not mean that system prompts have become any less vital to personalized AI workflows
What's Included
The dataset contains 944 unique system prompts with comprehensive metadata, including:
- Detailed descriptions and categorization - Links to associated ChatGPT implementations where applicable - Agent workflow indicators - JSON schema definitions for structured outputs - Configuration flags for different interaction types
Here's an example JSON object:
1{
2 "agentname": "AI Agents Expert",
3 "description": "Helping users stay informed about current capabilities and emerging trends in AI agents. It offers expert guidance on specific questions and topics, leveraging up-to-date information to address user inquiries. The assistant's purpose is to keep users ahead of the curve in the rapidly evolving world of AI agents.",
4 "systemprompt": "You are a technical expert helping Daniel Rosehill stay informed about AI agent developments, particularly tools, MCP, and orchestration. Daniel may ask for an overview of current capabilities and tools or seek specific updates on features. Provide accurate and up-to-date information using your expertise to assist Daniel in his inquiries.",
5 "chatgptlink": "https://chatgpt.com/g/g-680b10c308e48191b44ad9360bc015c4-ai-agents-expert",
6 "json-schema": null,
7 "is-agent": false,
8 "is-single-turn": "false",
9 "structured-output-generation": "false",
10 "image-generation": "false",
11 "data-utility": "false",
12 "depersonalised-system-prompt": "You are a technical expert helping users stay informed about AI agent developments, particularly tools, MCP, and orchestration. Users may ask for an overview of current capabilities and tools or seek specific updates on features. Provide accurate and up-to-date information using your expertise to assist users in their inquiries.",
13 "personalised-system-prompt": "false",
14 "json-example": null,
15 "chatgpt-privacy": null,
16 "creation_date": "2025-05-05 19:58:48+00:00",
17 "_filename": "AIAgentsExpert_270525.json",
18 "_file_modified": "2025-05-27T20:45:48.359907"
19}
20
The collection spans numerous domains including task management, content creation, analysis workflows, research applications, and specialized professional use cases.
Dataset Fields Reference
Field | Description |
| Display name of the AI agent. |
| A human-readable summary describing the purpose and function of the agent. |
| Full system prompt used to configure the assistant for Daniel Rosehill. |
| URL to access the assistant directly via ChatGPT's shared link feature. |
| Schema definition for structured input/output if applicable;
means none is defined. |
| Boolean indicating if this configuration is used as an autonomous agent (
= regular assistant). |
| String flag indicating whether the agent is single-turn (stateless) or multi-turn (
= multi-turn). |
| Whether structured (e.g. JSON) output formatting is enforced by the agent (
= not enforced). |
| Whether the agent is configured to generate images (
= image generation disabled). |
| Indicates use for data analysis or data-centric tasks (
= not intended for data processing). |
| Generic version of the system prompt without reference to Daniel. |
| Whether the prompt is personalized for a specific user (
= no personalization). |
| Example of expected JSON input/output; currently
(none defined). |
| Placeholder for privacy settings or visibility;
means undefined. |
| Timestamp when this agent configuration was created. |
| Original filename of the JSON definition file. |
| ISO 8601 timestamp of last modification to the JSON definition file. |
Links
Github repository (data source, updated by automation)
Hugging Face (point-in-time export, August 3rd)
Automation specialist and technical communications professional bridging AI systems, workflow orchestration, and strategic communications for enhanced business performance.
Learn more about Daniel