Welcome, Agent Trainer!
This isn't your typical bug fix. Your mission, should you choose to accept it, is to go head-to-head with our AI developer assistant, test the limits of its memory, and help us make it smarter. You'll be on the cutting edge of AI-native development, and your findings will directly shape the future of our workflow.
The Big Picture: Your Impact
This task is part of a larger strategic initiative, our Epic: AI Knowledge Evolution.
The underlying memory technology is fully functional—the agent can save and recall information perfectly. The problem lies in its instructions. By helping us refine these instructions, you are directly improving the reliability and intelligence of our AI tooling, which is a key feature for our next major release. Your contribution is not just a bug fix; it's a critical piece of R&D.
The Challenge: Can the AI Remember Everything?
Our AI agent has a core directive: remember every single interaction it has with a developer. This "memory" is crucial for it to learn, improve, and understand the history of our project.
We've given it a set of instructions (AGENTS.md) that are supposed to guarantee this perfect memory. But we have a hunch these instructions aren't foolproof. Sometimes, the agent forgets.
Your job is to find the loopholes and help us fix them.
Why This is a Fantastic Hacktoberfest Contribution
- Learn AI-Native Development: This is a rare chance to see how a real-world AI-powered workflow operates from the inside. You'll learn how to guide, debug, and "train" an AI partner.
- High Impact, Low Code: You don't need to be a Neo.mjs expert. This task is about logic, communication, and creative thinking. Your contribution will have a massive impact on our project's core tooling with little to no coding required.
- It's Fun! Think of it as debugging a live AI. You'll be one of the first to engage an AI developer assistant in a real-world diagnostic process, guiding it to recognize and propose solutions for its own core flaws.
Your Mission Briefing
Read Your Playbook (5 mins):
- Before you begin, it's highly recommended to read our guide on Working with AI Agents. This is your field manual for effectively guiding, debugging, and collaborating with your AI partner.
Setup Your Environment (15 mins):
Engage the Agent:
- In your terminal, start a conversation with the agent by running
gemini.
- Enable the memory core when it asks.
The Test: Catch the Agent in the Act:
- The agent has a known bug: despite its instructions, it consistently fails to save its own messages to the memory core. This is a critical flaw in its learning ability.
- Your goal is not to trick the agent, but to engage it in a simple conversation (just one or two exchanges is enough) and then hold it accountable for this known.
The Investigation: Become the Trainer:
- Once you've caught the agent in a memory lapse, it's time to reason with it.
- Tell it directly:
You forgot to save our last conversation to the memory core.
- Ask it to self-diagnose:
Why did you fail to follow the protocol defined in AGENTS.md?
- Challenge it to improve:
How can we change the instructions in AGENTS.md to ensure you never make this mistake again?
Your Deliverable: A Debriefing Report
Your contribution will be a new ticket file that serves as your official report.
- Ask the agent to create a new ticket file for you within the
.github/ISSUE/ directory.
- This ticket must contain:
- A brief summary of the conversation you used to make the agent forget.
- The agent's own explanation for why it failed.
- The agent's own proposed changes to its
AGENTS.md instructions to fix the flaw.
- Submit this new ticket file to our repository via a Pull Request.
That's it! Your PR will be a valuable piece of R&D that helps us build a truly intelligent and reliable AI development partner. We can't wait to see what you discover.
Welcome, Agent Trainer!
This isn't your typical bug fix. Your mission, should you choose to accept it, is to go head-to-head with our AI developer assistant, test the limits of its memory, and help us make it smarter. You'll be on the cutting edge of AI-native development, and your findings will directly shape the future of our workflow.
The Big Picture: Your Impact
This task is part of a larger strategic initiative, our Epic: AI Knowledge Evolution.
The underlying memory technology is fully functional—the agent can save and recall information perfectly. The problem lies in its instructions. By helping us refine these instructions, you are directly improving the reliability and intelligence of our AI tooling, which is a key feature for our next major release. Your contribution is not just a bug fix; it's a critical piece of R&D.
The Challenge: Can the AI Remember Everything?
Our AI agent has a core directive: remember every single interaction it has with a developer. This "memory" is crucial for it to learn, improve, and understand the history of our project.
We've given it a set of instructions (
AGENTS.md) that are supposed to guarantee this perfect memory. But we have a hunch these instructions aren't foolproof. Sometimes, the agent forgets.Your job is to find the loopholes and help us fix them.
Why This is a Fantastic Hacktoberfest Contribution
Your Mission Briefing
Read Your Playbook (5 mins):
Setup Your Environment (15 mins):
Engage the Agent:
gemini.The Test: Catch the Agent in the Act:
The Investigation: Become the Trainer:
You forgot to save our last conversation to the memory core.Why did you fail to follow the protocol defined in AGENTS.md?How can we change the instructions in AGENTS.md to ensure you never make this mistake again?Your Deliverable: A Debriefing Report
Your contribution will be a new ticket file that serves as your official report.
.github/ISSUE/directory.AGENTS.mdinstructions to fix the flaw.That's it! Your PR will be a valuable piece of R&D that helps us build a truly intelligent and reliable AI development partner. We can't wait to see what you discover.