Title: Automated agent production r&d
2025-01-05 19:19:36 +0800
-----
2025-01-04 04:36:08 -0800
had chatgpt write instructions for cursor to develop a system for creating agents, then gave those instructions to claude3.5 to look over and improve. claude3.5 voyaged to the hamlet. check out this outline.
SYSTEM CREATION INSTRUCTIONS (Opening Statement)
Purpose and Outcomes
- What This System Will Create
- Examples of Agents You Can Expect
- What You'll Receive
- Access to Results
- Practical Benefits
- Control and Customization
1. System Architecture
1.1 Core Components
1.2 Resource Management
1.3 Security Framework
2. LLM Collaboration Protocol
2.1 Interaction Rules
2.2 Quality Control
2.3 Version Control
3. Development Workflow
3.1 Concept Generation
3.2 Implementation Phase
4. Monitoring and Logging
4.1 Metrics Collection
4.2 Logging Configuration
4.3 Alerting Thresholds
5. Documentation Requirements
5.1 Code Documentation
5.2 Required Sections
6. Deployment Protocol
6.1 Containerization
6.2 Deployment Steps
6.3 Rollback Procedure
7. Error Handling
7.1 Required Error Cases
7.2 Recovery Procedures
8. Success Metrics
8.1 Quantitative Metrics
8.2 Qualitative Metrics
9. Maintenance
9.1 Regular Tasks
9.2 Update Protocol
10. Communication Protocol
10.1 Implementation Requirements for User Communication
(Including email_config structure)
11. Enhanced Logging System
11.1 Required Logging Implementation
11.2 Required Log Entry Format
11.3 Required Log Access Implementation
11.4 Required Log Analysis Features
11.5 Log Retention Implementation
12. System Integration
- Integration requirements
- Communication verification
- Workflow testing
- Error handling validation
13. Pre-Launch Checklist
- Security verification
- System configurations
- Testing completion
- Monitoring setup
14. Handover Requirements
- Initialization instructions
- Configuration details
- Documentation requirements
- Recovery procedures
END OF SYSTEM CREATION INSTRUCTIONS
2025-01-04 05:07:39 -0800
had claude look it over one more time and now cursor agent is working on it. I told it to do autoaccept it still prompts me for "accept" a lot.
2025-01-04 08:09:26 -0800
claude3.5 through the cursor agent seems to have built a big and pretty fancy system. it's not quite what I was going for but it looks like such a nice piece of work that I thought I'd try it. but now I'm in the debugging process and god damn I'm not so sure I want to go through this
2025-01-04 15:43:22 -0800
just now back from a break of several hours. the system generated based on claude3.5's instruction document is basically useless. I'm going to start fresh with the orignial instructions generated by chatgpt. gpt's doc is 832 words.claude3.5's was 3021 words. I think it's an easy guess what the problem was. the question now is whether cursor agent (which will be claude3.5) will build anything satisfactory with briefer instructions. it'll be better, but will it be satisfactory?
2025-01-04 15:55:28 -0800
okay starting that now
2025-01-04 17:32:15 -0800
going much better this time. the main concept seems to be better understood. we've got the collaborative llms working to create agents. they've created three so far and are working on a third test.
2025-01-04 17:53:50 -0800
moving along. now we've implemented testing, and of course that's a whole lot of fun. this part may take a while, but if we can get it right it'll be pretty epic
2025-01-04 18:38:50 -0800
I figured out that there are already ai-based debugging packages out there and I told this to the cursor agent. he said 'really can I see one?' so I cloned a git hub and showed it to him and his response was basically 'this is really interesting. let's not use it though, we should just steal all the ideas.' so I guess that's what we're doing. with credit of course. the agent assures me that he's keeping notes of all the methods that he's borrowing
of course the agent is having a heck of a time writing it so it works.
2025-01-05 03:16:05 -0800
gradually gradually getting there. in my mind I'm thinking about starting this over from scratch. but we're advancing at a pace that keeps me hopeful.