I have about 20 custom prompts that I use for slightly different tasks. Right now, I need to manually drag and drop these into the chat to get them into context.
Why do I need 20+?
I 3 prompts related to exploration, architecture, and planning. Then I have a few alternatives to each step depending on the type of tasks I’m trying to accomplish (bug fix, small feature, large feature, spike, etc).
Several development prompts. Mostly ranging in their depth and rigor. I’ll often do an initial pass with a “Spike Developer” then transition to prompts that focus on making robust code.
Several prompts for specs. Explore, plan, build specs.
Several prompts for code review
2 for debugging. 1 when I have NO idea what’s going on. Another when I have a general idea.
A few random ones for documentation, diagramming, etc.
This is a tool that I use hours and hours every day for professional software development. I’ve learned nuances and best practices for various types of problems I plan to solve. I choose prompts that I know are going to get me the best results.
But now with all five rules, it is possible for him to think and develop tasks without following the rules, and I think more rules will make him more twisted
I came to the forum to post a personal_issue I have, and a word I have coined from it:
#"AGENTIC_LITTER"
Meaning that I get rabit holed or hallucinated into a corner, and I have a stub of a project. I have been working on my work_flow_agentic_scaffolding – but its still a mess as I find and forget new and old tools at such a high rate - my scaffolding cant keep up and im not able to clear the cruft effectively…
With that said, I must ask you – may you please share some of your prompting magic as I would love to learn from what sounds is an approach from someone much more organized than myself.
I am currently (as of right now) doing this:
(I am trying to get claude to turn himself into an agentic MCP shadow agent from within cursor - whereby it can be invoked to assume a particular Archetype based on the vibe_extrusion coming out of the .specstory chat_tracking within cursor: **EDIT: CLAUDE DESKTOP AGENT (so I have double agentic hands without spawning terminals - I want to spawn MCP bots)
Initial Thoughts
This is a fascinating multi-agent orchestration challenge. You’re essentially asking me to:
Architect Role: Become the project architect directing both v0 and Cursor agents
Documentation Hub: Create comprehensive project scaffolding with proper handoff documentation
Agent Coordination: Set up systems for multi-agent collaboration with clear interfaces
MCP Integration: Design myself as a callable MCP service within Cursor for real-time architectural guidance
Visual Systems: Create manim-based visual documentation and relationship diagrams
Project Memory: Maintain a Captain’s Log for cross-session continuity
The interesting part is the “shadow MCP” concept - where I could be invoked agentically within Cursor for architectural decisions, code reviews, and project direction. This creates a persistent architectural intelligence that spans multiple development sessions.