Question on approach and effectiveness

[brycereynolds]

First off, thank you to the author and contributors for this tool and for releasing it to the public!

I think I'm missing a key aspect of this. I was using the individual commands to execute tasks (e.g., /two-stage-opportunity-pipeline:task-4 ), but I was seeing tokens get up to 20k or more and it would take a very long time, before writing a single file (completing a single task). I get there is some context setting going on here.

But then I run a command like this directly against claude...

Review @.claude/specs/two-stage-opportunity-pipeline/tasks.md. I want you to work through Phase 2. Review the requirements document @.claude/specs/two-stage-opportunity-pipeline/requirements.md and the
  @.claude/specs/two-stage-opportunity-pipeline/design.md for more context

=> 
⏺ Update Todos
  ⎿  ☒ Review existing Temporal workflow patterns and structure
     ☐ Task 4: Create discovery workflow interface and types
     ☐ Task 5: Implement discovery workflow logic
     ☐ Task 6: Create LangGraph discovery activity
     ☐ Task 7: Create candidate storage activity
     ☐ Task 8: Create validation workflow
     ☐ Task 9: Register workflows in temporal worker

I get the first task completed with about 2k tokens. In total it gets through tasks 4-9 with about 11k tokens and in a fraction of the time.

I thought perhaps that I've missed something critical here. Just curious to here your thoughts.

I really liked the generation of the requirements.md, design.md and tasks.md. I was doing something similar (not nearly as rigorous and just a "spec" document) in the past. But your approach is really solid and I can see myself adopting it going forward.

Thanks in advance for your time!

[Pimzino]

First off, thank you to the author and contributors for this tool and for releasing it to the public!

I think I'm missing a key aspect of this. I was using the individual commands to execute tasks (e.g., /two-stage-opportunity-pipeline:task-4 ), but I was seeing tokens get up to 20k or more and it would take a very long time, before writing a single file (completing a single task). I get there is some context setting going on here.

But then I run a command like this directly against claude...

Review @.claude/specs/two-stage-opportunity-pipeline/tasks.md. I want you to work through Phase 2. Review the requirements document @.claude/specs/two-stage-opportunity-pipeline/requirements.md and the

  @.claude/specs/two-stage-opportunity-pipeline/design.md for more context



=> 

⏺ Update Todos

  ⎿  ☒ Review existing Temporal workflow patterns and structure

     ☐ Task 4: Create discovery workflow interface and types

     ☐ Task 5: Implement discovery workflow logic

     ☐ Task 6: Create LangGraph discovery activity

     ☐ Task 7: Create candidate storage activity

     ☐ Task 8: Create validation workflow

     ☐ Task 9: Register workflows in temporal worker

I get the first task completed with about 2k tokens. In total it gets through tasks 4-9 with about 11k tokens and in a fraction of the time.

I thought perhaps that I've missed something critical here. Just curious to here your thoughts.

I really liked the generation of the requirements.md, design.md and tasks.md. I was doing something similar (not nearly as rigorous and just a "spec" document) in the past. But your approach is really solid and I can see myself adopting it going forward.

Thanks in advance for your time!

Hi Brey,

First of all, thank you so much for your kind words I really appreciate it.

Agreed there is some fine tuning required for the slash commands and the agents in particular and I was working on that but I've since put this project on the back burner as I am working on an McP version of the workflow which would support all ai agents with MCP support to allow everyone to enjoy this framework / workflow.

Happy for you to contribute to this project based on your testing with revised prompts, they are written in markdown format within sec/markdown if you wish and I will happy merge the pull request if I'm happy with the changes.

I think going forward an MCP version of this tool will be better in the long run to avoid the just Claude code support as I use multiple agentic tools but I also think from the wider community McP seems to be the go to.

McP will allow me to better handle token consumption which has been a massive request especially from pro and max5 users.

Thanks again
Pimzino

[brycereynolds]

Ya that makes sense. For sure I'll contribute as opportunities arise. Thanks again and gl!