Added review and implement commands #235
Conversation
There was a problem hiding this comment.
There’s room for making the instructions more robust and useful. A few suggestions:
- More instructions on task tracking
- Add references to project memory
- Add SDD phylosophy to context
- Avoid unnecessary Human-In-The-Loop interpolations. [NEEDS CLARIFICATION] from /spec stage is a reference.
- (Optional) Allow the agent to delegate some tasks to other agents (that works through mcp), and even run multithreaded requests for tasks marked with [P]
My idea is to make implementation stage agent more orchestrator-like instead of asking him to write a lot of independent modules consistently.
You will coordinate with other agents or tools (if available) for tasks like code writing, code review, testing, or deployment, maintaining a seamless CONSTITUTION.
From my own implementation.
You can check it out my as a reference: sdd-implementation-stage.md.
I didn’t start this PR with these changes because I’m still not sure whether my prompt is optimal for everyone. I also needed more time to test it.
|
I understand, as they are these commands are working for me properly, maybe if you merge your improved commands with this we will have a passable v1 |
|
Okay, i'll do it today |
The review command is based on the readme.md recommendation to ask Claude Code to review the tasks.md file.
The implement command is based on my own implementation methodology for SDD.