Lately, everyone has been sharing SKILL.md and AGENT.md files.
I think the more important layer is actually SPEC and MEMORY.
That is the idea behind this repo:
Ultimate Project combines two separate disciplines:
Beastmode -> the software engineering discipline
spec-kit -> the structure and organization of specs
Beastmode gives the engineering workflow.
spec-kit gives the project a clear spec foundation.
Putting them together creates a development loop where the system starts from specs, keeps context in memory, and follows a more structured implementation lifecycle.
For small project, a simple history file ./store-my-history in a dir works. For bigger projects, I think structured specs + engineering memory matter much more than prompt files alone.
I’d be curious how others here are handling spec organization, memory, and iteration flow in agent-driven development.
We’ve been running Rook-Ceph in production across multiple client environments. In one example, we built a setup with 8 refurbished Dell servers (128GB RAM, 8–14 JBOD disks each) over 10G networking. It supports geo-replication between sites and has been stable for over 2 years. Total hardware cost was under $100k.
Rook simplifies the operational overhead of Ceph quite a lot, especially in Kubernetes-native stacks. For teams with large data and HA requirements, it's been a solid on-prem alternative. SLA-backed managed services are also becoming more common, which helps reduce the operational burden even further.