11 points | by mbbah 4 days ago ago
2 comments
There's an interesting experiment here, with a related blog: https://github.com/grapeot/devin.cursorrules?tab=readme-ov-f...
Basically instead of a complex layer, it just uses .cursorrules as the memory. This was before MCPs, so it might be capable of more today.
We’ve been exploring typed task graphs as an alternative to shared memory. Turns coordination into lineage rather than state. Surprisingly scalable. Happy to compare notes.
There's an interesting experiment here, with a related blog: https://github.com/grapeot/devin.cursorrules?tab=readme-ov-f...
Basically instead of a complex layer, it just uses .cursorrules as the memory. This was before MCPs, so it might be capable of more today.
We’ve been exploring typed task graphs as an alternative to shared memory. Turns coordination into lineage rather than state. Surprisingly scalable. Happy to compare notes.