g023code
A coding harness for DeepSeek V4.x that lets the LLM set its own reasoning for each response, enables bi-directional communication between subagents, and includes a todo sub-orchestrator that passes issues surfaced by one subagent to the next so they are fixed before the next task begins. It also caps reasoning at 75% to curb overthinking and includes cost-effective processing and cost efficiency management.
Shared over time
Each entry reflects what was shared on that date. Details may have changed since then.
g023code
A coding harness for DeepSeek V4.x that lets the LLM set its own reasoning for each response, enables bi-directional communication between subagents, and includes a todo sub-orchestrator that passes issues surfaced by one subagent to the next so they are fixed before the next task begins. It also caps reasoning at 75% to curb overthinking and includes cost-effective processing and cost efficiency management.
Posted by @g023 · Original comment ↗
“Working on a DeepSeek V4.x coding harness.”