Skills & Expert Routing
Route a problem to the smallest useful set of reusable expertise.
Primitive already has Skills/Tools reuse paths and a deterministic Role Router. The broader target is adaptive expert selection rather than loading every skill or assigning fixed specialists.
Instead of asking every expert in the building about every question, Primitive tries to call only the few people who actually know something useful about this problem.
Skills are reusable procedures; routing decides which procedures or expert roles should be active for a particular task. The architectural goal is bounded, context-sensitive selection that can evolve with evidence, not a permanent roster of agents that must always participate.
Implemented routing
Workspace includes a Role Router and deterministic route planning. The reuse indexes also map Skills, Tools, MCP/invocation pathways and execution machinery.
Target expert selection
The target model uses contextual relevance to select a bounded expert neighborhood, with the current Primitive design treating seven as a maximum selection cap rather than a quorum.
Skill lifecycle
A mature skill should be inspectable, reusable, attributable and promotable/demotable based on evidence. The docs distinguish this lifecycle from raw prompts or one-off agent instructions.
Where this description comes from
These pointers are the authority behind the status and technical description. If implementation and documentation disagree, the canonical implementation or Control contract wins and this page should be corrected.