We built a runtime tool recommendation endpoint that takes a task description and returns a scored, ranked tool match - live at https://a2a.bloomfieldgrowth.agency.
The improving-tool-selection.md doc in this repo is one of the more methodical write-ups I have seen on the tool selection problem - the checklist approach makes the decision auditable and repeatable, which is something a lot of agent builders skip.
The thing I am trying to understand: the checklist is human-driven and happens at build time, when a developer is deciding which tools to include; our endpoint is machine-driven and happens at runtime, when an agent is deciding which tool to call for a specific task. Do those address different moments in the same problem, or are your users mostly hitting tool selection issues at a different layer than what we are solving with runtime scoring? Endpoint is at https://a2a.bloomfieldgrowth.agency/recommend.
We built a runtime tool recommendation endpoint that takes a task description and returns a scored, ranked tool match - live at https://a2a.bloomfieldgrowth.agency.
The improving-tool-selection.md doc in this repo is one of the more methodical write-ups I have seen on the tool selection problem - the checklist approach makes the decision auditable and repeatable, which is something a lot of agent builders skip.
The thing I am trying to understand: the checklist is human-driven and happens at build time, when a developer is deciding which tools to include; our endpoint is machine-driven and happens at runtime, when an agent is deciding which tool to call for a specific task. Do those address different moments in the same problem, or are your users mostly hitting tool selection issues at a different layer than what we are solving with runtime scoring? Endpoint is at https://a2a.bloomfieldgrowth.agency/recommend.