GA Chatbot Probe

GA Chatbot Probe

Use this skill when you need to test agent behaviour, compare routing decisions, or get music theory answers from the live GA chatbot (Ollama-backed, SemanticRouter-dispatched).

When to Use

• Testing that SemanticRouter picks the right agent for a given query
• Comparing how TheoryAgent vs CriticAgent answer the same question
• Getting a real chatbot answer as context while writing code
• Verifying a newly deployed agent change works end-to-end

Single Agent Query

curl -sk -X POST https://localhost:7001/api/ga/eval \
  -H "Content-Type: application/json" \
  -d "$(jq -n --arg q "What intervals make up a Cmaj7 chord?" \
       '{script: ("invoke \"agent.theoryAgent\" (Map.ofList [\"question\", box " + ($q | @json) + "])")}')"

Or in a GAL script:

invoke "agent.theoryAgent" (Map.ofList ["question", box "What intervals make up a Cmaj7 chord?"])

The SemanticRouter inside the GA API picks the best specialist (Theory / Tab / Critic / Composer) based on embedding similarity. The agent closure does not force a specific agent — routing still happens at the API level.

Fan-Out: Compare Multiple Agents

To send the same question to multiple agent closures in parallel and collect all answers:

invoke "agent.fanOut" (Map.ofList [
    "question",   box "Analyse the progression Am - F - C - G"
    "agentNames", box [| "agent.theoryAgent"; "agent.criticAgent" |]
])

Returns Map<string, string> — agent name → response. Display side-by-side.

Direct Chatbot Endpoint (no GAL)

For debugging routing metadata (agent selected, confidence, routing method):

curl -sk -X POST https://localhost:7001/api/chatbot/chat \
  -H "Content-Type: application/json" \
  -d '{"message": "Show me fingering for an open G chord", "useSemanticSearch": true}' \
  | python -m json.tool

Response shape:

{
  "naturalLanguageAnswer": "...",
  "candidates": [...],
  "routing": {
    "agentId": "tab-agent",
    "confidence": 0.91,
    "routingMethod": "embedding"
  }
}