feat: per-run model selection, usage in diagnosis results and a Claude Messages API provider

캠페인마다 진단 모델을 고를 수 있게 한다. 하네스는 runs.model 열을 추가해 POST /runs의
model을 저장하고, 활성 런의 model을 /diagnose 요청에 실어 보낸다. 에이전트는 요청의
model을 우선 사용하고(없으면 OPENAI_MODEL 기본값), 모델 id가 claude-로 시작하면
internal/anthropic(Claude Messages API 어댑터)로 라우팅한다. 어댑터는 Responses API
항목(message/function_call/function_call_output)을 system·tools(input_schema)·tool_use/
tool_result로 변환하고 previous_response_id 체인을 클라이언트 쪽에서 재현하며, Claude 5.5가
강제 tool_choice를 거부하므로 강제 호출은 auto + 지시문으로 바꾼다. 키는
ANTHROPIC_KEY_FILE(/run/secrets/anthropic-api-key)에서 읽고 비어 있으면 명확한 오류를 낸다.
Result에 usage(입력·출력·합계 토큰)를 추가해 하네스 diagnosis 이벤트에 비용 근거를 남긴다.

ai/internal/agent/agent_test.go

@@ -8,6 +8,7 @@ import (
 	"strings"
 	"testing"
 
+	"github.com/selfheal-lab/selfheal-lab-apps/ai/internal/anthropic"
 	"github.com/selfheal-lab/selfheal-lab-apps/ai/internal/diag"
 	"github.com/selfheal-lab/selfheal-lab-apps/ai/internal/harness"
 	"github.com/selfheal-lab/selfheal-lab-apps/ai/internal/openai"
@@ -81,3 +82,48 @@ func TestTargetFor(t *testing.T) {
 		t.Fatal("mig")
 	}
 }
+
+// A claude-* model id routes the same loop through the Anthropic adapter; the request model wins over Config.Model
+// and the summed usage is returned in the result.
+func TestDiagnoseRoutesClaudeModel(t *testing.T) {
+	var models []string
+	anth := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var b map[string]any
+		_ = json.NewDecoder(r.Body).Decode(&b)
+		models = append(models, b["model"].(string))
+		args := `{"root_cause_label":"db_primary_down","confidence":0.8,"evidence":["e"],"proposed_action":"escalate","target":"db-primary","rationale":"r"}`
+		w.Write([]byte(`{"id":"msg_a","type":"message","model":"claude-sonnet-5-5","stop_reason":"tool_use","content":[{"type":"tool_use","id":"toolu_a","name":"submit_diagnosis","input":` + args + `}],"usage":{"input_tokens":40,"output_tokens":8}}`))
+	}))
+	defer anth.Close()
+	oaiSrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/embeddings" {
+			w.Write([]byte(`{"data":[{"index":0,"embedding":[0.1,0.2]}]}`))
+			return
+		}
+		t.Errorf("openai must not be called for a claude-* model: %s", r.URL.Path)
+	}))
+	defer oaiSrv.Close()
+	mimir := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Write([]byte(`{"status":"success","data":{"resultType":"vector","result":[]}}`))
+	}))
+	defer mimir.Close()
+	oai := openai.New(oaiSrv.URL, func() (string, error) { return "test-key", nil })
+	tr := tools.New(tools.Config{MimirURL: mimir.URL, LokiURL: mimir.URL, EmbedModel: "e"}, nil, oai)
+	enums := diag.Enums{Labels: diag.DefaultLabels, Actions: diag.DefaultActions}
+	ag := New(Config{Model: "gpt-default", SystemPrompt: "sys", Topology: "topo", SLO: "slo", HitlURL: "http://127.0.0.1:1", MaxTurns: 3}, oai, tr, harness.New("http://127.0.0.1:1"), nil, enums, "pv1")
+	req := Request{Alert: map[string]any{"alertname": "DbPrimaryDown"}, Model: "claude-sonnet-5-5"}
+	if _, err := ag.Diagnose(context.Background(), req); err == nil || !strings.Contains(err.Error(), "anthropic provider not configured") {
+		t.Fatalf("without SetAnthropic a claude-* model must fail clearly, got %v", err)
+	}
+	ag.SetAnthropic(anthropic.New(anth.URL, func() (string, error) { return "k", nil }))
+	res, err := ag.Diagnose(context.Background(), req)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if res.Model != "claude-sonnet-5-5" || len(models) != 1 || models[0] != "claude-sonnet-5-5" || res.RootCauseLabel != "db_primary_down" {
+		t.Fatalf("routing: res=%+v models=%v", res, models)
+	}
+	if res.Usage["input_tokens"] != 40 || res.Usage["output_tokens"] != 8 || res.Usage["total_tokens"] != 48 || res.Usage["turns"] != 1 {
+		t.Fatalf("usage: %v", res.Usage)
+	}
+}