Files
selfheal-lab-apps/ai/internal/agent/agent_test.go
T
seon ed3d41d399 feat(ai): external responder for external-* model ids (reference runs)
모델 id가 external- 로 시작하면 ai-agent 가 LLM API 를 부르지 않고 각 턴(시스템
프롬프트·대화 이력·도구 정의·강제 제출 여부)을 GET /external/next 로 내어주고
POST /external/reply 로 답을 받는다. 진단 루프·프롬프트·도구·턴 상한은 다른 모델과
동일하므로, API 경로가 없는 모델(Claude Mythos 5.1 을 Claude Code 서브에이전트가
구동)의 참고 측정에 쓴다. 토큰 사용량은 알 수 없어 0 으로 기록한다.

external-* 진단은 EXTERNAL_DIAGNOSE_TIMEOUT_S(기본 1500초), 한 턴은
EXTERNAL_TURN_TIMEOUT_S(기본 1200초)까지 기다리고, 하네스는 같은 모델에 26분을
기다린다. /selfcheck 에 external_responder 상태(대기 턴·마지막 폴링 시각)를 추가.
2026-10-16 09:05:12 +09:00

152 lines
7.1 KiB
Go

package agent
import (
"context"
"encoding/json"
"net/http"
"net/http/httptest"
"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"
"github.com/selfheal-lab/selfheal-lab-apps/ai/internal/tools"
)
// A mock Responses API: turn 1 asks for get_probe_status, turn 2 submits the diagnosis. Verifies the loop wiring
// (function_call_output with the call_id, previous_response_id chaining, strict submit parsing).
func TestDiagnoseLoop(t *testing.T) {
var turns int
var sawOutput, sawPrev bool
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
}
var req openai.Request
_ = json.NewDecoder(r.Body).Decode(&req)
turns++
for _, it := range req.Input {
if it.Type == "function_call_output" && it.CallID == "call_1" && strings.Contains(it.Output, "probe_success") {
sawOutput = true
}
}
if req.PreviousResponseID == "resp_1" {
sawPrev = true
}
if r.Header.Get("Authorization") != "Bearer test-key" {
t.Errorf("missing bearer")
}
switch turns {
case 1:
w.Write([]byte(`{"id":"resp_1","status":"completed","output":[{"type":"function_call","call_id":"call_1","name":"get_probe_status","arguments":"{}"}],"usage":{"input_tokens":10,"output_tokens":5}}`))
default:
args := `{\"root_cause_label\":\"db_primary_down\",\"confidence\":0.8,\"evidence\":[\"probe_success db-primary=0\"],\"proposed_action\":\"escalate\",\"target\":\"db-primary\",\"rationale\":\"primary unreachable\"}`
w.Write([]byte(`{"id":"resp_2","status":"completed","output":[{"type":"function_call","call_id":"call_2","name":"submit_diagnosis","arguments":"` + args + `"}],"usage":{"input_tokens":20,"output_tokens":9}}`))
}
}))
defer oaiSrv.Close()
mimir := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if strings.HasSuffix(r.URL.Path, "/api/v1/query") {
w.Write([]byte(`{"status":"success","data":{"resultType":"vector","result":[{"metric":{"target":"db-primary"},"value":[1,"0"]},{"metric":{"target":"alb"},"value":[1,"1"]}]}}`))
return
}
w.Write([]byte(`{"status":"success","data":{"resultType":"matrix","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: "m", SystemPrompt: "sys", Topology: "topo", SLO: "slo", HitlURL: "http://127.0.0.1:1", MaxTurns: 5}, oai, tr, harness.New("http://127.0.0.1:1"), nil, enums, "pv1")
res, err := ag.Diagnose(context.Background(), Request{Alert: map[string]any{"alertname": "DbPrimaryDown"}})
if err != nil {
t.Fatal(err)
}
if res.RootCauseLabel != "db_primary_down" || res.Turns != 2 || !sawOutput || !sawPrev {
t.Fatalf("unexpected result: %+v sawOutput=%v sawPrev=%v", res, sawOutput, sawPrev)
}
ag.Dispose(context.Background(), Request{Alert: map[string]any{}}, res)
if res.Disposition != "escalated" {
t.Fatalf("escalate action must escalate, got %s", res.Disposition)
}
}
func TestTargetFor(t *testing.T) {
m := targetFor("k8s.rollout_restart", "demo/api")
if m["namespace"] != "demo" || m["deployment"] != "api" {
t.Fatalf("%v", m)
}
if targetFor("mig.resize", "")["mig"] != "selfheal-legacy-mig" {
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)
}
}
// Model routing: external-* goes to the external responder when one is set, and is refused when none is configured.
func TestProviderRouting(t *testing.T) {
a := New(Config{}, nil, nil, nil, nil, diag.Enums{}, "v")
if _, err := a.provider("external-claude-mythos-5-1"); err == nil || !strings.Contains(err.Error(), "external responder not configured") {
t.Fatalf("expected unconfigured error, got %v", err)
}
fake := &fakeResponder{}
a.SetExternal(fake)
if p, err := a.provider("external-claude-mythos-5-1"); err != nil || p != Responder(fake) {
t.Fatalf("external routing: %v %v", p, err)
}
if !IsExternal("external-x") || IsExternal("claude-x") || IsExternal("gpt-6-luna") {
t.Fatal("IsExternal prefix check")
}
}
type fakeResponder struct{}
func (*fakeResponder) CreateResponse(context.Context, openai.Request) (*openai.Response, error) {
return nil, nil
}