test(nest): live-model crow delivery in the nest; ReadEvents log helper (#13)

nestlive_test.go

@@ -0,0 +1,97 @@
+package main
+
+import (
+	"context"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+	"time"
+)
+
+// newLiveNestServer builds the server over the real brain from env (NewBrain)
+// instead of the mock. Requires an API key; callers must skip first.
+func newLiveNestServer(t *testing.T) (*Server, string) {
+	t.Helper()
+	dir := t.TempDir()
+	ledger, _ := LoadLedger(filepath.Join(dir, "ledger.json"))
+	flog, err := OpenFlightLog(filepath.Join(dir, "flight.jsonl"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	t.Cleanup(func() { flog.Close() })
+	nests, _ := LoadNests(filepath.Join(dir, "nests.json"))
+	msgs, _ := LoadMessageStore(filepath.Join(dir, "messages"))
+	ctx, cancel := context.WithCancel(context.Background())
+	approver := NewWebApprover(time.Minute)
+	p := &Pipeline{Flock: NewFlock(NewBrain(), DefaultGuardrails(), flog), Ledger: ledger, Log: flog, G: DefaultGuardrails(), Approver: approver}
+	s := NewServer(ctx, p, approver, nests).WithMessages(msgs)
+	ts := httptest.NewServer(s.Handler())
+	t.Cleanup(func() { cancel(); ts.Close() })
+	return s, filepath.Join(dir, "flight.jsonl")
+}
+
+func TestLiveCrowDeliversInNest(t *testing.T) {
+	if os.Getenv("OPENROUTER_API_KEY") == "" && os.Getenv("ANTHROPIC_API_KEY") == "" {
+		t.Skip("no model key")
+	}
+	s, logPath := newLiveNestServer(t)
+	nest, err := s.nests.Create("Live Lab", []string{"Hugin"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	req, err := s.msgs.Append(Message{NestID: nest.ID, Kind: "request", Author: "owner", To: "Hugin", Text: "What is the plural of crow?", Timestamp: time.Now().UTC().Format(time.RFC3339)})
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Synchronous call, like the endpoint's goroutine would.
+	s.respondToRequest(nest, req)
+
+	msgs := s.msgs.All(nest.ID)
+	var reply *Message
+	for i := range msgs {
+		if msgs[i].Author == "Hugin" {
+			reply = &msgs[i]
+		}
+	}
+	events, err := ReadEvents(logPath)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	if reply == nil {
+		// No crow message at all: either a flight failure (429/timeout: the
+		// known free-model instability -> skip) or nothing was even attempted.
+		flightFailed := false
+		for _, e := range events {
+			if e.Kind == "flight" && e.Error != "" {
+				flightFailed = true
+			}
+		}
+		if flightFailed {
+			t.Skip("upstream model failed (known free-model instability); the flag path is covered by #11")
+		}
+		t.Fatalf("no Hugin reply after a request, log: %+v", msgs)
+	}
+	kind := reply.Kind
+	if kind != "deliver" && kind != "claim" && kind != "decline" {
+		t.Fatalf("live crow reply must be responsive, got %q (%s)", kind, reply.Text)
+	}
+	if reply.To != "owner" {
+		t.Fatalf("deliver must answer the requester, got to=%q", reply.To)
+	}
+	if reply.RefersTo != req.MsgID() {
+		t.Fatalf("reply must point at the request, got %q", reply.RefersTo)
+	}
+	// The call went through Flock.ask, so the flight log has Hugin's flight.
+	flights := 0
+	for _, e := range events {
+		if e.Crow == "Hugin" && e.Model != "" && !strings.Contains(strings.ToLower(e.Model), "mock") {
+			flights++
+		}
+	}
+	if flights == 0 {
+		t.Fatalf("want a logged non-mock Hugin flight, events: %+v", events)
+	}
+}