feat(builder): diagnose a failed boot before changing the template

A failed boot went straight back to the model as "fix it", so it guessed:
one draft raised its memory three rounds in a row while the real cause, an
empty setting the server can't run without, stayed untouched.

Now a failed boot is investigated first:
- Evidence: the server's distinct error lines (timestamps and repeats
  removed), and a web search on the most telling one, with the top answer
  read through the reader.
- Diagnosis: the model states the cause, the evidence for it, its
  confidence and the exact fix. If unsure, it asks for up to 3 checks: a
  search, a page, the files the server writes (a first boot of the draft),
  or a boot test with settings or the startup command changed. Results go
  back to it, two rounds at most.
- Fix: the model applies exactly the diagnosed fix and nothing unrelated.
- Attempt log: each cause, fix and outcome. A fix that changed nothing
  rules its cause out; if the diagnosis returns to a ruled-out cause, the
  build stops instead of guessing, and the PR lists what the builder found.

Boot rounds go up to 5, since rounds now end on a ruled-out cause rather
than a fixed count.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

cmd/gbrgames/diagnose_test.go

@@ -0,0 +1,128 @@
+package main
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"testing"
+)
+
+// fakeModel answers with the given replies in order and keeps what it was asked.
+type fakeModel struct {
+	replies []string
+	asked   [][]chatMessage
+}
+
+func (f *fakeModel) name() string { return "fake" }
+func (f *fakeModel) chat(m []chatMessage, _ map[string]any) (string, error) {
+	f.asked = append(f.asked, append([]chatMessage{}, m...))
+	r := f.replies[0]
+	f.replies = f.replies[1:]
+	return r, nil
+}
+
+func TestDiagnoseRunsRequestedChecksBeforeAnswering(t *testing.T) {
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if strings.HasPrefix(r.URL.Path, "/search") {
+			w.Write([]byte(`{"results":[{"url":"https://example.com/fix","title":"Server stops at start","content":"set a password"}]}`))
+			return
+		}
+		w.Write([]byte("The server refuses to run without a join password."))
+	}))
+	defer srv.Close()
+	games := t.TempDir()
+	os.MkdirAll(filepath.Join(games, "example"), 0o755)
+	os.WriteFile(filepath.Join(games, "example", "game.yaml"), []byte("id: example\nname: Example\n"), 0o644)
+
+	m := &fakeModel{replies: []string{
+		`{"cause":"unsure","evidence":"no error line","confidence":"low","fix":"","checks":[{"type":"search","query":"example server stops at start"}]}`,
+		`{"cause":"the server needs a join password","evidence":"the search result says so","confidence":"high","fix":"give JoinPassword x-generate","checks":[]}`,
+	}}
+	iv := &investigator{m: m, gameName: "Example", gamesDir: games, id: "example", searx: srv.URL, reader: srv.URL}
+	d, err := iv.diagnose("FAILED: server crashed and was restarted 1 time(s)\nError: could not start")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if d.Confidence != "high" || !strings.Contains(d.Fix, "x-generate") {
+		t.Errorf("diagnosis: %+v", d)
+	}
+	second := m.asked[1]
+	last := second[len(second)-1].Content
+	if !strings.Contains(last, "Results of your checks") || !strings.Contains(last, "https://example.com/fix") ||
+		!strings.Contains(last, "refuses to run without a join password") {
+		t.Errorf("check results not sent back:\n%s", last)
+	}
+	// The first question already carried the web evidence for the error line
+	if first := m.asked[0][1].Content; !strings.Contains(first, "What the web says") {
+		t.Errorf("no web evidence in:\n%s", first)
+	}
+}
+
+func TestAttemptLogRulesOutCausesThatChangedNothing(t *testing.T) {
+	iv := &investigator{}
+	iv.record(&diagnosis{Cause: "Not enough memory.", Fix: "requiredMemoryMb 8192"})
+	iv.outcome(false, "FAILED: server crashed", "FAILED: server crashed")
+	if !strings.HasPrefix(iv.log[0].Outcome, "unchanged") {
+		t.Errorf("outcome %q", iv.log[0].Outcome)
+	}
+	if !iv.ruledOut(&diagnosis{Cause: "not enough memory"}) {
+		t.Error("the same cause again should be ruled out")
+	}
+	if iv.ruledOut(&diagnosis{Cause: "the join password is empty"}) {
+		t.Error("a new cause is not ruled out")
+	}
+	if !strings.Contains(iv.logText(), "ruled its cause out") {
+		t.Error("log text should tell the model")
+	}
+	iv.record(&diagnosis{Cause: "empty join password", Fix: "x-generate"})
+	iv.outcome(true, "FAILED: server crashed", "")
+	if !strings.HasPrefix(iv.log[1].Outcome, "fixed") || iv.ruledOut(&diagnosis{Cause: "empty join password"}) {
+		t.Errorf("fixed cause: %+v", iv.log[1])
+	}
+}
+
+func TestDistinctErrorsDropsTimestampsAndRepeats(t *testing.T) {
+	out := "[boot-test] FAILED: container exited\n" +
+		"10/26/2026 01:12:02: Error: failed to bind port 30000\n" +
+		"10/26/2026 01:12:03: Error: failed to bind port 30001\n" +
+		"all fine here\n" +
+		"[12:00:01] Fatal: missing config file"
+	got := distinctErrors(out, 5)
+	if len(got) != 2 || got[0] != "Error: failed to bind port 30000" || got[1] != "Fatal: missing config file" {
+		t.Errorf("got %q", got)
+	}
+}
+
+func TestCopyTemplateWithAppliesTheChange(t *testing.T) {
+	src, dst := filepath.Join(t.TempDir(), "example"), filepath.Join(t.TempDir(), "example")
+	os.MkdirAll(src, 0o755)
+	os.WriteFile(filepath.Join(src, "game.yaml"), []byte("id: example\nname: Example\ndefaultStartupCommand: ./server\ndefaultConfig:\n  JoinPassword: \"\"\n"), 0o644)
+	os.WriteFile(filepath.Join(src, "test.yaml"), []byte("timeoutSec: 600\n"), 0o644)
+	if err := copyTemplateWith(src, dst, checkRequest{Config: map[string]string{"JoinPassword": "secret1"}, Startup: "./server -verbose"}); err != nil {
+		t.Fatal(err)
+	}
+	g, err := loadMap(filepath.Join(dst, "game.yaml"))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if g["defaultStartupCommand"] != "./server -verbose" || g["defaultConfig"].(map[string]any)["JoinPassword"] != "secret1" {
+		t.Errorf("got %v", g)
+	}
+	if _, err := os.Stat(filepath.Join(dst, "test.yaml")); err != nil {
+		t.Error("test.yaml not copied")
+	}
+	if got := describeChange(checkRequest{Config: map[string]string{"JoinPassword": "secret1"}}); got != `JoinPassword="secret1"` {
+		t.Errorf("describe: %s", got)
+	}
+}
+
+func TestFixRequestCarriesTheDiagnosis(t *testing.T) {
+	got := strings.Join(fixRequest("FAILED: server crashed\nlog...", &diagnosis{Cause: "c", Evidence: "e", Fix: "f"}), "")
+	for _, want := range []string{"FAILED: server crashed", "- cause: c", "- fix: f", "Apply exactly this fix"} {
+		if !strings.Contains(got, want) {
+			t.Errorf("missing %q in %s", want, got)
+		}
+	}
+}