Merge pull request #1 from thepeterstone/claude/local-oss-model-agents-MEBqj

Local OSS models as a third runner (epic)

1 files changed, 241 insertions, 0 deletions

diff --git a/internal/executor/summary_synth_test.go b/internal/executor/summary_synth_test.go
new file mode 100644
index 0000000..7ad396d
--- /dev/null
+++ b/internal/executor/summary_synth_test.go
@@ -0,0 +1,241 @@
+package executor
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"net/http/httptest"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync/atomic"
+	"testing"
+
+	"github.com/thepeterstone/claudomator/internal/llm"
+	"github.com/thepeterstone/claudomator/internal/storage"
+)
+
+func writeStreamLog(t *testing.T, lines []string) string {
+	t.Helper()
+	dir := t.TempDir()
+	path := filepath.Join(dir, "stdout.log")
+	var sb strings.Builder
+	for _, l := range lines {
+		sb.WriteString(l)
+		sb.WriteString("\n")
+	}
+	if err := os.WriteFile(path, []byte(sb.String()), 0600); err != nil {
+		t.Fatal(err)
+	}
+	return path
+}
+
+func TestSynthesizeSummary_NilClient(t *testing.T) {
+	got := synthesizeSummary(context.Background(), nil, "/some/path")
+	if got != "" {
+		t.Errorf("nil client: want empty, got %q", got)
+	}
+}
+
+func TestSynthesizeSummary_EmptyPath(t *testing.T) {
+	c := &llm.Client{Endpoint: "http://unused", Model: "x"}
+	got := synthesizeSummary(context.Background(), c, "")
+	if got != "" {
+		t.Errorf("empty path: want empty, got %q", got)
+	}
+}
+
+func TestSynthesizeSummary_MissingFile(t *testing.T) {
+	c := &llm.Client{Endpoint: "http://unused", Model: "x"}
+	got := synthesizeSummary(context.Background(), c, "/nonexistent/file.log")
+	if got != "" {
+		t.Errorf("missing file: want empty, got %q", got)
+	}
+}
+
+func TestSynthesizeSummary_EmptyAssistantContent(t *testing.T) {
+	// Log contains only system/result events — no assistant text. The function
+	// should short-circuit without calling the LLM.
+	path := writeStreamLog(t, []string{
+		`{"type":"system","subtype":"init"}`,
+		`{"type":"result","subtype":"success","total_cost_usd":0}`,
+	})
+
+	var calls int32
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&calls, 1)
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintln(w, `{"choices":[{"message":{"content":"should not be returned"},"finish_reason":"stop"}],"usage":{}}`)
+	}))
+	defer srv.Close()
+	c := &llm.Client{Endpoint: srv.URL + "/v1", Model: "x"}
+
+	got := synthesizeSummary(context.Background(), c, path)
+	if got != "" {
+		t.Errorf("empty content: want empty, got %q", got)
+	}
+	if atomic.LoadInt32(&calls) != 0 {
+		t.Errorf("LLM should not be called for empty assistant content")
+	}
+}
+
+func TestSynthesizeSummary_LLMSuccess(t *testing.T) {
+	path := writeStreamLog(t, []string{
+		`{"type":"assistant","message":{"content":[{"type":"text","text":"Ran the tests."}]}}`,
+		`{"type":"assistant","message":{"content":[{"type":"text","text":"Fixed the import."}]}}`,
+		`{"type":"result","subtype":"success"}`,
+	})
+
+	var capturedUser string
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		var body struct {
+			Messages []struct {
+				Role, Content string
+			} `json:"messages"`
+		}
+		json.NewDecoder(r.Body).Decode(&body)
+		for _, m := range body.Messages {
+			if m.Role == "user" {
+				capturedUser = m.Content
+			}
+		}
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintln(w, `{"choices":[{"message":{"content":"  Agent ran tests and fixed an import.  "},"finish_reason":"stop"}],"usage":{}}`)
+	}))
+	defer srv.Close()
+	c := &llm.Client{Endpoint: srv.URL + "/v1", Model: "x"}
+
+	got := synthesizeSummary(context.Background(), c, path)
+	if got != "Agent ran tests and fixed an import." {
+		t.Errorf("summary: got %q", got)
+	}
+	if !strings.Contains(capturedUser, "Ran the tests.") {
+		t.Errorf("user prompt missing first assistant text; got: %s", capturedUser)
+	}
+	if !strings.Contains(capturedUser, "Fixed the import.") {
+		t.Errorf("user prompt missing second assistant text; got: %s", capturedUser)
+	}
+}
+
+// TestPool_HandleRunResult_LLMSummaryFallback verifies the Pool falls back to
+// LLM-synthesized summary when extractSummary returns empty.
+func TestPool_HandleRunResult_LLMSummaryFallback(t *testing.T) {
+	// stdout has assistant text but no "## Summary" heading.
+	stdoutPath := writeStreamLog(t, []string{
+		`{"type":"assistant","message":{"content":[{"type":"text","text":"Did the work without writing a summary section."}]}}`,
+	})
+
+	llmSrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintln(w, `{"choices":[{"message":{"content":"Synthesized summary."},"finish_reason":"stop"}],"usage":{}}`)
+	}))
+	defer llmSrv.Close()
+
+	store := newMinimalMockStore()
+	pool := newPoolWithMockStore(store)
+	pool.LLM = &llm.Client{Endpoint: llmSrv.URL + "/v1", Model: "x"}
+
+	tk := makeTask("synth-summary")
+	store.tasks[tk.ID] = tk
+	exec := &storage.Execution{ID: "e-synth", TaskID: tk.ID, Status: "RUNNING", StdoutPath: stdoutPath}
+
+	pool.handleRunResult(context.Background(), tk, exec, nil, "claude")
+
+	id, summary, ok := store.lastSummaryUpdate()
+	if !ok {
+		t.Fatalf("expected UpdateTaskSummary to be called")
+	}
+	if id != tk.ID {
+		t.Errorf("summary recorded for wrong task: %q", id)
+	}
+	if summary != "Synthesized summary." {
+		t.Errorf("summary: got %q", summary)
+	}
+
+	// Drain the result channel so the test exits cleanly.
+	<-pool.resultCh
+}
+
+// TestPool_HandleRunResult_ExtractSummaryWins verifies the LLM is NOT called
+// when the agent already wrote a "## Summary" section.
+func TestPool_HandleRunResult_ExtractSummaryWins(t *testing.T) {
+	stdoutPath := writeStreamLog(t, []string{
+		`{"type":"assistant","message":{"content":[{"type":"text","text":"## Summary\nAgent wrote its own summary."}]}}`,
+	})
+
+	var llmCalls int32
+	llmSrv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		atomic.AddInt32(&llmCalls, 1)
+		w.Header().Set("Content-Type", "application/json")
+		fmt.Fprintln(w, `{"choices":[{"message":{"content":"should not be used"},"finish_reason":"stop"}],"usage":{}}`)
+	}))
+	defer llmSrv.Close()
+
+	store := newMinimalMockStore()
+	pool := newPoolWithMockStore(store)
+	pool.LLM = &llm.Client{Endpoint: llmSrv.URL + "/v1", Model: "x"}
+
+	tk := makeTask("agent-summary")
+	store.tasks[tk.ID] = tk
+	exec := &storage.Execution{ID: "e-agent", TaskID: tk.ID, Status: "RUNNING", StdoutPath: stdoutPath}
+
+	pool.handleRunResult(context.Background(), tk, exec, nil, "claude")
+
+	if got := atomic.LoadInt32(&llmCalls); got != 0 {
+		t.Errorf("LLM should not be called when ## Summary is present; got %d calls", got)
+	}
+	_, summary, ok := store.lastSummaryUpdate()
+	if !ok {
+		t.Fatalf("expected UpdateTaskSummary")
+	}
+	if summary != "Agent wrote its own summary." {
+		t.Errorf("summary: got %q (want extractSummary output)", summary)
+	}
+	<-pool.resultCh
+}
+
+func TestSynthesizeSummary_LLMFailure_ReturnsEmpty(t *testing.T) {
+	path := writeStreamLog(t, []string{
+		`{"type":"assistant","message":{"content":[{"type":"text","text":"Did something."}]}}`,
+	})
+	srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		http.Error(w, "boom", http.StatusInternalServerError)
+	}))
+	defer srv.Close()
+	c := &llm.Client{Endpoint: srv.URL + "/v1", Model: "x"}
+
+	got := synthesizeSummary(context.Background(), c, path)
+	if got != "" {
+		t.Errorf("LLM failure: want empty, got %q", got)
+	}
+}
+
+// TestReadAssistantTextTail_TailingLargeFile verifies the seek-to-tail
+// behavior drops early content but keeps later assistant text.
+func TestReadAssistantTextTail_TailingLargeFile(t *testing.T) {
+	dir := t.TempDir()
+	path := filepath.Join(dir, "stdout.log")
+	f, err := os.Create(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Write a ton of garbage assistant lines, then a final marker.
+	for i := 0; i < 500; i++ {
+		fmt.Fprintf(f, `{"type":"assistant","message":{"content":[{"type":"text","text":"filler line that should be in the early part of a large file %04d"}]}}`+"\n", i)
+	}
+	fmt.Fprintln(f, `{"type":"assistant","message":{"content":[{"type":"text","text":"FINAL_MARKER_LINE"}]}}`)
+	f.Close()
+
+	got := readAssistantTextTail(path, 4*1024) // 4 KB cap
+	if !strings.Contains(got, "FINAL_MARKER_LINE") {
+		t.Errorf("tail should contain final line; got: %s", got)
+	}
+	if strings.Contains(got, "filler line that should be in the early part of a large file 0000") {
+		end := 200
+		if len(got) < end {
+			end = len(got)
+		}
+		t.Errorf("tail should NOT contain very-early line; got first 200 chars: %s", got[:end])
+	}
+}