fix(llm): address PR review feedback

- Gate dashboard refresh key on insightsWanted() (checks both
  insightsEnabled and llmClient) instead of insightsEnabled alone
- Validate context_length >= 0 during config load with clear error
- Update DataDump() doc comment to reflect that entity IDs are now
  included

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: cpcloud

internal/app/dashboard_test.go

@@ -11,10 +11,20 @@ import (
 
 	"github.com/charmbracelet/lipgloss"
 	"github.com/cpcloud/micasa/internal/data"
+	"github.com/cpcloud/micasa/internal/llm"
 	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 )
 
+// testDummyLLMClient returns a non-nil LLM client for tests that need
+// insightsWanted() to return true. Not connected to a real server.
+func testDummyLLMClient(t *testing.T) *llm.Client {
+	t.Helper()
+	c, err := llm.NewClient("llamacpp", "http://127.0.0.1:1", "test", "", time.Second, 0)
+	require.NoError(t, err)
+	return c
+}
+
 // nonEmptyDashboard returns a minimal dashboardData that is not empty,
 // for tests that just need the dashboard overlay to render.
 func nonEmptyDashboard() dashboardData {
@@ -1415,6 +1425,7 @@ func TestInsightsRefreshKey_MarksStale(t *testing.T) {
 	t.Parallel()
 	m := newTestModel(t)
 	m.insightsEnabled = true
+	m.llmClient = testDummyLLMClient(t)
 	m.width = 120
 	m.height = 40
 

internal/app/model.go

@@ -679,7 +679,7 @@ func (m *Model) handleDashboardKeys(key tea.KeyMsg) (tea.Cmd, bool) {
 		// Block column movement on dashboard.
 		return nil, true
 	case keyR:
-		if m.insightsEnabled {
+		if m.insightsWanted() {
 			cmd := m.refreshInsights()
 			m.setStatusInfo("refreshing insights")
 			return cmd, true

internal/config/config.go

@@ -504,6 +504,14 @@ func LoadFromPath(path string) (Config, error) {
 		)
 	}
 
+	// Validate context_length.
+	if cfg.LLM.ContextLength < 0 {
+		return cfg, fmt.Errorf(
+			"llm.context_length: must be >= 0, got %d",
+			cfg.LLM.ContextLength,
+		)
+	}
+
 	// Validate timeouts.
 	if cfg.LLM.Timeout != "" {
 		d, err := time.ParseDuration(cfg.LLM.Timeout)

internal/config/config_test.go

@@ -861,6 +861,17 @@ context_length = 65536
 	assert.Equal(t, 65536, cfg.LLM.ExtractionConfig().ContextLength)
 }
 
+func TestContextLengthRejectsNegative(t *testing.T) {
+	path := writeConfig(t, `[llm]
+provider = "ollama"
+model = "qwen3"
+context_length = -1
+`)
+	_, err := LoadFromPath(path)
+	require.Error(t, err)
+	assert.Contains(t, err.Error(), "context_length")
+}
+
 func TestExtractionConfigInheritsBase(t *testing.T) {
 	path := writeConfig(t, `[llm]
 provider = "ollama"

internal/data/query.go

@@ -261,8 +261,8 @@ func firstWord(s string) string {
 //
 // The output is optimized for small LLMs: null/empty values are omitted,
 // money columns (ending in "_ct") are formatted as dollars, and internal
-// columns (id, created_at, updated_at, deleted_at) are excluded to reduce
-// noise.
+// columns (created_at, updated_at, deleted_at) are excluded to reduce
+// noise. Entity IDs are included so the LLM can reference specific records.
 func (s *Store) DataDump() string {
 	names, err := s.TableNames()
 	if err != nil {