fix: subagent file modifications invisible to checkpoint detection (ENT-297)

cmd/entire/cli/agent/claudecode/transcript.go

@@ -427,3 +427,49 @@ func CalculateTotalTokenUsage(transcriptPath string, startLine int, subagentsDir
 
 	return mainUsage, nil
 }
+
+// ExtractAllModifiedFiles extracts files modified by both the main agent and
+// any subagents spawned via the Task tool. It parses the main transcript from
+// startLine, collects modified files from the main agent, then reads each
+// subagent's transcript from subagentsDir to collect their modified files too.
+// The result is a deduplicated list of all modified file paths.
+func ExtractAllModifiedFiles(transcriptPath string, startLine int, subagentsDir string) ([]string, error) {
+	if transcriptPath == "" {
+		return nil, nil
+	}
+
+	// Parse main transcript once
+	transcript, err := parseTranscriptFromLine(transcriptPath, startLine)
+	if err != nil {
+		return nil, fmt.Errorf("failed to parse transcript: %w", err)
+	}
+
+	// Collect modified files from main agent
+	fileSet := make(map[string]bool)
+	var files []string
+	for _, f := range ExtractModifiedFiles(transcript) {
+		if !fileSet[f] {
+			fileSet[f] = true
+			files = append(files, f)
+		}
+	}
+
+	// Find spawned subagents and collect their modified files
+	agentIDs := ExtractSpawnedAgentIDs(transcript)
+	for agentID := range agentIDs {
+		agentPath := filepath.Join(subagentsDir, fmt.Sprintf("agent-%s.jsonl", agentID))
+		agentTranscript, err := parseTranscriptFromLine(agentPath, 0)
+		if err != nil {
+			// Subagent transcript may not exist yet or may have been cleaned up
+			continue
+		}
+		for _, f := range ExtractModifiedFiles(agentTranscript) {
+			if !fileSet[f] {
+				fileSet[f] = true
+				files = append(files, f)
+			}
+		}
+	}
+
+	return files, nil
+}

cmd/entire/cli/agent/claudecode/transcript_test.go

@@ -3,6 +3,7 @@ package claudecode
 import (
 	"encoding/json"
 	"os"
+	"strings"
 	"testing"
 
 	"github.com/entireio/cli/cmd/entire/cli/transcript"
@@ -618,3 +619,257 @@ func TestCalculateTotalTokenUsage_PerCheckpoint(t *testing.T) {
 		t.Errorf("From line 4: got APICallCount=%d, want 1", usage3.APICallCount)
 	}
 }
+
+// writeJSONLFile is a test helper that writes JSONL transcript lines to a file.
+func writeJSONLFile(t *testing.T, path string, lines ...string) {
+	t.Helper()
+	var buf strings.Builder
+	for _, line := range lines {
+		buf.WriteString(line)
+		buf.WriteByte('\n')
+	}
+	if err := os.WriteFile(path, []byte(buf.String()), 0o600); err != nil {
+		t.Fatalf("failed to write JSONL file %s: %v", path, err)
+	}
+}
+
+// makeWriteToolLine returns a JSONL assistant line with a Write tool_use for the given file.
+func makeWriteToolLine(t *testing.T, uuid, filePath string) string {
+	t.Helper()
+	data := mustMarshal(t, map[string]interface{}{
+		"content": []map[string]interface{}{
+			{
+				"type":  "tool_use",
+				"id":    "toolu_" + uuid,
+				"name":  "Write",
+				"input": map[string]string{"file_path": filePath},
+			},
+		},
+	})
+	line := mustMarshal(t, map[string]interface{}{
+		"type":    "assistant",
+		"uuid":    uuid,
+		"message": json.RawMessage(data),
+	})
+	return string(line)
+}
+
+// makeEditToolLine returns a JSONL assistant line with an Edit tool_use for the given file.
+func makeEditToolLine(t *testing.T, uuid, filePath string) string {
+	t.Helper()
+	data := mustMarshal(t, map[string]interface{}{
+		"content": []map[string]interface{}{
+			{
+				"type":  "tool_use",
+				"id":    "toolu_" + uuid,
+				"name":  "Edit",
+				"input": map[string]string{"file_path": filePath},
+			},
+		},
+	})
+	line := mustMarshal(t, map[string]interface{}{
+		"type":    "assistant",
+		"uuid":    uuid,
+		"message": json.RawMessage(data),
+	})
+	return string(line)
+}
+
+// makeTaskToolUseLine returns a JSONL assistant line with a Task tool_use (spawning a subagent).
+func makeTaskToolUseLine(t *testing.T, uuid, toolUseID string) string {
+	t.Helper()
+	data := mustMarshal(t, map[string]interface{}{
+		"content": []map[string]interface{}{
+			{
+				"type":  "tool_use",
+				"id":    toolUseID,
+				"name":  "Task",
+				"input": map[string]string{"prompt": "do something"},
+			},
+		},
+	})
+	line := mustMarshal(t, map[string]interface{}{
+		"type":    "assistant",
+		"uuid":    uuid,
+		"message": json.RawMessage(data),
+	})
+	return string(line)
+}
+
+// makeTaskResultLine returns a JSONL user line with a tool_result containing agentId.
+func makeTaskResultLine(t *testing.T, uuid, toolUseID, agentID string) string {
+	t.Helper()
+	data := mustMarshal(t, map[string]interface{}{
+		"content": []map[string]interface{}{
+			{
+				"type":        "tool_result",
+				"tool_use_id": toolUseID,
+				"content":     "agentId: " + agentID,
+			},
+		},
+	})
+	line := mustMarshal(t, map[string]interface{}{
+		"type":    "user",
+		"uuid":    uuid,
+		"message": json.RawMessage(data),
+	})
+	return string(line)
+}
+
+func TestExtractAllModifiedFiles_IncludesSubagentFiles(t *testing.T) {
+	t.Parallel()
+
+	tmpDir := t.TempDir()
+	transcriptPath := tmpDir + "/transcript.jsonl"
+	subagentsDir := tmpDir + "/tasks/toolu_task1"
+
+	if err := os.MkdirAll(subagentsDir, 0o755); err != nil {
+		t.Fatalf("failed to create subagents dir: %v", err)
+	}
+
+	// Main transcript: Write to main.go + Task call spawning subagent "sub1"
+	writeJSONLFile(t, transcriptPath,
+		makeWriteToolLine(t, "a1", "/repo/main.go"),
+		makeTaskToolUseLine(t, "a2", "toolu_task1"),
+		makeTaskResultLine(t, "u1", "toolu_task1", "sub1"),
+	)
+
+	// Subagent transcript: Write to helper.go + Edit to utils.go
+	writeJSONLFile(t, subagentsDir+"/agent-sub1.jsonl",
+		makeWriteToolLine(t, "sa1", "/repo/helper.go"),
+		makeEditToolLine(t, "sa2", "/repo/utils.go"),
+	)
+
+	files, err := ExtractAllModifiedFiles(transcriptPath, 0, subagentsDir)
+	if err != nil {
+		t.Fatalf("ExtractAllModifiedFiles() error: %v", err)
+	}
+
+	if len(files) != 3 {
+		t.Errorf("expected 3 files, got %d: %v", len(files), files)
+	}
+
+	wantFiles := map[string]bool{
+		"/repo/main.go":   true,
+		"/repo/helper.go": true,
+		"/repo/utils.go":  true,
+	}
+	for _, f := range files {
+		if !wantFiles[f] {
+			t.Errorf("unexpected file %q in result", f)
+		}
+		delete(wantFiles, f)
+	}
+	for f := range wantFiles {
+		t.Errorf("missing expected file %q", f)
+	}
+}
+
+func TestExtractAllModifiedFiles_DeduplicatesAcrossAgents(t *testing.T) {
+	t.Parallel()
+
+	tmpDir := t.TempDir()
+	transcriptPath := tmpDir + "/transcript.jsonl"
+	subagentsDir := tmpDir + "/tasks/toolu_task1"
+
+	if err := os.MkdirAll(subagentsDir, 0o755); err != nil {
+		t.Fatalf("failed to create subagents dir: %v", err)
+	}
+
+	// Main transcript: Write to shared.go + Task call
+	writeJSONLFile(t, transcriptPath,
+		makeWriteToolLine(t, "a1", "/repo/shared.go"),
+		makeTaskToolUseLine(t, "a2", "toolu_task1"),
+		makeTaskResultLine(t, "u1", "toolu_task1", "sub1"),
+	)
+
+	// Subagent transcript: Also modifies shared.go (same file as main)
+	writeJSONLFile(t, subagentsDir+"/agent-sub1.jsonl",
+		makeEditToolLine(t, "sa1", "/repo/shared.go"),
+	)
+
+	files, err := ExtractAllModifiedFiles(transcriptPath, 0, subagentsDir)
+	if err != nil {
+		t.Fatalf("ExtractAllModifiedFiles() error: %v", err)
+	}
+
+	if len(files) != 1 {
+		t.Errorf("expected 1 file (deduplicated), got %d: %v", len(files), files)
+	}
+	if len(files) > 0 && files[0] != "/repo/shared.go" {
+		t.Errorf("expected /repo/shared.go, got %q", files[0])
+	}
+}
+
+func TestExtractAllModifiedFiles_NoSubagents(t *testing.T) {
+	t.Parallel()
+
+	tmpDir := t.TempDir()
+	transcriptPath := tmpDir + "/transcript.jsonl"
+
+	// Main transcript: Write to a file, no Task calls
+	writeJSONLFile(t, transcriptPath,
+		makeWriteToolLine(t, "a1", "/repo/solo.go"),
+	)
+
+	files, err := ExtractAllModifiedFiles(transcriptPath, 0, tmpDir+"/nonexistent")
+	if err != nil {
+		t.Fatalf("ExtractAllModifiedFiles() error: %v", err)
+	}
+
+	if len(files) != 1 {
+		t.Errorf("expected 1 file, got %d: %v", len(files), files)
+	}
+	if len(files) > 0 && files[0] != "/repo/solo.go" {
+		t.Errorf("expected /repo/solo.go, got %q", files[0])
+	}
+}
+
+func TestExtractAllModifiedFiles_SubagentOnlyChanges(t *testing.T) {
+	t.Parallel()
+
+	tmpDir := t.TempDir()
+	transcriptPath := tmpDir + "/transcript.jsonl"
+	subagentsDir := tmpDir + "/tasks/toolu_task1"
+
+	if err := os.MkdirAll(subagentsDir, 0o755); err != nil {
+		t.Fatalf("failed to create subagents dir: %v", err)
+	}
+
+	// Main transcript: ONLY a Task call, no direct file modifications
+	// This is the key bug scenario - if we only look at the main transcript,
+	// we miss all the subagent's file changes entirely.
+	writeJSONLFile(t, transcriptPath,
+		makeTaskToolUseLine(t, "a1", "toolu_task1"),
+		makeTaskResultLine(t, "u1", "toolu_task1", "sub1"),
+	)
+
+	// Subagent transcript: Write to two files
+	writeJSONLFile(t, subagentsDir+"/agent-sub1.jsonl",
+		makeWriteToolLine(t, "sa1", "/repo/subagent_file1.go"),
+		makeWriteToolLine(t, "sa2", "/repo/subagent_file2.go"),
+	)
+
+	files, err := ExtractAllModifiedFiles(transcriptPath, 0, subagentsDir)
+	if err != nil {
+		t.Fatalf("ExtractAllModifiedFiles() error: %v", err)
+	}
+
+	if len(files) != 2 {
+		t.Errorf("expected 2 files from subagent, got %d: %v", len(files), files)
+	}
+
+	wantFiles := map[string]bool{
+		"/repo/subagent_file1.go": true,
+		"/repo/subagent_file2.go": true,
+	}
+	for _, f := range files {
+		if !wantFiles[f] {
+			t.Errorf("unexpected file %q in result", f)
+		}
+		delete(wantFiles, f)
+	}
+	for f := range wantFiles {
+		t.Errorf("missing expected file %q", f)
+	}
+}

cmd/entire/cli/hooks_claudecode_handlers.go

@@ -221,8 +221,15 @@ func commitWithMetadata() error { //nolint:maintidx // already present in codeba
 	}
 	fmt.Fprintf(os.Stderr, "Extracted summary to: %s\n", sessionDir+"/"+paths.SummaryFileName)
 
-	// Get modified files from transcript
-	modifiedFiles := extractModifiedFiles(transcript)
+	// Get modified files from transcript (main agent + subagents).
+	// Subagent transcripts live in <transcriptDir>/<modelSessionID>/subagents/
+	subagentsDir := filepath.Join(filepath.Dir(transcriptPath), input.SessionID, "subagents")
+	modifiedFiles, subagentErr := claudecode.ExtractAllModifiedFiles(transcriptPath, transcriptOffset, subagentsDir)
+	if subagentErr != nil {
+		fmt.Fprintf(os.Stderr, "Warning: failed to extract modified files with subagents: %v\n", subagentErr)
+		// Fall back to main transcript only
+		modifiedFiles = extractModifiedFiles(transcript)
+	}
 
 	// Generate commit message from last user prompt
 	lastPrompt := ""
@@ -321,11 +328,9 @@ func commitWithMetadata() error { //nolint:maintidx // already present in codeba
 	// Calculate token usage for this checkpoint (Claude Code specific)
 	var tokenUsage *agent.TokenUsage
 	if transcriptPath != "" {
-		// Subagents are stored in a subagents/ directory next to the main transcript
-		subagentsDir := filepath.Join(filepath.Dir(transcriptPath), sessionID, "subagents")
-		usage, err := claudecode.CalculateTotalTokenUsage(transcriptPath, transcriptLinesAtStart, subagentsDir)
-		if err != nil {
-			fmt.Fprintf(os.Stderr, "Warning: failed to calculate token usage: %v\n", err)
+		usage, tokenErr := claudecode.CalculateTotalTokenUsage(transcriptPath, transcriptLinesAtStart, subagentsDir)
+		if tokenErr != nil {
+			fmt.Fprintf(os.Stderr, "Warning: failed to calculate token usage: %v\n", tokenErr)
 		} else {
 			tokenUsage = usage
 		}