fix: add engine_counts to logs summary using aw_info.json, preventing false-positive Copilot classification (#26359)

Author: Copilot

.github/workflows/audit-workflows.md

@@ -66,6 +66,10 @@ Use the agentic-workflows MCP tool `logs` with parameters:
 Output is saved to: /tmp/gh-aw/aw-mcp/logs
 ```
 
+**Engine Classification**: Use `summary.engine_counts` from the `logs` tool output to report engine usage. Each run also has an `agent` field (e.g., `"copilot"`, `"claude"`, `"codex"`). Both are derived from the `engine_id` field in `aw_info.json`, which is the authoritative source for engine type.
+
+**IMPORTANT**: Do NOT infer engine type by scanning `.lock.yml` files. Lock files contain the word `copilot` in allowed-domains lists and workflow source paths regardless of which engine the workflow uses, causing false positives.
+
 **Analyze**: Review logs for:
 - Missing tools (patterns, frequency, legitimacy)
 - Errors (tool execution, MCP failures, auth, timeouts, resources)

pkg/cli/logs_report.go

@@ -67,6 +67,11 @@ type LogsSummary struct {
 	TotalEpisodes          int     `json:"total_episodes" console:"header:Total Episodes"`
 	HighConfidenceEpisodes int     `json:"high_confidence_episodes" console:"header:High Confidence Episodes"`
 	TotalGitHubAPICalls    int     `json:"total_github_api_calls,omitempty" console:"header:Total GitHub API Calls,format:number,omitempty"`
+	// EngineCounts maps engine_id (from aw_info.json) to the number of runs using that engine.
+	// Use this field to accurately classify engine types — do NOT infer engines by scanning
+	// lock files, which contain the word "copilot" in allowed-domains and workflow-source paths
+	// regardless of which engine the workflow actually uses.
+	EngineCounts map[string]int `json:"engine_counts,omitempty" console:"-"`
 }
 
 // RunData contains information about a single workflow run
@@ -134,6 +139,11 @@ func buildLogsData(processedRuns []ProcessedRun, outputDir string, continuation
 	var totalMissingData int
 	var totalSafeItems int
 	var totalGitHubAPICalls int
+	// engineCounts tracks the number of runs per engine_id, sourced from aw_info.json.
+	// This is the authoritative engine classification — do not infer engine type from
+	// lock file contents, which contain "copilot" in allowed-domains and source paths
+	// regardless of which engine the workflow uses.
+	engineCounts := make(map[string]int)
 
 	// Build runs data
 	// Initialize as empty slice to ensure JSON marshals to [] instead of null
@@ -175,6 +185,10 @@ func buildLogsData(processedRuns []ProcessedRun, outputDir string, continuation
 		if awContext == nil {
 			awContext = pr.AwContext
 		}
+		// Accumulate engine counts from aw_info.json data (authoritative source).
+		if agentID != "" {
+			engineCounts[agentID]++
+		}
 
 		comparison := buildAuditComparisonForProcessedRuns(pr, processedRuns)
 
@@ -255,6 +269,9 @@ func buildLogsData(processedRuns []ProcessedRun, outputDir string, continuation
 		TotalSafeItems:       totalSafeItems,
 		TotalGitHubAPICalls:  totalGitHubAPICalls,
 	}
+	if len(engineCounts) > 0 {
+		summary.EngineCounts = engineCounts
+	}
 
 	episodes, edges := buildEpisodeData(runs, processedRuns)
 	for _, episode := range episodes {

pkg/cli/logs_report_test.go

@@ -3,6 +3,8 @@
 package cli
 
 import (
+	"os"
+	"path/filepath"
 	"testing"
 	"time"
 )
@@ -908,3 +910,53 @@ func TestDeriveRunClassification(t *testing.T) {
 		})
 	}
 }
+
+// TestBuildLogsDataEngineCountsFromAwInfo verifies that engine_counts in the summary
+// is populated from aw_info.json data (the authoritative engine source), not from
+// lock file string matching.
+func TestBuildLogsDataEngineCountsFromAwInfo(t *testing.T) {
+	createRunDir := func(engineID string) string {
+		dir := t.TempDir()
+		awInfo := `{"engine_id":"` + engineID + `","engine_name":"Test","workflow_name":"test","created_at":"2024-01-01T00:00:00Z"}`
+		if err := os.WriteFile(filepath.Join(dir, "aw_info.json"), []byte(awInfo), 0600); err != nil {
+			t.Fatalf("Failed to write aw_info.json: %v", err)
+		}
+		return dir
+	}
+
+	claudeDir := createRunDir("claude")
+	claudeDir2 := createRunDir("claude")
+	copilotDir := createRunDir("copilot")
+
+	processedRuns := []ProcessedRun{
+		{Run: WorkflowRun{DatabaseID: 1, WorkflowName: "wf-claude-1", LogsPath: claudeDir}},
+		{Run: WorkflowRun{DatabaseID: 2, WorkflowName: "wf-claude-2", LogsPath: claudeDir2}},
+		{Run: WorkflowRun{DatabaseID: 3, WorkflowName: "wf-copilot", LogsPath: copilotDir}},
+	}
+
+	data := buildLogsData(processedRuns, "/tmp/logs", nil)
+
+	if data.Summary.EngineCounts == nil {
+		t.Fatal("EngineCounts should not be nil when runs have aw_info.json")
+	}
+	if got := data.Summary.EngineCounts["claude"]; got != 2 {
+		t.Errorf("Expected 2 claude runs, got %d", got)
+	}
+	if got := data.Summary.EngineCounts["copilot"]; got != 1 {
+		t.Errorf("Expected 1 copilot run, got %d", got)
+	}
+	// Verify individual RunData.Agent fields also reflect the engine from aw_info.json
+	agentsByID := make(map[int64]string)
+	for _, run := range data.Runs {
+		agentsByID[run.DatabaseID] = run.Agent
+	}
+	if agentsByID[1] != "claude" {
+		t.Errorf("Run 1: expected agent=claude, got %q", agentsByID[1])
+	}
+	if agentsByID[2] != "claude" {
+		t.Errorf("Run 2: expected agent=claude, got %q", agentsByID[2])
+	}
+	if agentsByID[3] != "copilot" {
+		t.Errorf("Run 3: expected agent=copilot, got %q", agentsByID[3])
+	}
+}