feat: Add Google Gemini CLI as experimental agentic engine

.github/workflows/smoke-gemini.md

@@ -0,0 +1,67 @@
+---
+description: Smoke test workflow that validates Gemini engine functionality twice daily
+on:
+  schedule: every 12h
+  workflow_dispatch:
+  pull_request:
+    types: [labeled]
+    names: ["smoke"]
+permissions:
+  contents: read
+  issues: read
+  pull-requests: read
+name: Smoke Gemini
+engine: gemini
+strict: true
+imports:
+  - shared/gh.md
+  - shared/reporting.md
+network:
+  allowed:
+    - defaults
+    - github
+tools:
+  cache-memory: true
+  github:
+    toolsets: [repos, pull_requests]
+  edit:
+  bash:
+    - "*"
+safe-outputs:
+    add-comment:
+      hide-older-comments: true
+      max: 2
+    create-issue:
+      expires: 2h
+      close-older-issues: true
+    add-labels:
+      allowed: [smoke-gemini]
+    messages:
+      footer: "> ✨ *[{workflow_name}]({run_url}) — Powered by Gemini*"
+      run-started: "✨ Gemini awakens... [{workflow_name}]({run_url}) begins its journey on this {event_type}..."
+      run-success: "🚀 [{workflow_name}]({run_url}) **MISSION COMPLETE!** Gemini has spoken. ✨"
+      run-failure: "⚠️ [{workflow_name}]({run_url}) {status}. Gemini encountered unexpected challenges..."
+timeout-minutes: 10
+---
+
+# Smoke Test: Gemini Engine Validation
+
+**CRITICAL EFFICIENCY REQUIREMENTS:**
+- Keep ALL outputs extremely short and concise. Use single-line responses.
+- NO verbose explanations or unnecessary context.
+- Minimize file reading - only read what is absolutely necessary for the task.
+
+## Test Requirements
+
+1. **GitHub MCP Testing**: Use GitHub MCP tools to fetch details of exactly 2 merged pull requests from ${{ github.repository }} (title and number only)
+2. **File Writing Testing**: Create a test file `/tmp/gh-aw/agent/smoke-test-gemini-${{ github.run_id }}.txt` with content "Smoke test passed for Gemini at $(date)" (create the directory if it doesn't exist)
+3. **Bash Tool Testing**: Execute bash commands to verify file creation was successful (use `cat` to read the file back)
+4. **Build gh-aw**: Run `GOCACHE=/tmp/go-cache GOMODCACHE=/tmp/go-mod make build` to verify the agent can successfully build the gh-aw project. If the command fails, mark this test as ❌ and report the failure.
+
+## Output
+
+Add a **very brief** comment (max 5-10 lines) to the current pull request with:
+- ✅ or ❌ for each test result
+- Overall status: PASS or FAIL
+
+If all tests pass, use the `add_labels` safe-output tool to add the label `smoke-gemini` to the pull request.

cmd/gh-aw/main_entry_test.go

@@ -40,6 +40,11 @@ func TestValidateEngine(t *testing.T) {
 			engine:    "copilot",
 			expectErr: false,
 		},
+		{
+			name:      "valid gemini engine",
+			engine:    "gemini",
+			expectErr: false,
+		},
 		{
 			name:       "invalid engine",
 			engine:     "gpt4",
@@ -82,11 +87,11 @@ func TestValidateEngine(t *testing.T) {
 					return
 				}
 
-				// Check that error message contains the expected format
-				// Error may include "Did you mean" suggestions, so we check if it starts with the base message
-				expectedMsg := fmt.Sprintf("invalid engine value '%s'. Must be 'claude', 'codex', or 'copilot'", tt.engine)
-				if tt.errMessage != "" && !strings.HasPrefix(err.Error(), expectedMsg) {
-					t.Errorf("validateEngine(%q) error message = %v, want to start with %v", tt.engine, err.Error(), expectedMsg)
+				// Check that error message contains the expected format.
+				// The engine list is dynamic, so only check the prefix.
+				expectedPrefix := fmt.Sprintf("invalid engine value '%s'. Must be", tt.engine)
+				if tt.errMessage != "" && !strings.HasPrefix(err.Error(), expectedPrefix) {
+					t.Errorf("validateEngine(%q) error message = %v, want to start with %v", tt.engine, err.Error(), expectedPrefix)
 				}
 			} else {
 				if err != nil {

pkg/cli/completions_test.go

@@ -228,7 +228,7 @@ func TestValidEngineNames(t *testing.T) {
 	assert.NotEmpty(t, engines, "Engine names list should not be empty")
 
 	// Verify expected engines are present
-	expectedEngines := []string{"copilot", "claude", "codex"}
+	expectedEngines := []string{"copilot", "claude", "codex", "gemini"}
 	for _, expected := range expectedEngines {
 		assert.Contains(t, engines, expected, "Expected engine '%s' to be in the list", expected)
 	}
@@ -245,7 +245,7 @@ func TestCompleteEngineNames(t *testing.T) {
 		{
 			name:       "empty prefix returns all engines",
 			toComplete: "",
-			wantLen:    3, // copilot, claude, codex
+			wantLen:    4, // copilot, claude, codex, gemini
 		},
 		{
 			name:       "c prefix returns claude, codex, copilot",
@@ -753,7 +753,7 @@ func TestValidEngineNamesConsistency(t *testing.T) {
 	assert.Len(t, thirdCall, len(secondCall), "Engine names list length should be consistent")
 
 	// Verify all expected engines are present in all calls
-	expectedEngines := []string{"copilot", "claude", "codex"}
+	expectedEngines := []string{"copilot", "claude", "codex", "gemini"}
 	for _, engine := range expectedEngines {
 		assert.Contains(t, firstCall, engine, "Expected engine '%s' in first call", engine)
 		assert.Contains(t, secondCall, engine, "Expected engine '%s' in second call", engine)

pkg/constants/constants.go

@@ -303,6 +303,9 @@ const (
 
 	// CopilotLLMGatewayPort is the port for the Copilot LLM gateway
 	CopilotLLMGatewayPort = 10002
+
+	// GeminiLLMGatewayPort is the port for the Gemini LLM gateway
+	GeminiLLMGatewayPort = 10003
 )
 
 // DefaultMCPRegistryURL is the default MCP registry URL.
@@ -339,12 +342,16 @@ const (
 	EnvVarModelAgentCodex = "GH_AW_MODEL_AGENT_CODEX"
 	// EnvVarModelAgentCustom configures the default Custom model for agent execution
 	EnvVarModelAgentCustom = "GH_AW_MODEL_AGENT_CUSTOM"
+	// EnvVarModelAgentGemini configures the default Gemini model for agent execution
+	EnvVarModelAgentGemini = "GH_AW_MODEL_AGENT_GEMINI"
 	// EnvVarModelDetectionCopilot configures the default Copilot model for detection
 	EnvVarModelDetectionCopilot = "GH_AW_MODEL_DETECTION_COPILOT"
 	// EnvVarModelDetectionClaude configures the default Claude model for detection
 	EnvVarModelDetectionClaude = "GH_AW_MODEL_DETECTION_CLAUDE"
 	// EnvVarModelDetectionCodex configures the default Codex model for detection
 	EnvVarModelDetectionCodex = "GH_AW_MODEL_DETECTION_CODEX"
+	// EnvVarModelDetectionGemini configures the default Gemini model for detection
+	EnvVarModelDetectionGemini = "GH_AW_MODEL_DETECTION_GEMINI"
 
 	// Common environment variable names used across all engines
 
@@ -373,6 +380,9 @@ const (
 // DefaultCodexVersion is the default version of the OpenAI Codex CLI
 const DefaultCodexVersion Version = "0.104.0"
 
+// DefaultGeminiVersion is the default version of the Google Gemini CLI
+const DefaultGeminiVersion Version = "0.29.0"
+
 // DefaultGitHubMCPServerVersion is the default version of the GitHub MCP server Docker image
 const DefaultGitHubMCPServerVersion Version = "v0.30.3"
 

pkg/parser/schemas/main_workflow_schema.json

@@ -6848,17 +6848,17 @@
       "oneOf": [
         {
           "type": "string",
-          "enum": ["claude", "codex", "copilot"],
-          "description": "Simple engine name: 'claude' (default, Claude Code), 'copilot' (GitHub Copilot CLI), or 'codex' (OpenAI Codex CLI)"
+          "enum": ["claude", "codex", "copilot", "gemini"],
+          "description": "Simple engine name: 'claude' (default, Claude Code), 'copilot' (GitHub Copilot CLI), 'codex' (OpenAI Codex CLI), or 'gemini' (Google Gemini CLI - experimental)"
         },
         {
           "type": "object",
           "description": "Extended engine configuration object with advanced options for model selection, turn limiting, environment variables, and custom steps",
           "properties": {
             "id": {
               "type": "string",
-              "enum": ["claude", "codex", "copilot"],
-              "description": "AI engine identifier: 'claude' (Claude Code), 'codex' (OpenAI Codex CLI), or 'copilot' (GitHub Copilot CLI)"
+              "enum": ["claude", "codex", "copilot", "gemini"],
+              "description": "AI engine identifier: 'claude' (Claude Code), 'codex' (OpenAI Codex CLI), 'copilot' (GitHub Copilot CLI), or 'gemini' (Google Gemini CLI - experimental)"
             },
             "version": {
               "type": ["string", "number"],

pkg/workflow/agentic_engine.go

@@ -325,6 +325,7 @@ func NewEngineRegistry() *EngineRegistry {
 	registry.Register(NewClaudeEngine())
 	registry.Register(NewCodexEngine())
 	registry.Register(NewCopilotEngine())
+	registry.Register(NewGeminiEngine())
 
 	agenticEngineLog.Printf("Registered %d engines", len(registry.engines))
 	return registry

pkg/workflow/agentic_engine_test.go

@@ -9,10 +9,21 @@ import (
 func TestEngineRegistry(t *testing.T) {
 	registry := NewEngineRegistry()
 
-	// Test that built-in engines are registered
+	// Test that built-in engines are registered - check for specific IDs rather than exact count
+	// to avoid brittleness when new engines are added
 	supportedEngines := registry.GetSupportedEngines()
-	if len(supportedEngines) != 3 {
-		t.Errorf("Expected 3 supported engines, got %d", len(supportedEngines))
+	expectedEngineIDs := []string{"claude", "codex", "copilot", "gemini"}
+	for _, engineID := range expectedEngineIDs {
+		found := false
+		for _, id := range supportedEngines {
+			if id == engineID {
+				found = true
+				break
+			}
+		}
+		if !found {
+			t.Errorf("Expected engine '%s' to be registered", engineID)
+		}
 	}
 
 	// Test getting engines by ID

pkg/workflow/domains.go

@@ -99,6 +99,16 @@ var ClaudeDefaultDomains = []string{
 	"ts-ocsp.ws.symantec.com",
 }
 
+// GeminiDefaultDomains are the default domains required for Google Gemini CLI authentication and operation
+var GeminiDefaultDomains = []string{
+	"*.googleapis.com",
+	"generativelanguage.googleapis.com",
+	"github.com",
+	"host.docker.internal",
+	"raw.githubusercontent.com",
+	"registry.npmjs.org",
+}
+
 // PlaywrightDomains are the domains required for Playwright browser downloads
 // These domains are needed when Playwright MCP server initializes in the Docker container
 var PlaywrightDomains = []string{
@@ -510,6 +520,12 @@ func GetClaudeAllowedDomainsWithToolsAndRuntimes(network *NetworkPermissions, to
 	return mergeDomainsWithNetworkToolsAndRuntimes(ClaudeDefaultDomains, network, tools, runtimes)
 }
 
+// GetGeminiAllowedDomainsWithToolsAndRuntimes merges Gemini default domains with NetworkPermissions, HTTP MCP server domains, and runtime ecosystem domains
+// Returns a deduplicated, sorted, comma-separated string suitable for AWF's --allow-domains flag
+func GetGeminiAllowedDomainsWithToolsAndRuntimes(network *NetworkPermissions, tools map[string]any, runtimes map[string]any) string {
+	return mergeDomainsWithNetworkToolsAndRuntimes(GeminiDefaultDomains, network, tools, runtimes)
+}
+
 // GetBlockedDomains returns the blocked domains from network permissions
 // Returns empty slice if no network permissions configured or no domains blocked
 // The returned list is sorted and deduplicated