Agent-Native/adapters/native/exec_tools_test.go

package native

import (
	"context"
	"os"
	osexec "os/exec"
	"path/filepath"
	"strings"
	"testing"

	"agent-native/exec"
	"agent-native/runtime/llm"
	"agent-native/runtime/mcp"
	"agent-native/runtime/subagent"
	"agent-native/runtime/tools"
)

func TestBackendExecToolsWiredWithOptionsAndRequestEnv(t *testing.T) {
	bash, err := osexec.LookPath("bash")
	if err != nil {
		t.Fatalf("bash not available: %v", err)
	}
	dir := t.TempDir()
	b := New(Options{
		ModelOptions: ModelOptions{
			Model: "fake",
			Stream: script(
				llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
					ID: "c1", Name: "run_command",
					Arguments: `{"command":"echo role=$ROLE_FLAG step=$STEP_FLAG > exec_out.txt"}`,
				}}},
				llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
					ID: "c2", Name: "write_file",
					Arguments: `{"path":"output.md","content":"done"}`,
				}}},
				llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd},
			),
		},
		ToolOptions: ToolOptions{
			Exec: &tools.ExecConfig{Bash: bash, Env: []string{"PATH=/usr/bin:/bin", "ROLE_FLAG=backend"}},
		},
	})
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: dir, Prompt: "run", Targets: []string{"output.md"},
		Env: map[string]string{"STEP_FLAG": "request"},
	})
	if err != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v logs=%s", err, res, res.Logs)
	}
	data, err := os.ReadFile(filepath.Join(dir, "exec_out.txt"))
	if err != nil {
		t.Fatal(err)
	}
	if !strings.Contains(string(data), "role=backend") || !strings.Contains(string(data), "step=request") {
		t.Fatalf("env merge wrong: %q", data)
	}
}

func TestBackendExecToolsAbsentWithoutOptions(t *testing.T) {
	stream, c := captureStream()
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
	})
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: t.TempDir(), Prompt: "x", Targets: []string{"output.md"},
		ToolProfile: exec.ToolProfileReasoning,
	})
	if err != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v", err, res)
	}
	// reasoning 档本无工具；改用 workspace 档断言 decls 不含执行工具。
	stream2, c2 := captureStream()
	b2 := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream2,
		},
	})
	_, _ = b2.Execute(context.Background(), exec.Request{
		WorkDir: t.TempDir(), Prompt: "x", Targets: []string{"output.md"},
		ToolProfile: exec.ToolProfileWorkspace,
	})
	for _, names := range c2.toolDecls {
		for _, n := range names {
			if n == "run_command" || n == "run_python" {
				t.Fatalf("exec tools must not register without Options.Exec, got %v", names)
			}
		}
	}
	_ = c
}

func TestBackendAllowlistCanDisableExecTools(t *testing.T) {
	bash, _ := osexec.LookPath("bash")
	stream, c := captureStream()
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
		ToolOptions: ToolOptions{
			Exec: &tools.ExecConfig{Bash: bash, Env: []string{"PATH=/usr/bin:/bin"}},
		},
	})
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: t.TempDir(), Prompt: "x", Targets: []string{"output.md"},
		ToolAllowlist: []string{"read_file", "write_file", "edit_file", "list_dir", "grep", "glob", "multi_edit"},
	})
	if err != nil {
		t.Fatal(err)
	}
	if res.Success {
		t.Fatalf("write-only allowlist without write? (unexpected) res=%+v", res)
	}
	for _, names := range c.toolDecls {
		for _, n := range names {
			if n == "run_command" {
				t.Fatalf("allowlist must be able to strip run_command, got %v", names)
			}
		}
	}
}

// fakeSearchProvider 与 tools 包测试同构的最小搜索替身。
type fakeSearchProvider struct{ query string }

func (f *fakeSearchProvider) Search(_ context.Context, req tools.SearchRequest) (tools.SearchResponse, error) {
	f.query = req.Query
	return tools.SearchResponse{Answer: "ans", Results: []tools.SearchResult{{Title: "T", Content: "C", Score: 0.9}}}, nil
}

func TestBackendSearchToolsWiredViaOptions(t *testing.T) {
	provider := &fakeSearchProvider{}
	stream, c := captureStream(
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c1", Name: "web_search", Arguments: `{"query":"golang generics"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c2", Name: "write_file",
			Arguments: `{"path":"output.md","content":"done"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd},
	)
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
		ToolOptions: ToolOptions{
			Search: provider,
		},
	})
	dir := t.TempDir()
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: dir, Prompt: "search", Targets: []string{"output.md"},
	})
	if err != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v logs=%s", err, res, res.Logs)
	}
	if provider.query != "golang generics" {
		t.Fatalf("provider should receive the query, got %q", provider.query)
	}
	// 搜索工具确实出现在 decls 里。
	found := false
	for _, names := range c.toolDecls {
		for _, n := range names {
			if n == "web_search" || n == "tavily_search" {
				found = true
			}
		}
	}
	if !found {
		t.Fatal("search tools must be declared when Options.Search is set")
	}
}

func TestBackendSearchToolsAbsentWithoutOptions(t *testing.T) {
	stream, c := captureStream()
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
	})
	_, _ = b.Execute(context.Background(), exec.Request{
		WorkDir: t.TempDir(), Prompt: "x", Targets: []string{"output.md"},
		ToolProfile: exec.ToolProfileWorkspace,
	})
	for _, names := range c.toolDecls {
		for _, n := range names {
			if n == "web_search" || n == "tavily_search" {
				t.Fatalf("search tools must not register without Options.Search, got %v", names)
			}
		}
	}
}

func TestBackendMCPToolsWiredViaOptions(t *testing.T) {
	exe, err := osexec.LookPath(os.Args[0])
	if err != nil {
		t.Fatal(err)
	}
	// os.Args[0] 在 go test 下指向测试二进制——用其自身作为 MCP server。
	mgr, err := mcp.NewManager([]mcp.ServerConfig{{
		Name: "fake", Command: exe,
		Env: append(os.Environ(), "AGENT_NATIVE_TEST_MCP_STDIO=1"),
	}})
	if err != nil {
		t.Fatal(err)
	}
	defer mgr.Close()
	if err := mgr.RegisterInto(context.Background(), tools.NewRegistry(t.TempDir())); err != nil {
		t.Fatal(err)
	}
	stream, c := captureStream(
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c1", Name: "mcp__fake__echo", Arguments: `{"text":"via-backend"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c2", Name: "write_file",
			Arguments: `{"path":"output.md","content":"done"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd},
	)
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
		ToolOptions: ToolOptions{
			MCP: mgr,
		},
	})
	dir := t.TempDir()
	res, err2 := b.Execute(context.Background(), exec.Request{
		WorkDir: dir, Prompt: "call mcp", Targets: []string{"output.md"},
	})
	if err2 != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v logs=%s", err2, res, res.Logs)
	}
	found := false
	for _, names := range c.toolDecls {
		for _, n := range names {
			if n == "mcp__fake__echo" {
				found = true
			}
		}
	}
	if !found {
		t.Fatal("mcp bridge tools must be declared to the model")
	}
}

func TestBackendSkillsIndexAndLoader(t *testing.T) {
	skillsRoot := t.TempDir()
	skillDir := filepath.Join(skillsRoot, "docx-author")
	if err := os.MkdirAll(skillDir, 0o755); err != nil {
		t.Fatal(err)
	}
	if err := os.WriteFile(filepath.Join(skillDir, "SKILL.md"), []byte("---\nname: docx-author\ndescription: 生成文档交付物\n---\n\n配方正文"), 0o644); err != nil {
		t.Fatal(err)
	}
	stream, c := captureStream(
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c1", Name: "load_skill", Arguments: `{"name":"docx-author"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c2", Name: "write_file", Arguments: `{"path":"output.md","content":"done"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd},
	)
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
		ToolOptions: ToolOptions{
			SkillsRoots: []string{skillsRoot},
		},
	})
	dir := t.TempDir()
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: dir, Prompt: "写", Targets: []string{"output.md"}, Skills: []string{"docx-author"},
	})
	if err != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v logs=%s", err, res, res.Logs)
	}
	if !strings.Contains(c.system, "- docx-author — 生成文档交付物") {
		t.Fatalf("skill index missing in system prompt:\n%s", c.system)
	}
	if !strings.Contains(c.system, "Declared for this task") || !strings.Contains(c.system, "docx-author") {
		t.Fatalf("declared skills missing:\n%s", c.system)
	}
	sawLoader := false
	for _, names := range c.toolDecls {
		for _, n := range names {
			if n == "load_skill" {
				sawLoader = true
			}
		}
	}
	if !sawLoader {
		t.Fatal("load_skill must be declared when skills exist")
	}
}

func TestBackendDeclaredSkillMissingFailsClosed(t *testing.T) {
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: script(),
		},
		ToolOptions: ToolOptions{
			SkillsRoots: []string{t.TempDir()},
		},
	})
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: t.TempDir(), Prompt: "x", Targets: []string{"output.md"},
		Skills: []string{"ghost-skill"},
	})
	if err != nil {
		t.Fatal(err)
	}
	if res.Success || !strings.Contains(res.ErrorReason, "declared skill not installed: ghost-skill") {
		t.Fatalf("missing declared skill must fail-closed, got %+v", res)
	}
}

func TestBackendWorkspaceSkillsRootAutoMerged(t *testing.T) {
	stream, c := captureStream(
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c1", Name: "write_file", Arguments: `{"path":"output.md","content":"done"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd},
	)
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
	})
	dir := t.TempDir()
	wsSkill := filepath.Join(dir, ".agent", "skills", "ws-skill")
	if err := os.MkdirAll(wsSkill, 0o755); err != nil {
		t.Fatal(err)
	}
	if err := os.WriteFile(filepath.Join(wsSkill, "SKILL.md"), []byte("---\nname: ws-skill\ndescription: 工作区技能\n---\n正文"), 0o644); err != nil {
		t.Fatal(err)
	}
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: dir, Prompt: "x", Targets: []string{"output.md"},
	})
	if err != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v", err, res)
	}
	if !strings.Contains(c.system, "ws-skill") {
		t.Fatalf("workspace skills root must be merged into index:\n%s", c.system)
	}
}

func TestBackendSubagentWiredViaOptions(t *testing.T) {
	var delegated *subagent.Request
	stream, c := captureStream(
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c1", Name: "delegate_task",
			Arguments: `{"goal":"查证数据","write_scope":"research/","done_criteria":"三来源"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, StopReason: llm.StopToolCalls, ToolCalls: []llm.ToolCall{{
			ID: "c2", Name: "write_file",
			Arguments: `{"path":"output.md","content":"done"}`,
		}}},
		llm.Message{Role: llm.RoleAssistant, Content: "ok", StopReason: llm.StopEnd},
	)
	b := New(Options{
		ModelOptions: ModelOptions{
			Model:  "fake",
			Stream: stream,
		},
		ToolOptions: ToolOptions{
			Subagent: func(ctx context.Context, req subagent.Request) (subagent.Result, error) {
				delegated = &req
				return subagent.Result{Success: true, Text: "结论已查明"}, nil
			},
		},
	})
	dir := t.TempDir()
	res, err := b.Execute(context.Background(), exec.Request{
		WorkDir: dir, Prompt: "分工", Targets: []string{"output.md"},
	})
	if err != nil || !res.Success {
		t.Fatalf("execute: err=%v res=%+v logs=%s", err, res, res.Logs)
	}
	if delegated == nil || delegated.Goal != "查证数据" {
		t.Fatalf("factory not invoked with fields: %+v", delegated)
	}
	sawDelegate := false
	for _, names := range c.toolDecls {
		for _, n := range names {
			if n == "delegate_task" {
				sawDelegate = true
			}
		}
	}
	if !sawDelegate {
		t.Fatal("delegate_task must be declared when Subagent factory is set")
	}
}