Add answer worker

Subsystem for message processing: parses messages, generates LLM
responses, and replies with SMTP.

Introduces:
- answer: message processing worker.
- llm: OpenAI API compatible client with support for tool execution.
- message: message parsing and response logic.
- tool: converts YAML configuration into executable subprocesses.
- smtp: simple config and client wrapper for sending email.
This commit is contained in:
dwrz
2026-01-04 21:01:57 +00:00
parent c53ee5f6ad
commit 486cc5fa52
9 changed files with 3228 additions and 0 deletions

235
internal/tool/tools.go Normal file
View File

@@ -0,0 +1,235 @@
// Package tool provides a registry for external tools that can be invoked by
// LLMs.
//
// The package bridges YAML configuration to exec.CommandContext, allowing
// tools to be defined declaratively without writing Go code. Each tool
// specifies a command, argument templates using Go's text/template syntax,
// JSON Schema parameters for LLM input, and an optional execution timeout.
//
// The registry validates all tool definitions at construction time,
// failing fast on configuration errors. At execution time, it expands
// argument templates with LLM-provided JSON, runs the subprocess, and
// returns stdout.
package tool
import (
"bytes"
"context"
"encoding/json"
"fmt"
"os/exec"
"text/template"
"time"
"github.com/sashabaranov/go-openai"
)
// fn provides template functions available to argument templates.
var fn = template.FuncMap{
"json": func(v any) string {
b, _ := json.Marshal(v)
return string(b)
},
}
// Tool represents an external tool that can be invoked by LLMs.
// Tools are executed as subprocesses with templated arguments.
type Tool struct {
// Name uniquely identifies the tool within a registry.
Name string `yaml:"name"`
// Description explains the tool's purpose for the LLM.
Description string `yaml:"description"`
// Command is the executable path or name.
Command string `yaml:"command"`
// Arguments are Go templates expanded with LLM-provided parameters.
// Empty results after expansion are filtered out.
Arguments []string `yaml:"arguments"`
// Parameters is a JSON Schema describing expected input from the LLM.
Parameters map[string]any `yaml:"parameters"`
// Timeout limits execution time (e.g., "30s", "5m").
// Empty means no timeout.
Timeout string `yaml:"timeout"`
// timeout is the parsed duration, set during registry construction.
timeout time.Duration `yaml:"-"`
}
// OpenAI converts the tool to an OpenAI function definition for API calls.
func (t Tool) OpenAI() openai.Tool {
return openai.Tool{
Type: openai.ToolTypeFunction,
Function: &openai.FunctionDefinition{
Name: t.Name,
Description: t.Description,
Parameters: t.Parameters,
},
}
}
// ParseArguments expands argument templates with the provided JSON data.
// The args parameter should be a JSON object string; empty string or "{}"
// results in an empty data map. Templates producing empty strings are
// filtered from the result, allowing conditional arguments.
func (t Tool) ParseArguments(args string) ([]string, error) {
var data = map[string]any{}
if args != "" && args != "{}" {
if err := json.Unmarshal([]byte(args), &data); err != nil {
return nil, fmt.Errorf(
"invalid arguments JSON: %w", err,
)
}
}
var result []string
for _, v := range t.Arguments {
tmpl, err := template.New("").Funcs(fn).Parse(v)
if err != nil {
return nil, fmt.Errorf(
"invalid template %q: %w", v, err,
)
}
var buf bytes.Buffer
if err := tmpl.Execute(&buf, data); err != nil {
return nil, fmt.Errorf(
"execute template %q: %w", v, err,
)
}
// Filter out empty strings (unused conditional arguments).
if s := buf.String(); s != "" {
result = append(result, s)
}
}
return result, nil
}
// Validate checks that the tool definition is complete and valid.
// It verifies required fields are present, the timeout (if specified)
// is parseable, and all argument templates are syntactically valid.
func (t Tool) Validate() error {
if t.Name == "" {
return fmt.Errorf("missing name")
}
if t.Description == "" {
return fmt.Errorf("missing description")
}
if t.Command == "" {
return fmt.Errorf("missing command")
}
if t.Timeout != "" {
if _, err := time.ParseDuration(t.Timeout); err != nil {
return fmt.Errorf("invalid timeout: %v", err)
}
}
for _, arg := range t.Arguments {
if _, err := template.New("").Funcs(fn).Parse(arg); err != nil {
return fmt.Errorf("invalid argument template")
}
}
return nil
}
// Registry holds tools indexed by name and handles their execution.
// It validates all tools at construction time to fail fast on
// configuration errors.
type Registry struct {
tools map[string]*Tool
}
// NewRegistry creates a registry from the provided tool definitions.
// Returns an error if any tool fails validation or if duplicate names exist.
func NewRegistry(tools []Tool) (*Registry, error) {
var r = &Registry{
tools: make(map[string]*Tool),
}
for _, t := range tools {
if err := t.Validate(); err != nil {
return nil, fmt.Errorf("invalid tool: %v", err)
}
if t.Timeout != "" {
d, err := time.ParseDuration(t.Timeout)
if err != nil {
return nil, fmt.Errorf(
"parse timeout: %v", err,
)
}
t.timeout = d
}
if _, exists := r.tools[t.Name]; exists {
return nil, fmt.Errorf(
"duplicate tool name: %s", t.Name,
)
}
r.tools[t.Name] = &t
}
return r, nil
}
// Get returns a tool by name and a boolean indicating if it was found.
func (r *Registry) Get(name string) (*Tool, bool) {
tool, ok := r.tools[name]
return tool, ok
}
// List returns all registered tool names in arbitrary order.
func (r *Registry) List() []string {
names := make([]string, 0, len(r.tools))
for name := range r.tools {
names = append(names, name)
}
return names
}
// Execute runs a tool by name with the provided JSON arguments.
// It expands argument templates, executes the command as a subprocess,
// and returns stdout on success. The context can be used for cancellation;
// tool-specific timeouts are applied on top of any context deadline.
func (r *Registry) Execute(
ctx context.Context, name string, args string,
) (string, error) {
tool, ok := r.tools[name]
if !ok {
return "", fmt.Errorf("unknown tool: %s", name)
}
// Evaluate argument templates.
cmdArgs, err := tool.ParseArguments(args)
if err != nil {
return "", fmt.Errorf("parse arguments: %w", err)
}
// If defined, use the timeout.
if tool.timeout > 0 {
var cancel context.CancelFunc
ctx, cancel = context.WithTimeout(ctx, tool.timeout)
defer cancel()
}
// Setup and run the command.
var (
stdout, stderr bytes.Buffer
cmd = exec.CommandContext(
ctx, tool.Command, cmdArgs...,
)
)
cmd.Stdout = &stdout
cmd.Stderr = &stderr
if err := cmd.Run(); err != nil {
if ctx.Err() == context.DeadlineExceeded && tool.timeout > 0 {
return "", fmt.Errorf(
"tool %s timed out after %v",
name, tool.timeout,
)
}
return "", fmt.Errorf(
"tool %s: %w\nstderr: %s",
name, err, stderr.String(),
)
}
return stdout.String(), nil
}

374
internal/tool/tools_test.go Normal file
View File

@@ -0,0 +1,374 @@
package tool
import (
"context"
"slices"
"testing"
"time"
)
func TestToolValidate(t *testing.T) {
tests := []struct {
name string
tool Tool
wantErr bool
}{
{
name: "valid tool",
tool: Tool{
Name: "echo",
Description: "echoes input",
Command: "echo",
Arguments: []string{"{{.message}}"},
},
wantErr: false,
},
{
name: "valid tool with timeout",
tool: Tool{
Name: "slow",
Description: "slow command",
Command: "sleep",
Arguments: []string{"1"},
Timeout: "5s",
},
wantErr: false,
},
{
name: "missing name",
tool: Tool{
Description: "test",
Command: "echo",
},
wantErr: true,
},
{
name: "missing description",
tool: Tool{
Name: "test",
Command: "echo",
},
wantErr: true,
},
{
name: "missing command",
tool: Tool{
Name: "test",
Description: "test",
},
wantErr: true,
},
{
name: "invalid timeout",
tool: Tool{
Name: "test",
Description: "test",
Command: "echo",
Timeout: "invalid",
},
wantErr: true,
},
{
name: "invalid template",
tool: Tool{
Name: "test",
Description: "test",
Command: "echo",
Arguments: []string{"{{.unclosed"},
},
wantErr: true,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
err := tt.tool.Validate()
if (err != nil) != tt.wantErr {
t.Errorf("Validate() error = %v, wantErr %v", err, tt.wantErr)
}
})
}
}
func TestToolParseArguments(t *testing.T) {
tests := []struct {
name string
tool Tool
args string
want []string
wantErr bool
}{
{
name: "simple substitution",
tool: Tool{
Arguments: []string{"{{.message}}"},
},
args: `{"message": "hello"}`,
want: []string{"hello"},
},
{
name: "multiple arguments",
tool: Tool{
Arguments: []string{"-n", "{{.count}}", "{{.file}}"},
},
args: `{"count": "10", "file": "test.txt"}`,
want: []string{"-n", "10", "test.txt"},
},
{
name: "conditional with if",
tool: Tool{
Arguments: []string{"{{.required}}", "{{if .optional}}{{.optional}}{{end}}"},
},
args: `{"required": "value"}`,
want: []string{"value"},
},
{
name: "json function",
tool: Tool{
Arguments: []string{`{{json .data}}`},
},
args: `{"data": {"key": "value"}}`,
want: []string{`{"key":"value"}`},
},
{
name: "empty JSON object",
tool: Tool{
Arguments: []string{"fixed"},
},
args: "{}",
want: []string{"fixed"},
},
{
name: "empty string args",
tool: Tool{
Arguments: []string{"fixed"},
},
args: "",
want: []string{"fixed"},
},
{
name: "invalid JSON",
tool: Tool{
Arguments: []string{"{{.x}}"},
},
args: "not json",
wantErr: true,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
got, err := tt.tool.ParseArguments(tt.args)
if (err != nil) != tt.wantErr {
t.Errorf("ParseArguments() error = %v, wantErr %v", err, tt.wantErr)
return
}
if !tt.wantErr && !slices.Equal(got, tt.want) {
t.Errorf("ParseArguments() = %v, want %v", got, tt.want)
}
})
}
}
func TestNewRegistry(t *testing.T) {
tests := []struct {
name string
tools []Tool
wantErr bool
}{
{
name: "valid registry",
tools: []Tool{
{Name: "a", Description: "a", Command: "echo"},
{Name: "b", Description: "b", Command: "cat"},
},
wantErr: false,
},
{
name: "empty registry",
tools: []Tool{},
wantErr: false,
},
{
name: "duplicate names",
tools: []Tool{
{Name: "dup", Description: "first", Command: "echo"},
{Name: "dup", Description: "second", Command: "cat"},
},
wantErr: true,
},
{
name: "invalid tool",
tools: []Tool{
{Name: "", Description: "missing name", Command: "echo"},
},
wantErr: true,
},
}
for _, tt := range tests {
t.Run(tt.name, func(t *testing.T) {
_, err := NewRegistry(tt.tools)
if (err != nil) != tt.wantErr {
t.Errorf("NewRegistry() error = %v, wantErr %v", err, tt.wantErr)
}
})
}
}
func TestRegistryGet(t *testing.T) {
r, err := NewRegistry([]Tool{
{Name: "echo", Description: "echo", Command: "echo"},
})
if err != nil {
t.Fatalf("NewRegistry() error = %v", err)
}
tool, ok := r.Get("echo")
if !ok {
t.Error("Get(echo) returned false, want true")
}
if tool.Name != "echo" {
t.Errorf("Get(echo).Name = %q, want %q", tool.Name, "echo")
}
_, ok = r.Get("nonexistent")
if ok {
t.Error("Get(nonexistent) returned true, want false")
}
}
func TestRegistryList(t *testing.T) {
r, err := NewRegistry([]Tool{
{Name: "a", Description: "a", Command: "echo"},
{Name: "b", Description: "b", Command: "cat"},
{Name: "c", Description: "c", Command: "ls"},
})
if err != nil {
t.Fatalf("NewRegistry() error = %v", err)
}
names := r.List()
if len(names) != 3 {
t.Errorf("List() returned %d names, want 3", len(names))
}
slices.Sort(names)
want := []string{"a", "b", "c"}
if !slices.Equal(names, want) {
t.Errorf("List() = %v, want %v", names, want)
}
}
func TestRegistryExecute(t *testing.T) {
r, err := NewRegistry([]Tool{
{
Name: "echo",
Description: "echo message",
Command: "echo",
Arguments: []string{"{{.message}}"},
},
{
Name: "cat",
Description: "cat stdin",
Command: "cat",
Arguments: []string{},
},
})
if err != nil {
t.Fatalf("NewRegistry() error = %v", err)
}
ctx := context.Background()
// Successful execution.
out, err := r.Execute(ctx, "echo", `{"message": "hello world"}`)
if err != nil {
t.Errorf("Execute(echo) error = %v", err)
}
if out != "hello world\n" {
t.Errorf("Execute(echo) = %q, want %q", out, "hello world\n")
}
// Unknown tool.
_, err = r.Execute(ctx, "unknown", "{}")
if err == nil {
t.Error("Execute(unknown) expected error, got nil")
}
// Invalid JSON arguments.
_, err = r.Execute(ctx, "echo", "not json")
if err == nil {
t.Error("Execute with invalid JSON expected error, got nil")
}
}
func TestRegistryExecuteTimeout(t *testing.T) {
r, err := NewRegistry([]Tool{
{
Name: "slow",
Description: "slow command",
Command: "sleep",
Arguments: []string{"10"},
Timeout: "50ms",
},
})
if err != nil {
t.Fatalf("NewRegistry() error = %v", err)
}
ctx := context.Background()
start := time.Now()
_, err = r.Execute(ctx, "slow", "{}")
elapsed := time.Since(start)
if err == nil {
t.Error("Execute(slow) expected timeout error, got nil")
}
if elapsed > time.Second {
t.Errorf("Execute(slow) took %v, expected ~50ms timeout", elapsed)
}
}
func TestRegistryExecuteFailure(t *testing.T) {
r, err := NewRegistry([]Tool{
{
Name: "fail",
Description: "always fails",
Command: "false",
},
})
if err != nil {
t.Fatalf("NewRegistry() error = %v", err)
}
_, err = r.Execute(context.Background(), "fail", "{}")
if err == nil {
t.Error("Execute(fail) expected error, got nil")
}
}
func TestToolOpenAI(t *testing.T) {
tool := Tool{
Name: "test",
Description: "test tool",
Command: "echo",
Parameters: map[string]any{
"type": "object",
"properties": map[string]any{
"message": map[string]any{
"type": "string",
"description": "the message",
},
},
},
}
openaiTool := tool.OpenAI()
if openaiTool.Function.Name != "test" {
t.Errorf("OpenAI().Function.Name = %q, want %q", openaiTool.Function.Name, "test")
}
if openaiTool.Function.Description != "test tool" {
t.Errorf("OpenAI().Function.Description = %q, want %q", openaiTool.Function.Description, "test tool")
}
}