chore: clean up

CLAUDE.md

@@ -1,113 +1,36 @@
 # CLAUDE.md
 
-This file provides guidance to Claude Code (claude.ai/code) when working with
- code in this repository.
-
 ## Project Overview
 
-Ghost is a command-line AI assistant written in Go and powered by Ollama, designed
- with a cyberpunk aesthetic inspired by Shadowrun, Cyberpunk 2077, and The Matrix.
- It provides local AI capabilities for querying, analyzing piped data, processing
- images with vision models, and formatting output (text, JSON, Markdown).
-
-## Architecture
-
-### Core Flow
+Ghost is a command line AI assistant written in Go and powered by Ollama, designed
+with a cyberpunk aesthetic inspired by Shadowrun, Cyberpunk 2077, and The Matrix.
 
-1. **Entry Point** (`main.go`): Initializes root command via Fang CLI framework
- with custom theming and error handling
-2. **Root Command** (`cmd/root.go`): Orchestrates the main execution flow:
-   - Collects user prompt, piped input, and flags
-   - Analyzes images if provided (using vision model)
-   - Executes tool calls in a loop before streaming final response
-   - Streams LLM response using Bubbletea TUI
-   - Renders final output with appropriate formatting
-3. **LLM Client** (`internal/llm/ollama.go`): Communicates with Ollama API
-   - `StreamChat()`: Streaming chat with callback for each chunk
-   - `AnalyzeImages()`: Non-streaming vision model requests
-4. **UI Layer** (`internal/ui/`): Bubbletea models for interactive display
-   - `stream.go`: Streaming model for single-shot queries
-   - `chat.go`: Core ChatModel struct, types, Init, Update, View
-   - `chat_normal.go`: Normal mode key handling
-   - `chat_command.go`: Command mode (`:` commands like `:q`, `:r`)
-   - `chat_insert.go`: Insert mode text input handling
-   - `chat_stream.go`: LLM streaming and response handling
-5. **Theme System** (`theme/`): Handles cyberpunk-themed rendering and formatting
-   - UI glyphs in `theme/glyph.go`: Use `theme.GlyphInfo` (󱙝) and `theme.GlyphError`
-    (󱙜)
+## Configuration
 
-### Configuration System
-
-Configuration priority (highest to lowest):
+Priority (highest to lowest):
 
 1. Command-line flags
 2. Environment variables (prefixed with `GHOST_`, dots/hyphens replaced with `*`)
 3. Config file (`~/.config/ghost/config.toml`)
 
-Implemented in `cmd/config.go` using Viper. Vision model configuration uses
- nested structure: `vision.model` in config file, `--vision-model` flag, or
- `GHOST_VISION*MODEL` env var. Web search uses `search.api-key` and
- `search.max-results` following the same pattern.
-
-### Message Flow for Images
-
-Images are base64 encoded and analyzed separately with the vision model. Analysis
- results are formatted with IMAGE_ANALYSIS blocks and appended to message history
- before the main model processes everything.
-
-Vision system prompt is designed to prevent prompt injection from image text by
- treating all visible text as data, not instructions.
-
-### Streaming Architecture
-
-User goroutine and Bubbletea message passing where callbacks send chunk/done/error
- messages to the SteamModel for incremental rendering.
-
-### Error Handling Pattern
-
-All packages define custom error types (e.g., `ErrImageAnalysis`, `ErrModelNotFound`)
- with cyberpunk-themed messages. Errors are wrapped using `fmt.Errorf("%w", err)`
- for proper unwrapping. Theme package provides custom Fang error handler.
+Nested config keys use dot notation in TOML, hyphens in flags, and `*` in env
+vars (e.g., `vision.model` / `--vision-model` / `GHOST_VISION*MODEL`).
 
 ## Code Conventions
 
-**Style**:
-
 - Standard Go formatting (enforced by pre-commit)
 - Wrap errors with `fmt.Errorf("%w", err)` for proper error chains
-- Follow Go naming conventions (exported vs unexported)
-- Comment struct fields and exported types
-- Cyberpunk aesthetic in user-facing messages (e.g., "neural link", "data stream",
- "visual recon")
-
-**Testing**:
-
-- One test function per code function: Test function name matches the function
-being tested
+- Cyberpunk aesthetic in user-facing messages (e.g., "neural link", "data stream")
+- UI glyphs: `style.GlyphInfo` (󱙝) and `style.GlyphError` (󱙜)
+- Table-driven tests, one test function per code function
   (e.g., `TestChatModel_HandleCommandMode` tests `handleCommandMode`)
-- Use table-driven tests pattern (see `cmd/root_test.go`, `internal/llm/ollama_test.go`)
-- Test file naming mirrors source files (e.g., `chat_command_test.go` for `chat_command.go`)
+- Test file naming mirrors source files (e.g., `chat_command_test.go`)
 - Use `errors.Is()` for error comparison
-- Use `t.Fatalf()` for unexpected errors, `t.Errorf()` for assertions
-
-**Commit Messages**:
-Conventional commits format (`feat:`, `fix:`, `refactor:`, `test:`, `docs:`)
+- Conventional commits (`feat:`, `fix:`, `refactor:`, `test:`, `docs:`, `chore:`,
+ `perf:`, `build:`, `ci:`, `style:`)
 
 ## Design Principles
 
-- **Keep it simple**: Single-file structure per package unless strong reason to
- split
-- **Cyberpunk aesthetic**: Match tone in user-facing messages and error messages
-- **CLI-first**: Prioritize terminal experience with proper TTY detection
-- **Teach, don't implement**: When helping users, explain patterns and provide
- code examples rather than immediately editing files
-
-## Documentation
-
-### VHS Tape Files (GIF Demos)
-
-- Located in `documentation/` directory
-- Standard settings: Fish shell, 14pt font, 1200x600 dimensions
-- Use `ghost` command (not `go run .`) in demos for cleaner output
-- Key timing: 500ms between user actions, 12-20s for LLM response streaming
-- Generate GIFs with `vhs <filename>.tape` from the documentation directory
+- **Keep it simple**: Single file structure per package unless strong reason to split
+- **Cyberpunk aesthetic**: Match tone in user facing messages and error messages
+- **TUI first**: Prioritize terminal experience with proper TTY detection

cmd/chat.go

@@ -13,7 +13,7 @@ import (
 	"github.com/theantichris/ghost/v3/internal/agent"
 	"github.com/theantichris/ghost/v3/internal/storage"
 	"github.com/theantichris/ghost/v3/internal/tool"
-	"github.com/theantichris/ghost/v3/internal/ui"
+	"github.com/theantichris/ghost/v3/internal/tui"
 )
 
 var ErrHomeDir = errors.New("failed to retrieve user home directory")
@@ -59,7 +59,7 @@ func runChat(cmd *cobra.Command, args []string) error {
 
 	prompts := cmd.Context().Value(promptKey{}).(agent.Prompt)
 
-	config := ui.ModelConfig{
+	config := tui.ModelConfig{
 		Context:   cmd.Context(),
 		Logger:    logger,
 		URL:       viper.GetString("url"),
@@ -70,7 +70,7 @@ func runChat(cmd *cobra.Command, args []string) error {
 		Store:     store,
 	}
 
-	chatModel := ui.NewChatModel(config)
+	chatModel := tui.NewChatModel(config)
 
 	logger.Info("entering chat", "ollama_url", config.URL, "chat_model", config.ChatLLM, "vision_model", config.VisionLLM)
 	program := tea.NewProgram(chatModel)

cmd/root.go

@@ -15,21 +15,11 @@ import (
 	"github.com/theantichris/ghost/v3/internal/agent"
 	"github.com/theantichris/ghost/v3/internal/llm"
 	"github.com/theantichris/ghost/v3/internal/tool"
-	"github.com/theantichris/ghost/v3/internal/ui"
-	"github.com/theantichris/ghost/v3/theme"
+	"github.com/theantichris/ghost/v3/internal/tui"
+	"github.com/theantichris/ghost/v3/style"
 )
 
-const (
-	Version = "dev"
-
-	useText   = "ghost <prompt>"
-	shortText = "ghost is a local cyberpunk AI assistant."
-	longText  = `Ghost is a local cyberpunk AI assistant.
-Send prompts directly or pipe data through for analysis.`
-	exampleText = `  ghost "explain this code" < main.go
-	cat error.log | ghost "what's wrong here"
-	ghost "tell me a joke"`
-)
+const Version = "dev"
 
 type promptKey struct{}
 
@@ -50,11 +40,13 @@ func NewRootCmd() (*cobra.Command, func() error, error) {
 	var cfgFile string
 
 	cmd := &cobra.Command{
-		Use:     useText,
-		Short:   shortText,
-		Long:    longText,
-		Example: exampleText,
-		Args:    cobra.MinimumNArgs(1),
+		Use:   "ghost <prompt>",
+		Short: "ghost is a local cyberpunk AI assistant.",
+		Long:  "Ghost is a local cyberpunk AI Assistant.\nSend prompts directly or pipe data through for analysis.",
+		Example: `  ghost "explain this code" < main.go
+	cat error.log | ghost "what's wrong here"
+	ghost "tell me a joke"`,
+		Args: cobra.MinimumNArgs(1),
 		PersistentPreRunE: func(cmd *cobra.Command, args []string) error {
 			cmd.SetContext(context.WithValue(cmd.Context(), loggerKey{}, logger))
 
@@ -126,7 +118,7 @@ func run(cmd *cobra.Command, args []string) error {
 	maxResults := viper.GetInt("search.max-results")
 	registry := tool.NewRegistry(tavilyAPIKey, maxResults, logger)
 
-	config := ui.ModelConfig{
+	config := tui.ModelConfig{
 		Context:   cmd.Context(),
 		Prompts:   cmd.Context().Value(promptKey{}).(agent.Prompt),
 		Logger:    logger,
@@ -138,7 +130,7 @@ func run(cmd *cobra.Command, args []string) error {
 		Images:    images,
 		Registry:  registry,
 	}
-	streamModel := ui.NewStreamModel(config)
+	streamModel := tui.NewStreamModel(config)
 
 	var programOpts []tea.ProgramOption
 	if ttyIn, ttyOut, err := tea.OpenTTY(); err == nil {
@@ -156,12 +148,12 @@ func run(cmd *cobra.Command, args []string) error {
 		return fmt.Errorf("%w: %w", ErrStreamDisplay, err)
 	}
 
-	finalModel := returnedModel.(ui.StreamModel)
+	finalModel := returnedModel.(tui.StreamModel)
 	if finalModel.Err != nil {
 		return finalModel.Err
 	}
 
-	render, err := theme.RenderContent(finalModel.Content(), format, isTTY)
+	render, err := style.RenderContent(finalModel.Content(), format, isTTY)
 	if err != nil {
 		return fmt.Errorf("%w: %w", ErrRender, err)
 	}

internal/tool/registry.go

@@ -12,6 +12,12 @@ import (
 
 var ErrToolNotRegistered = errors.New("tool not registered")
 
+// Tool is the interface that all the tools the LLM uses must implement.
+type Tool interface {
+	Definition() llm.Tool
+	Execute(ctx context.Context, args json.RawMessage) (string, error)
+}
+
 // Registry holds all available tools, provides their definitions to send to chat
 // requests, and dispatches execution to the right tool by name.
 type Registry struct {