smashimo/oh-my-pi
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
a4bfb5795cd3a1acefd80b351f3776fab7ec44ef
oh-my-pi/packages/coding-agent/DEVELOPMENT.md
T
can1357 a4bfb5795c refactor(coding-agent): renamed 'complete' tool to 'submit_result' for clarity and consistency 
- Renamed 'complete' tool to 'submit_result' throughout codebase for clarity and consistency.
- Updated all tool references, function names, and variable names from 'complete' to 'submit_result' in executor, render, and tools modules.
- Renamed CompleteTool class to SubmitResultTool and CompleteDetails interface to SubmitResultDetails.
- Updated agent prompts and documentation to reference the new 'submit_result' tool name.
- Reorganized task.md prompt to move critical guidance to the top and restructure instructions for better clarity.
2026-01-28 03:28:03 +01:00

728 lines
32 KiB
Markdown
Raw Blame History

# coding-agent Development Guide
This document describes the architecture and development workflow for the coding-agent package.
## Architecture Overview
The coding-agent is structured into distinct layers:
```
┌─────────────────────────────────────────────────────────────┐
│ CLI Layer │
│ cli.ts → main.ts → cli/args.ts, cli/file-processor.ts │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Mode Layer │
│ modes/interactive-mode.ts modes/print-mode.ts modes/rpc/│
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ Core Layer │
│ session/agent-session.ts, sdk.ts (SDK wrapper) │
│ session/session-manager.ts, config/model-resolver.ts, etc.│
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ External Dependencies │
│ @oh-my-pi/pi-agent-core (Agent core) │
│ @oh-my-pi/pi-ai (models, providers) │
│ @oh-my-pi/pi-tui (TUI components) │
└─────────────────────────────────────────────────────────────┘
```
## Directory Structure
```
src/
├── cli.ts # CLI entry point (shebang, calls main)
├── main.ts # Main orchestration, argument handling, mode routing
├── index.ts # Public API exports (SDK)
├── config.ts # APP_NAME, VERSION, paths (getAgentDir, etc.)
├── migrations.ts # Session/config migration logic
├── sdk.ts # SDK wrapper for programmatic usage
├── system-prompt.ts # buildSystemPrompt(), loadProjectContextFiles()
├── cursor.ts # Cursor exec bridge (flattened from execution/cursor/)
├── cli/ # CLI-specific utilities
│ ├── args.ts # parseArgs(), printHelp(), Args interface
│ ├── file-processor.ts # processFileArguments() for @file args
│ ├── list-models.ts # --list-models implementation
│ ├── plugin-cli.ts # Plugin management CLI
│ ├── session-picker.ts # selectSession() TUI for --resume
│ └── update-cli.ts # Self-update CLI
├── capability/ # Capability system (extension types)
│ ├── index.ts # Main capability registry and discovery
│ ├── context-file.ts # Context file capability
│ ├── extension.ts # Extension capability
│ ├── hook.ts # Hook capability
│ ├── instruction.ts # Instruction capability
│ ├── mcp.ts # MCP capability
│ ├── prompt.ts # Prompt capability
│ ├── rule.ts # Rulebook rule capability
│ ├── skill.ts # Skill capability
│ ├── slash-command.ts # Slash command capability
│ ├── system-prompt.ts # System prompt capability
│ └── tool.ts # Tool capability
├── config/ # Configuration management
│ ├── keybindings.ts # Keybinding configuration
│ ├── model-registry.ts # Model registry and configuration
│ ├── model-resolver.ts # resolveModelScope(), restoreModelFromSession()
│ ├── prompt-templates.ts # Prompt template loading and rendering
│ └── settings-manager.ts # SettingsManager class - user preferences
├── discovery/ # Extension discovery from multiple sources
│ ├── index.ts # Main discovery orchestration
│ ├── builtin.ts # Built-in extensions
│ ├── claude.ts # Claude.md discovery
│ ├── cline.ts # Cline .mcp.json discovery
│ ├── codex.ts # .codex discovery
│ ├── cursor.ts # Cursor .cursorrules discovery
│ ├── gemini.ts # Gemini .config discovery
│ ├── github.ts # GitHub extension discovery
│ ├── mcp-json.ts # MCP JSON discovery
│ ├── vscode.ts # VSCode extension discovery
│ ├── windsurf.ts # Windsurf discovery
│ └── helpers.ts # Discovery helper functions
├── exa/ # Exa MCP tools (22 tools)
│ ├── index.ts # Exa tools exports
│ ├── company.ts # Company search
│ ├── linkedin.ts # LinkedIn search
│ ├── mcp-client.ts # MCP client for Exa
│ ├── render.ts # Exa result rendering
│ ├── researcher.ts # Research tools
│ ├── search.ts # Search tools
│ ├── types.ts # Exa types
│ └── websets.ts # Websets tools
├── exec/ # Bash/shell execution
│ ├── bash-executor.ts # executeBash() with streaming, abort
│ └── exec.ts # Process execution utilities
├── export/ # Export functionality
│ ├── custom-share.ts # Custom sharing
│ ├── ttsr.ts # Text-to-speech/speech-to-text utilities
│ └── html/ # Session export to HTML
│ └── index.ts # HTML export logic
├── extensibility/ # Extension and customization systems
│ ├── skills.ts # loadSkills(), skill discovery
│ ├── slash-commands.ts # loadSlashCommands()
│ │
│ ├── custom-commands/ # Custom command loading system
│ │ └── types.ts # CustomCommand types
│ │
│ ├── custom-tools/ # Custom tool loading system
│ │ ├── index.ts # Custom tool exports
│ │ ├── types.ts # CustomToolFactory, CustomToolDefinition
│ │ ├── loader.ts # loadCustomTools()
│ │ └── wrapper.ts # Tool wrapper utilities
│ │
│ ├── extensions/ # Extension system
│ │ ├── index.ts # Extension exports
│ │ ├── types.ts # Extension types
│ │ ├── loader.ts # Extension loading
│ │ ├── runner.ts # Extension event dispatch
│ │ └── wrapper.ts # Extension wrappers
│ │
│ ├── hooks/ # Hook system
│ │ ├── index.ts # Hook exports
│ │ ├── types.ts # HookAPI, HookContext, event types
│ │ ├── loader.ts # loadHooks()
│ │ └── runner.ts # runHook() event dispatch
│ │
│ └── plugins/ # Plugin system
│ └── (plugin loading and management)
├── internal-urls/ # Internal URL protocol handlers
│ ├── index.ts # Router and exports
│ ├── rule-protocol.ts # Rule protocol handler
│ └── skill-protocol.ts # Skill protocol handler
├── ipy/ # Python/Jupyter execution
│ ├── executor.ts # executePython(), kernel execution
│ ├── gateway-coordinator.ts # Gateway coordination
│ ├── kernel.ts # PythonKernel class
│ ├── modules.ts # Python module discovery
│ ├── prelude.ts # Python prelude (TS bindings)
│ └── prelude.py # Python prelude script
├── lsp/ # LSP integration
│ ├── index.ts # LSP exports
│ ├── client.ts # LSP client
│ ├── config.ts # LSP configuration
│ ├── render.ts # LSP result rendering
│ ├── utils.ts # LSP utilities
│ └── clients/ # Language-specific clients
│ ├── index.ts # Client exports
│ ├── biome-client.ts # Biome LSP
│ └── lsp-linter-client.ts # Generic linter
├── mcp/ # MCP (Model Context Protocol) integration
│ ├── index.ts # MCP exports
│ ├── config.ts # MCP configuration
│ ├── loader.ts # MCP server loading
│ ├── manager.ts # MCPManager class
│ ├── tool-bridge.ts # Tool bridging
│ ├── tool-cache.ts # Tool caching
│ ├── types.ts # MCP types
│ └── transports/ # MCP transports
│ ├── http.ts # HTTP transport
│ └── stdio.ts # Stdio transport
├── modes/ # Run mode implementations
│ ├── index.ts # Re-exports InteractiveMode, runPrintMode, runRpcMode
│ ├── interactive-mode.ts # InteractiveMode class (main TUI)
│ ├── print-mode.ts # Non-interactive mode
│ ├── types.ts # Mode types
│ │
│ ├── components/ # TUI components
│ │ ├── index.ts # Component exports
│ │ ├── assistant-message.ts # Agent response rendering
│ │ ├── bash-execution.ts # Bash output display
│ │ ├── custom-editor.ts # Multi-line input editor
│ │ ├── dynamic-border.ts # Adaptive border rendering
│ │ ├── footer.ts # Status bar / footer
│ │ ├── hook-input.ts # Hook input dialog
│ │ ├── hook-selector.ts # Hook selection UI
│ │ ├── login-dialog.ts # OAuth login dialog
│ │ ├── model-selector.ts # Model picker
│ │ ├── session-selector.ts # Session browser for --resume
│ │ ├── theme-selector.ts # Theme picker
│ │ ├── thinking-selector.ts # Thinking level picker
│ │ ├── tool-execution.ts # Tool call/result rendering
│ │ ├── user-message.ts # User message rendering
│ │ ├── status-line/ # Status line components
│ │ └── extensions/ # Extension UI components
│ │
│ ├── controllers/ # TUI controllers
│ │ ├── command-controller.ts # Command handling
│ │ ├── event-controller.ts # Event handling
│ │ ├── input-controller.ts # Input handling
│ │ └── selector-controller.ts # Selector handling
│ │
│ ├── theme/ # Theme system
│ │ ├── theme.ts # Theme loading, getEditorTheme()
│ │ └── defaults/ # Default themes
│ │
│ ├── utils/ # Mode utilities
│ │ └── ui-helpers.ts # UI helper functions
│ │
│ └── rpc/ # RPC mode for programmatic control
│ ├── rpc-mode.ts # runRpcMode() - JSON stdin/stdout protocol
│ ├── rpc-types.ts # RpcCommand, RpcResponse types
│ └── rpc-client.ts # RpcClient class
├── patch/ # File patching/editing
│ ├── index.ts # EditTool and exports
│ ├── applicator.ts # Patch application
│ ├── diff.ts # Diff utilities
│ ├── shared.ts # Shared utilities
│ └── render.ts # Patch rendering
├── prompts/ # Prompt templates
│ ├── system/ # System prompts
│ ├── agents/ # Agent-specific prompts
│ ├── compaction/ # Compaction prompts
│ └── tools/ # Tool-specific prompts
├── session/ # Session management
│ ├── agent-session.ts # AgentSession class - THE central abstraction
│ ├── agent-storage.ts # Agent storage utilities
│ ├── auth-storage.ts # AuthStorage class - API keys and OAuth
│ ├── artifacts.ts # Artifact management
│ ├── history-storage.ts # History storage
│ ├── messages.ts # Message types and transformers
│ ├── session-manager.ts # SessionManager class - JSONL persistence
│ ├── session-storage.ts # Session storage utilities
│ ├── storage-migration.ts # Storage migration
│ ├── streaming-output.ts # OutputSink with spill-to-disk for large outputs
│ └── compaction/ # Context compaction system
│ └── index.ts # Compaction logic, summary generation
├── ssh/ # SSH execution
│ ├── connection-manager.ts # SSH connection management
│ ├── ssh-executor.ts # executeSSH() with streaming
│ └── sshfs-mount.ts # SSHFS mounting
├── task/ # Task/subagent spawning
│ ├── index.ts # TaskTool exports
│ ├── agents.ts # Agent definitions
│ ├── commands.ts # Task commands
│ ├── discovery.ts # Task discovery
│ ├── executor.ts # In-process task execution
│ ├── render.ts # Task rendering
│ ├── subprocess-tool-registry.ts # Subprocess tool registry
│ ├── parallel.ts # Task concurrency helpers
│ └── template.ts # Task templating
├── tools/ # Built-in tool implementations
│ ├── index.ts # Tool exports, BUILTIN_TOOLS, createTools
│ ├── ask.ts # User input tool
│ ├── bash.ts # Bash command execution
│ ├── bash-interceptor.ts # Bash command interception
│ ├── calculator.ts # Calculator tool
│ ├── submit-result.ts # Submit result tool
│ ├── context.ts # Tool context utilities
│ ├── fetch.ts # URL content fetching
│ ├── find.ts # File search by glob
│ ├── gemini-image.ts # Gemini image generation
│ ├── grep.ts # Content search (regex/literal)
│ ├── ls.ts # Directory listing
│ ├── notebook.ts # Jupyter notebook editing
│ ├── output-meta.ts # OutputMetaBuilder, wrapToolWithMetaNotice
│ ├── output-utils.ts # TailBuffer, allocateOutputArtifact helpers
│ ├── path-utils.ts # Path resolution utilities
│ ├── python.ts # Python tool (delegates to ipy/)
│ ├── read.ts # File reading (text and images)
│ ├── renderers.ts # Tool output renderers
│ ├── render-utils.ts # Rendering utilities
│ ├── review.ts # Code review tools
│ ├── ssh.ts # SSH tool (delegates to ssh/)
│ ├── todo-write.ts # Todo management
│ ├── tool-errors.ts # Tool error types
│ ├── tool-result.ts # ToolResultBuilder fluent API
│ ├── truncate.ts # Output truncation utilities
│ └── write.ts # File writing
├── utils/ # Generic utilities
│ ├── changelog.ts # parseChangelog(), getNewEntries()
│ ├── event-bus.ts # Event bus for tool communication
│ ├── file-mentions.ts # File mention detection
│ ├── frontmatter.ts # Frontmatter parsing
│ ├── image-convert.ts # Image format conversion
│ ├── image-resize.ts # Image resizing utilities
│ ├── mime.ts # MIME type detection
│ ├── shell.ts # getShellConfig()
│ ├── terminal-notify.ts # Terminal notification utilities
│ ├── timings.ts # Performance timing utilities
│ ├── title-generator.ts # Session title generation
│ ├── tools-manager.ts # ensureTool() - download fd, etc.
│ └── utils.ts # Generic utilities
├── vendor/ # Vendored dependencies
│ └── photon/ # Photon image processing
└── web/ # Web tools (merged web-scrapers + web-search)
├── scrapers/ # Web scraping
│ ├── index.ts # Scraper exports
│ ├── types.ts # Scraper types
│ ├── utils.ts # Scraper utilities
│ └── (domain-specific scrapers)
└── search/ # Web search
├── index.ts # WebSearchTool exports
├── auth.ts # Search auth
├── render.ts # Search result rendering
└── providers/ # Search providers
├── anthropic.ts # Anthropic search
├── exa.ts # Exa search
└── perplexity.ts # Perplexity search
```
## Key Abstractions
### AgentSession (session/agent-session.ts)
The central abstraction that wraps the SDK Agent with:
- Session persistence (via SessionManager)
- Settings persistence (via SettingsManager)
- Model cycling with scoped models
- Context compaction
- Bash command execution
- Message queuing
- Hook integration
- Custom tool loading
- Extension/capability system integration
All three modes (interactive, print, rpc) use AgentSession.
### SDK (sdk.ts)
Wrapper around `@oh-my-pi/pi-agent-core` that provides a simplified interface for creating and managing agents programmatically. Used by AgentSession and available as a public API through index.ts exports.
### InteractiveMode (modes/interactive-mode.ts)
Handles TUI rendering and user interaction:
- Subscribes to AgentSession events
- Renders messages, tool executions, streaming
- Manages editor, selectors, key handlers
- Delegates all business logic to AgentSession
### RPC Mode (modes/rpc/)
Headless operation via JSON protocol over stdin/stdout:
- **rpc-mode.ts**: `runRpcMode()` function that listens for JSON commands on stdin and emits responses/events on stdout
- **rpc-types.ts**: Typed protocol definitions (`RpcCommand`, `RpcResponse`, `RpcSessionState`)
- **rpc-client.ts**: `RpcClient` class for spawning the agent as a subprocess and controlling it programmatically
The RPC mode exposes the full AgentSession API via JSON commands. See [docs/rpc.md](docs/rpc.md) for protocol documentation.
### SessionManager (session/session-manager.ts)
Handles session persistence:
- JSONL format for append-only writes
- Session file location management
- Message loading/saving
- Model/thinking level persistence
### SettingsManager (config/settings-manager.ts)
Handles user preferences:
- Default model/provider
- Theme selection
- Queue mode
- Thinking block visibility
- Compaction settings
- Hook/custom tool paths
- Thinking budgets (`thinkingBudgets` setting for custom token budgets per level)
- Image blocking (`blockImages` setting to prevent images from being sent to LLM)
### Hook System (extensibility/hooks/)
Extensibility layer for intercepting agent behavior:
- **loader.ts**: Discovers and loads hooks from `~/.omp/agent/hooks/`, `.omp/hooks/`, and CLI
- **runner.ts**: Dispatches events to registered hooks
- **types.ts**: Event types (`session`, `tool_call`, `tool_result`, `message`, `error`, `user_bash`)
See [docs/hooks.md](docs/hooks.md) for full documentation.
### Extension System Architecture
The extension system uses a shared runtime pattern:
1. **ExtensionRuntime** (`extensibility/extensions/types.ts`): Shared state and action handlers for all extensions
2. **Extension**: Per-extension registration data (handlers, tools, commands, shortcuts)
3. **ExtensionAPI**: Per-extension API that writes registrations to Extension and delegates actions to runtime
4. **ExtensionRunner**: Orchestrates event dispatch and provides context to handlers
Key extension events:
- `before_agent_start`: Receives `systemPrompt` and can return full replacement (not just append)
- `user_bash`: Intercept `!`/`!!` commands for custom execution (e.g., remote SSH)
- `session_shutdown`: Cleanup notification before exit
### Custom Tools (extensibility/custom-tools/)
System for adding LLM-callable tools:
- **loader.ts**: Discovers and loads tools from `~/.omp/agent/tools/`, `.omp/tools/`, and CLI
- **types.ts**: `CustomToolFactory`, `CustomToolDefinition`, `CustomToolResult`
See [docs/custom-tools.md](docs/custom-tools.md) for full documentation.
### Skills (extensibility/skills.ts)
On-demand capability packages:
- Discovers SKILL.md files from multiple locations
- Provides specialized workflows and instructions
- Loaded when task matches description
See [docs/skills.md](docs/skills.md) for full documentation.
### Capability System (capability/)
Unified extension system that discovers and loads capabilities from multiple sources:
- **Extension Discovery** (discovery/): Discovers extensions from Claude.md, .cursorrules, .codex, MCP servers, etc.
- **Capability Types**: Hooks, tools, context files, rules, skills, slash commands, system prompts, etc.
- **Multi-source**: Global (~/.omp/), project (.omp/), and built-in capabilities
See [docs/extensions.md](docs/extensions.md) for full documentation.
### Tool Output Infrastructure (tools/)
Standardized system for handling tool outputs with truncation, streaming, and artifact storage.
#### Core Components
**OutputSink** (`session/streaming-output.ts`): Line-buffered output collector with automatic spill-to-disk.
- Tracks total lines/bytes as data flows through
- Spills to artifact file when memory threshold exceeded
- Produces `OutputSummary` with truncation metadata
- Used by executors (bash, python, ssh) for streaming command output
**TailBuffer** (`tools/output-utils.ts`): Simple rolling buffer keeping the last N bytes.
- Used for UI preview during streaming
- Handles UTF-8 boundaries correctly
- Lightweight alternative to OutputSink for non-spilling tools
**ToolResultBuilder** (`tools/tool-result.ts`): Fluent builder for constructing `AgentToolResult`.
- Chains `.text()`, `.truncation*()`, `.limits()`, `.sourcePath()` methods
- Automatically populates `details.meta` with structured metadata
- Produces well-formed result via `.done()`
**OutputMetaBuilder** (`tools/output-meta.ts`): Fluent builder for `OutputMeta` structure.
- Methods for truncation info, limits, source paths, diagnostics
- Accepts `TruncationResult` (from truncate.ts) or `OutputSummary` (from OutputSink)
- Returns `undefined` if no metadata to report (empty case)
**wrapToolWithMetaNotice** (`tools/output-meta.ts`): Tool wrapper applied to all built-in tools.
- Automatically appends truncation/limit notices from `details.meta`
- Catches exceptions and renders them via `ToolError.render()`
#### Standard Pattern for Streaming Tools
Tools that produce potentially large streaming output (bash, python, ssh):
```typescript
async execute(...): Promise<AgentToolResult<MyDetails>> {
// 1. Allocate artifact path for full output storage
const { artifactPath, artifactId } = await allocateOutputArtifact(this.session, "toolname");
// 2. Create tail buffer for UI preview
const tailBuffer = createTailBuffer(DEFAULT_MAX_BYTES);
// 3. Execute with streaming callback
const result = await executeCommand({
artifactPath,
artifactId,
onChunk: (chunk) => {
tailBuffer.append(chunk);
onUpdate?.({
content: [{ type: "text", text: tailBuffer.text() }],
details: {},
});
},
});
// 4. Build result with truncation metadata
return toolResult<MyDetails>({})
.text(result.output)
.truncationFromSummary(result, { direction: "tail" })
.done();
}
```
#### Standard Pattern for Non-Streaming Tools
Tools that produce output in one shot (grep, find, ls):
```typescript
async execute(...): Promise<AgentToolResult<MyDetails>> {
const { items, limitReached } = await doWork();
const output = formatItems(items);
return toolResult<MyDetails>({})
.text(output)
.limits({ resultLimit: limitReached ? effectiveLimit : undefined })
.done();
}
```
#### Key Principles
1. **Always allocate artifact before execution** — ensures path exists for spill
2. **Pass onChunk to OutputSink/executor** — required for live UI updates during streaming
3. **Use ToolResultBuilder for all results** — ensures consistent metadata structure
4. **Close file sinks on error** — prevent descriptor leaks in failure paths
5. **Truncation direction matters** — `"tail"` for command output (show recent), `"head"` for file reads (show beginning)
## Development Workflow
### Running in Development
Run the CLI directly with bun (this is a bun-based project):
```bash
# From monorepo root
bun run dev
# Or run directly
bun packages/coding-agent/src/cli.ts
# With arguments
bun packages/coding-agent/src/cli.ts --help
bun packages/coding-agent/src/cli.ts -p "Hello"
# RPC mode
bun packages/coding-agent/src/cli.ts --mode rpc --no-session
```
### Type Checking
```bash
# From monorepo root (runs biome + tsgo type check)
bun run check
# From packages/coding-agent
bun run check
```
### Building
```bash
# Type check and build (from packages/coding-agent)
bun run build
# Build standalone binary
bun run build:binary
```
### Testing
```bash
# Run tests (from packages/coding-agent)
bun test
# Run specific test pattern
bun test --testNamePattern="RPC"
# Run RPC example interactively
bun test/rpc-example.ts
```
### Managed Binaries
Tools like `fd` and `rg` are auto-downloaded to `~/.omp/bin/` (migrated from `~/.omp/agent/tools/`).
## Adding New Features
### Adding a New Slash Command
1. If it's a UI-only command (e.g., `/theme`), add handler in `interactive-mode.ts` `setupEditorSubmitHandler()`
2. If it needs session logic, add method to `AgentSession` and call from mode
### Adding a New Tool
1. Create tool factory in `tools/` following existing patterns (e.g., `createMyTool(session: ToolSession)`)
2. Export factory and types from `tools/index.ts`
3. Add to `BUILTIN_TOOLS` map in `tools/index.ts`
4. Add tool prompt template to `prompts/tools/` if needed
5. Tool will automatically be included in system prompt
6. Use `ToolResultBuilder` for results and `OutputMeta` for truncation/limit metadata (see "Tool Output Infrastructure" section)
7. For streaming tools: use `allocateOutputArtifact()` + `createTailBuffer()` pattern with `onChunk` callback
### Adding a New Hook Event
1. Add event type to hook event types in `extensibility/hooks/types.ts`
2. Add emission point in relevant code (AgentSession, tool wrapper, etc.)
3. Update `docs/hooks.md` with the new event type
### Adding a New RPC Command
1. Add command type to `RpcCommand` union in `modes/rpc/rpc-types.ts`
2. Add response type to `RpcResponse` union in `modes/rpc/rpc-types.ts`
3. Add handler case in `handleCommand()` switch in `modes/rpc/rpc-mode.ts`
4. Add client method in `RpcClient` class in `modes/rpc/rpc-client.ts`
5. Update `docs/rpc.md` with the new command
### Adding a New Selector
1. Create component in `modes/components/`
2. Use `showSelector()` helper in `interactive-mode.ts`:
```typescript
private showMySelector(): void {
this.showSelector((done) => {
const selector = new MySelectorComponent(
// ... params
(result) => {
// Handle selection
done();
this.showStatus(`Selected: ${result}`);
},
() => {
done();
this.ui.requestRender();
},
);
return { component: selector, focus: selector.getSelectList() };
});
}
```
### Adding a New Extension Source
1. Create discovery module in `discovery/` (e.g., `my-source.ts`)
2. Implement discovery functions that return capability objects
3. Add to discovery chain in `discovery/index.ts`
4. Update `docs/extension-loading.md` with the new source
### Adding a New Capability Type
1. Create capability module in `capability/` (e.g., `my-capability.ts`)
2. Define capability type and schema
3. Add to capability registry in `capability/index.ts`
4. Add loader/handler in relevant core module
5. Update `docs/extensions.md` with the new capability type
### Adding a New Keybinding
1. Add the action name to `AppAction` type in `config/keybindings.ts`
2. Add default binding to `DEFAULT_APP_KEYBINDINGS`
3. Add to `APP_ACTIONS` array
4. Handle the action in `CustomEditor` or `InteractiveMode`
Example: The `dequeue` action (`Alt+Up`) restores queued messages to the editor.
## Code Style
- TypeScript with strict type checking (tsgo)
- No `any` types unless absolutely necessary
- No inline dynamic imports
- Formatting via Biome (`bun run check` or `bun run fix`)
- Keep InteractiveMode focused on UI, delegate logic to AgentSession
- Use event bus for tool/extension communication
- Components should override `invalidate()` to rebuild on theme changes
## Package Structure
This is part of a monorepo with the following packages:
- `@oh-my-pi/pi-coding-agent` (this package) - Main CLI and TUI
- `@oh-my-pi/pi-agent-core` - Core agent implementation
- `@oh-my-pi/pi-tui` - TUI components
- `@oh-my-pi/pi-ai` - External AI provider library
## CLI Flags
Key CLI flags for development:
- `--no-tools`: Disable all built-in tools (extension-only setups)
- `--no-extensions`: Disable extension discovery (explicit `-e` paths still work)
- `--no-skills`: Disable skill discovery
- `--session <id>`: Resume by session ID prefix (UUID match) or path
## SDK Exports
The SDK (`src/index.ts`) exports run modes for programmatic usage:
- `InteractiveMode`: Full TUI mode
- `runPrintMode()`: Non-interactive, process messages and exit
- `runRpcMode()`: JSON stdin/stdout protocol
Extension types and utilities are also exported for building custom extensions.
## Documentation
See the `docs/` directory for detailed documentation:
- `docs/sdk.md` - SDK usage and examples
- `docs/rpc.md` - RPC protocol documentation
- `docs/hooks.md` - Hook system documentation
- `docs/extensions.md` - Extension system documentation
- `docs/custom-tools.md` - Custom tool development
- `docs/skills.md` - Skill system documentation
- `docs/compaction.md` - Context compaction system
- `docs/session.md` - Session management
- `docs/theme.md` - Theme customization
- `docs/tui.md` - TUI architecture
Reference in New Issue View Git Blame Copy Permalink