Core Package

@cdoing/core — The foundation layer containing tools, permissions, sandbox, context providers, indexing, and all core infrastructure.

Overview

The core package is the dependency-free foundation of the entire system. It has no internal package dependencies and exports everything needed by the ai, cli, and vscode-extension packages.

packages/core/src/ ├── index.ts # All exports ├── tools/ # 21 built-in tools │ ├── registry.ts # Central tool registry │ ├── file-read.ts # File reading (text, images, PDFs) │ ├── file-write.ts # File creation/overwrite │ ├── file-edit.ts # Find-and-replace editing │ ├── multi-edit.ts # Atomic batch edits │ ├── file-delete.ts # Permission-controlled deletion │ ├── file-run.ts # Auto-detect & run 14 languages │ ├── shell-exec.ts # Shell command execution │ ├── glob-search.ts # File pattern matching │ ├── grep-search.ts # Content search with regex │ ├── list-dir.ts # Directory listing │ ├── view-repo-map.ts # Repository tree view │ ├── view-diff.ts # Git diff viewer │ ├── web-fetch.ts # URL content fetching │ ├── web-search.ts # DuckDuckGo search │ ├── codebase-search.ts # FTS5 + vector search │ └── system-info.ts # Runtime info query ├── permissions/ # Permission engine │ └── index.ts # PermissionManager (738 lines) ├── sandbox/ # Filesystem & network sandbox │ ├── manager.ts # Orchestrator │ ├── filesystem.ts # Path checking │ ├── network.ts # Domain allowlisting │ └── types.ts # Configuration types ├── context-providers/ # 10 @ mention providers │ ├── registry.ts # Provider registry │ ├── git.ts # @git context │ ├── diff.ts # @diff context │ └── folder.ts # @folder context ├── indexing/ # Codebase indexing │ ├── indexer.ts # Main orchestration │ ├── database.ts # SQLite schema + queries │ ├── chunker.ts # Code-aware chunking │ └── types.ts # TypeScript interfaces ├── hooks/ # Pre/post tool hooks │ └── index.ts ├── rules/ # Project rules │ └── manager.ts ├── plan/ # Plan mode (read-only) │ └── manager.ts ├── mcp/ # Model Context Protocol │ └── manager.ts ├── effort/ # Effort level control │ └── index.ts ├── agents/ # Multi-agent coordination │ └── coordinator.ts └── utils/ # Utilities ├── path-safety.ts # Resolve paths safely ├── path-matching.ts # Glob pattern matching ├── shell-paths.ts # Extract paths from commands ├── search-match.ts # Find + replace, unified diff ├── gitignore.ts # .gitignore loader ├── project-config.ts # Config file loader ├── memory.ts # Conversation memory store └── todo.ts # Task tracking store

Tool Registry

All 21 tools are managed by the central ToolRegistry which provides:

register(tool) — Register a new tool instance
get(name) — Retrieve a tool by name
getAll() — Get all registered tools
execute(name, params) — Run a tool with parameters
getDefinitions() — Get JSON Schema definitions for LLM binding

Each tool implements the BaseTool interface with a ToolDefinition (name, description, parameters as JSON Schema) and an execute() method.

Context Providers

Context providers allow users to inject additional information into the conversation using @mentions:

Provider	Trigger	Description
Terminal	@terminal	Last terminal command output
Tree	@tree [path] [depth]	Workspace file tree
URL	@url <url>	Fetch and attach web page content
Codebase	@codebase <query>	Full-text + semantic search
Open Files	@open	All open editor files (VS Code only)
Problems	@problems	File diagnostics/errors (VS Code only)
Clipboard	@clipboard	Clipboard contents
File	@file <path>	Include a specific file
Git	@git	Git context (commits, branch info)
Diff	@diff	Current working changes

Indexing System

The indexing system provides intelligent code search using SQLite FTS5 for full-text search and optional vector embeddings for semantic search.

How Indexing Works

File Scanning

Scans the workspace respecting .gitignore. Skips node_modules, .git, dist, build, and files over 1MB. Supports 40+ file extensions.

Incremental Updates (SHA-256)

Content hashing detects changes. New files are chunked and indexed, modified files are re-indexed, unchanged files are skipped.

Code-Aware Chunking

Code files are split at function/class boundaries. Markdown splits at headers. Basic files split at ~17 lines (~384 tokens per chunk).

Search Pipeline

FTS5 search (BM25 ranking with 10x path boost) provides 35% of results. Vector embedding search provides 65%. Results are deduplicated by path + line.

Database Schema

-- File chunks with line ranges
CREATE TABLE chunks (
  id INTEGER PRIMARY KEY,
  path TEXT NOT NULL,
  content TEXT NOT NULL,
  startLine INTEGER,
  endLine INTEGER,
  hash TEXT NOT NULL
);

-- FTS5 virtual table (trigram tokenizer, BM25)
CREATE VIRTUAL TABLE fts USING fts5(
  path, content, tokenize='trigram'
);

-- Vector embeddings (JSON storage)
CREATE TABLE embeddings (
  chunk_id INTEGER REFERENCES chunks(id),
  vector TEXT NOT NULL  -- JSON array
);

-- Incremental index tracking
CREATE TABLE index_catalog (
  path TEXT PRIMARY KEY,
  hash TEXT NOT NULL,
  indexed_at TEXT
);

Utilities

Utility	Purpose
path-safety.ts	Resolve paths safely, prevent directory traversal escape
path-matching.ts	Glob pattern matching for permission rules
shell-paths.ts	Extract file paths from shell commands for permission checking
search-match.ts	Find-and-replace engine with unified diff support
gitignore.ts	Parse and apply .gitignore patterns
project-config.ts	Load project configuration files
memory.ts	Persistent conversation memory storage
todo.ts	Task tracking with status management

External Dependencies

Minimal Dependencies

The core package deliberately keeps its external dependency count low: better-sqlite3 (database), chalk (colors), glob (file patterns), minimatch (glob matching).