Skip to main content
The qckfx Agent SDK comes with a comprehensive set of built-in tools that provide agents with powerful capabilities for file system operations, shell execution, and code analysis. These tools are automatically available to every agent instance.

File System Tools

FileReadTool

Reads the contents of files in the filesystem with support for partial reading and line numbering.
const result = await agent.processQuery('Read the package.json file');
Features:
  • Handles text files with various encodings
  • Supports partial file reading with line offset and count
  • 500KB hard cap for performance and safety
  • Line numbering for easier code review
  • Pagination support for large files

FileEditTool

Modifies existing files by replacing specific content with precise targeting.
const result = await agent.processQuery('Replace the old function with a new implementation');
Features:
  • Ensures precise targeting of text to be replaced
  • Preserves file structure and formatting
  • Maintains file encodings during edits
  • Requires exact match for safety
  • Supports deletion by using empty string for replaceCode

FileWriteTool

Creates new files with specified content or completely replaces existing ones.
const result = await agent.processQuery('Create a new README.md file');
Features:
  • Creates new files or overwrites existing ones
  • Automatic parent directory creation
  • Various text encoding support
  • Safety checks for overwrite operations

LSTool

Lists files and directories with detailed information and filtering options.
const result = await agent.processQuery('List all files in the src directory');
Features:
  • Directory exploration capabilities
  • Hidden file visibility control
  • Detailed file information (size, dates, permissions)
  • Entry count limiting for performance

Search and Discovery Tools

GlobTool

Fast file pattern matching tool for finding files by name patterns across the codebase.
const result = await agent.processQuery('Find all TypeScript files in the project');
Features:
  • Powerful glob patterns for flexible matching
  • Cross-codebase file searching
  • Hidden file inclusion control
  • Directory filtering options
  • Result limiting for performance

GrepTool

Fast content search tool that searches file contents using patterns or regular expressions.
const result = await agent.processQuery('Find all functions that contain "create" in their name');
Features:
  • Regular expression support
  • Recursive directory traversal
  • File type filtering
  • Case-sensitive/insensitive options
  • Result limiting for performance

Shell Execution Tools

BashTool

Executes shell commands in the agent’s environment with full shell feature support.
const result = await agent.processQuery('Install npm dependencies');
Features:
  • Maintains state between command executions
  • Supports all standard shell features
  • Working directory specification
  • 100KB output limit per stream (stdout/stderr)
  • Environment variable persistence
  • Always requires permission for security
BashTool always requires user permission before execution for security reasons. This cannot be overridden even in fast-edit mode.

Coding Tools

ClaudeTool 🚀

The ultimate coding companion - Integrates Anthropic’s Claude Code CLI for agentic coding assistance directly within your qckfx agents.
💰 Huge Cost Savings for Anthropic Users! If you’re already on Anthropic Pro or Max plan, Claude Code can save you considerable money compared to paying API fees. This is especially powerful for multi-model workflows where you can use o3 or Gemini for planning and then have Sonnet in Claude Code drive the implementation.
const result = await agent.processQuery('Refactor the authentication module for better security');
Key Features:
  • Built-in by default - Available on every agent unless explicitly disabled
  • Automatic rate limit handling - Gracefully falls back to other tools when rate limits are hit
  • Multi-model workflows - Perfect for o3/Gemini planning → Sonnet implementation patterns
  • Headless operation - Runs claude -p "<prompt>" in headless mode
  • Cost optimization - Leverages your existing Anthropic subscription
Perfect for:
  • Code refactoring and optimization
  • Bug fixes and debugging
  • Feature implementation
  • Code review and analysis
  • Architecture improvements
The Claude tool is enabled by default on all agents. To disable it, explicitly list the tools you want in your agent configuration and exclude claude from the list.
The Claude tool requires the claude CLI to be installed and configured on your system. It also requires permission for execution as it runs shell commands.

Advanced Tools

BatchTool

Executes multiple tool invocations in a single request for improved performance and reduced latency.
const result = await agent.processQuery('Read multiple configuration files at once');
Features:
  • Parallel execution when possible
  • Serial execution for dependent operations
  • Reduced context usage and latency
  • Collected results from all invocations
  • Respects individual tool permissions

ThinkTool

Allows the model to perform self-reflection and complex reasoning without external actions.
const result = await agent.processQuery('Think about the best approach to refactor this code');
Features:
  • Self-reflection capabilities
  • Complex reasoning support
  • No external system changes
  • Thought logging and tracking
  • Brainstorming and analysis support

SubAgentTool (Beta)

Allows the use of specialized sub-agents for specific tasks.
const result = await agent.processQuery('Use a specialized agent to analyze this codebase');
Features:
  • Specialized agent delegation
  • Task-specific expertise
  • Isolated execution context
  • Natural language interface
  • Beta feature with ongoing development

Tool Categories

Tools are organized into categories for better organization and permission management:
  • FILE_OPERATIONS - FileReadTool, FileEditTool, FileWriteTool, LSTool
  • SEARCH_DISCOVERY - GlobTool, GrepTool
  • SHELL_EXECUTION - BashTool, ClaudeTool
  • BATCH_OPERATIONS - BatchTool
  • REFLECTION - ThinkTool
  • SUB_AGENTS - SubAgentTool

Permission System

Tools integrate with the agent’s permission system:
  • Normal Mode - Requests permission for potentially dangerous operations
  • Fast-Edit Mode - Allows file operations without approval, but shell commands still require permission
  • Dangerous Mode - Allows all operations without permission (use only in secure, sandboxed environments)
Use the appropriate permission mode for your use case. Fast-edit mode is ideal for development workflows, while dangerous mode should only be used in secure, automated environments.

Best Practices

File Operations

  • Use FileReadTool instead of cat, head, or tail for more reliable results
  • Include sufficient context in FileEditTool’s searchCode to ensure unique matches
  • Use FileWriteTool for creating new files, FileEditTool for targeted edits

Search Operations

  • Prefer GlobTool over find command for file discovery
  • Prefer GrepTool over grep command for content searching
  • Use specific patterns to limit results and improve performance

Shell Execution

  • Use BashTool sparingly and only when necessary
  • Prefer specialized tools (FileReadTool, GlobTool, etc.) over shell commands
  • Always verify commands in safe environments before using dangerous mode

Coding Assistance

  • Use ClaudeTool for complex coding tasks that benefit from agentic assistance
  • Leverage multi-model workflows: o3/Gemini for planning, Claude Code for implementation
  • Take advantage of cost savings if you have an Anthropic Pro or Max subscription
  • Let the tool handle rate limiting automatically with graceful fallbacks

Performance

  • Use BatchTool for multiple independent operations
  • Limit search results with maxResults parameters
  • Use specific patterns and paths to reduce search scope
I