Overview

Relevant source files

• .github/workflows/claude.yml
• CLAUDE.md
• README.md
• main.go

This document provides a high-level introduction to the dump CLI tool, its architecture, and core capabilities. The dump tool aggregates content from local files, URLs, and tmux panes into structured formats (XML or Markdown) for consumption by Large Language Models.

For detailed information on specific subsystems:

• Core implementation details: see CLI Tool Architecture
• Content acquisition mechanisms: see Content Sources
• Filtering and transformation logic: see Processing Pipeline
• Build and deployment procedures: see Build and Deployment
• Development workflows: see Development

Purpose and Design

The dump tool serves as a context aggregator for LLM workflows. It recursively traverses directories, fetches web content via the Exa API, and captures tmux terminal panes, producing unified output suitable for LLM context windows. The tool implements concurrent processing with bounded worker pools and respects .gitignore rules while providing extensive filtering capabilities.

Sources: main.go1-24 README.md1-21 CLAUDE.md1-13

System Architecture

The application is implemented as a single-file Go program with a clear separation between acquisition, processing, and output stages.

Component Overview

Sources: main.go26-43 main.go99-125 main.go631-802 main.go804-825

Content Source Types

The tool acquires content from three distinct sources, each with its own processing mechanism.

Source	Entry Function	Data Structure	Concurrency Model
Local Files	processDirectory()	Item	Per-directory goroutines
URLs	fetchURLContent()	Item	Bounded worker pool (max 3)
Tmux Panes	capturePaneContent()	TmuxPaneItem	Bounded worker pool (max 6)

Local File System

Local files are processed through processDirectory216-330 which uses filepath.WalkDir to recursively traverse directories. The function respects .gitignore patterns via the go-gitignore library and filters files based on glob patterns and extensions with OR semantics.

Sources: main.go216-330 main.go47-67 main.go69-76 main.go189-214

URL Fetching

URLs are fetched concurrently through fetchURLsConcurrently332-372 which implements a bounded worker pool with rate limiting (350ms between requests). The function calls fetchURLContent526-582 to interact with the Exa API endpoint at https://api.exa.ai/contents. The API key is read from the EXA_API_KEY environment variable.

Sources: main.go332-372 main.go526-582 main.go45 main.go126-142

Tmux Pane Capture

Tmux panes are captured via fetchTmuxConcurrently470-524 which resolves selectors like current, all, %1, or 0.1 through resolveTmuxSelectors392-437 Content is captured using capturePaneContent454-467 which executes tmux capture-pane with configurable history lines.

Sources: main.go470-524 main.go392-437 main.go440-451 main.go454-467 main.go375-389

Processing Pipeline

The processing pipeline applies multiple filtering stages to content before output.

Filtering Stages

1. Binary Detection: isTextFile47-67 reads the first 512 bytes and validates UTF-8 encoding
2. Gitignore Rules: buildIgnoreList69-76 creates a matcher from .gitignore files
3. Pattern Matching: compilePatterns78-88 compiles glob patterns using gobwas/glob
4. Extension Matching: Case-insensitive extension filtering with OR semantics main.go273-276
5. Line Filtering: filterContent169-187 applies regex patterns to exclude matching lines

Sources: main.go47-67 main.go69-76 main.go78-97 main.go169-187 main.go263-281

Output Formats

The tool supports two output formats controlled by the --out-fmt flag.

Format Specification

Format	Flag Value	File Tag	Tmux Tag	Tree Tag
XML (default)	xml	<document path='...'>	<tmux_pane id='...'>	<tree path='...'>
Markdown	md	```path	```shell	```tree

Formatting Functions

The formatItem144-154 function handles both file and URL output, using <document> or custom XML tags. The formatTmuxItem156-166 function formats tmux panes with session metadata. Tree visualization is rendered by formatTreeOutput621-629 using formatTreeNode594-619 for recursive traversal.

Sources: main.go144-154 main.go156-166 main.go594-629

CLI Interface

The CLI is built using the Cobra framework with the main command defined in rootCmd804-825

Command-Line Flags

All flags are registered in init827-848 and processed in runDump631-802 The tool defaults to the current directory when no sources are specified.

Sources: main.go26-41 main.go804-825 main.go827-848 main.go631-802

Concurrency Model

The tool implements concurrent processing for performance while maintaining bounded resource usage.

Goroutine Pool	Max Workers	Implementation	Purpose
Directory Processing	Unlimited (1 per dir)	sync.WaitGroup	Parallel directory traversal
URL Fetching	3	Bounded worker pool	Rate-limited API requests
Tmux Capture	6	Bounded worker pool	Parallel pane capture

The runDump631-802 function orchestrates three concurrent operations:

• Directory processing at main.go698-751 using a goroutine per directory
• URL fetching at main.go643-657 with fetchURLsConcurrently
• Tmux capture at main.go769-785 with fetchTmuxConcurrently

Results are streamed through channels and written to stdout as they become available.

Sources: main.go631-802 main.go332-372 main.go470-524 main.go698-751

Dependencies

The tool relies on three external Go libraries:

Package	Purpose	Usage Location
github.com/spf13/cobra	CLI framework	main.go23 main.go804-825
github.com/gobwas/glob	Glob pattern matching	main.go21 main.go78-88
github.com/sabhiram/go-gitignore	Gitignore parsing	main.go22 main.go69-76

External system dependencies:

• Exa API: Requires EXA_API_KEY environment variable for URL fetching
• tmux: Must be in PATH for pane capture functionality

Sources: main.go3-24 main.go45 go.mod