why-does-this-exist

`wde` - Decode the why behind legacy code

A CLI tool that traces git blame → PRs → issues → reviews and explains legacy code decisions in plain English using AI.

$ wde src/utils/parser.ts:142

Tracing blame... ████████████████ Done
Fetching PR #234... ████████████████ Done
Fetching issues... ████████████████ Done
Generating explanation...

┌─────────────────────────────────────────────────────────────────────────┐
│                                                                         │
│  This null check was added to fix a production crash (Issue #198)       │
│  where parser.config could be undefined when called before init().      │
│  The team considered making init() synchronous but rejected it due      │
│  to startup performance concerns noted in PR #234 review comments.      │
│  Watch out: removing this check will break lazy initialization.         │
│                                                                         │
│  Sources: commit a3f9c21 • PR #234 • Issue #198                         │
│                                                                         │
└─────────────────────────────────────────────────────────────────────────┘

The Problem

Every developer eventually stares at code where:

The commit message says fix
The author left the company
There are zero comments
The what is obvious, but the why is buried

The answer exists somewhere — scattered across a PR description, three issue threads, and review comments from 2 years ago. wde reconstructs that trail automatically.

Why not just use Claude Code CLI?

You could ask Claude Code: "Why does this code exist?" But here's what you'd have to do:

# With Claude Code CLI (manual process)
claude
> read src/utils/parser.ts
> run git blame -L 142,142 src/utils/parser.ts
> "what's the PR for commit a3f9c21?"
> run gh pr view 234
> "what issues were linked?"
> run gh issue view 198
> "now explain why this code exists"

With wde, it's one command:

wde src/utils/parser.ts:142

The Difference

	Claude Code CLI	`wde`
Purpose	General-purpose AI assistant	Built specifically for code archaeology
Context gathering	Manual — you fetch and paste	Automatic — traces blame → PR → issues
Workflow	Interactive session	Single command, instant answer
Token usage	High (full conversation)	Low (optimized prompt ~8k tokens)
Output	Conversational	Structured with source citations
Cost per query	~$0.05-0.15	~$0.01 (or free with Ollama)
CI/Tooling integration	Limited	`--json` flag for automation
Provider flexibility	Claude only	Claude, GPT, or local LLMs

When to use what

Use `wde` when...	Use Claude Code when...
"Why was this written this way?"	"Refactor this function"
"What problem did this solve?"	"Add a new feature"
"Is it safe to delete this?"	"Fix this bug"
"What were the alternatives considered?"	"Write tests for this"

wde is a scalpel. Claude Code is a Swiss Army knife. Use the right tool for the job.

How Context is Gathered

wde works in layers — local git data is always available, GitHub API adds richer context when accessible:

Layer 1: Local Git (No auth required)

Everything from your cloned repo:

git blame -L 142,142 file.ts    # → commit SHA, author, date
git show <sha>                   # → commit message + full diff
git log --oneline               # → commit history

This alone is useful — commit messages and diffs often contain valuable context.

Layer 2: GitHub API (Auth required for private repos)

The real gold is in PR discussions:

Data	Where it lives	Why it matters
PR description	GitHub only	Detailed explanation of why
Review comments	GitHub only	Debates, alternatives considered
Linked issues	GitHub only	Original bug reports, requirements
Issue comments	GitHub only	User reports, reproduction steps

Authentication Logic

┌─────────────────────────────────────────────────────────┐
│  AI Provider                                            │
│    ├─ Anthropic → Needs API key (keychain or env)       │
│    ├─ OpenAI → Needs API key (keychain or env)          │
│    └─ Ollama → No API key needed (local)                │
└─────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────┐
│  GitHub Context                                         │
│    ├─ Token set → Full context (local + PR + issues)    │
│    └─ No token                                          │
│         ├─ Public repo → Full context (rate-limited)    │
│         └─ Private repo → Local-only mode               │
└─────────────────────────────────────────────────────────┘

Flags for Control

# Full context (default) - uses GitHub API if available
wde src/file.ts:42

# Local only - skip GitHub entirely, works offline
wde src/file.ts:42 --local

No auth? No problem. The tool degrades gracefully — you still get explanations based on commit history and diffs.

Features

Feature	Description
Git Blame Tracing	Automatically identifies the commit that introduced a line
PR Context Fetching	Retrieves PR title, body, labels, and all review comments
Issue Linking	Extracts and fetches all linked issues (`Fixes #123`, `Closes #456`)
Multi-Provider AI	Choose between Claude, GPT, or local LLMs via Ollama
Secure Auth	API keys stored in system keychain, not config files
Streaming Output	See AI response as it's generated
Function Lookup	Use `--fn functionName` to trace by function instead of line number
JSON Output	`--json` flag for editor/tooling integration

Installation

# Using npm
npm install -g why-does-this-exist

# Using bun
bun install -g why-does-this-exist

# Or run directly with npx
npx why-does-this-exist src/file.ts:42

Setup

Run the interactive setup command:

wde auth

This will guide you through:

Choose your AI provider — Anthropic (Claude), OpenAI (GPT), or Ollama (local)
Enter your API key — Securely stored in your system keychain (not in config files!)
Optional: GitHub token — For private repo PR/issue context

$ wde auth

  wde - Authentication Setup

? Select AI provider:
  > Anthropic (Claude) - Best for reasoning over messy PR/issue text (recommended)
    OpenAI (GPT) - GPT-4o and other OpenAI models
    Ollama (Local) - Run locally with Llama, Mistral, etc. (free, no API key)

? Enter your Anthropic API key: sk-ant-***
✓ API key stored securely in system keychain

? Set up GitHub token? (optional, for private repos)
  > Yes
    No

? Enter your GitHub token: ghp_***
✓ GitHub token stored securely

✓ Setup complete! Run: wde src/file.ts:42

Where are credentials stored?

Credentials are stored in your system keychain — not in plain text files:

Platform	Storage
macOS	Keychain Access
Windows	Credential Manager
Linux	libsecret (GNOME Keyring / KDE Wallet)

This means:

✅ Safe to commit your ~/.config/wde/config.json (contains only preferences, no secrets)
✅ Credentials are encrypted at rest
✅ Works with existing system security policies

Environment Variable Override

You can also use environment variables (useful for CI/CD):

export ANTHROPIC_API_KEY="sk-ant-..."  # or OPENAI_API_KEY
export GITHUB_TOKEN="ghp_..."          # optional

Environment variables take precedence over keychain credentials.

Usage

# Explain a specific line
wde src/utils/parser.ts:142

# Explain a function (fuzzy matches function name)
wde --fn parseConfig src/utils/parser.ts

# Output as JSON (for editor integrations)
wde src/file.ts:50 --json

# Show full context sent to AI
wde src/file.ts:50 --verbose

# Use a different AI provider
wde src/file.ts:50 --provider openai
wde src/file.ts:50 --provider ollama

# Use a different model
wde src/file.ts:50 --model claude-haiku-4-20250514
wde src/file.ts:50 --provider openai --model gpt-4o-mini

Flags

Flag	Description
`--fn <name>`	Find function by name instead of line number
`--provider <name>`	AI provider: `anthropic`, `openai`, or `ollama`
`--local`	Use only local git data (no GitHub API, works offline)
`--json`	Output structured JSON
`--verbose`	Show the full context trail sent to AI
`--model <model>`	Model to use (default depends on provider)
`--help`	Show help
`--version`	Show version

Subcommands

Command	Description
`wde auth`	Interactive setup for API keys and GitHub token
`wde auth --status`	Check current credential status
`wde auth --logout`	Clear all stored credentials

How It Works

┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│  CLI Input   │────▶│  Git Blame   │────▶│  PR Number   │
│  file:line   │     │  + Commit    │     │  Extraction  │
└──────────────┘     └──────────────┘     └──────────────┘
                                                  │
                                                  ▼
┌──────────────┐     ┌──────────────┐     ┌──────────────┐
│   Output     │◀────│ AI Provider  │◀────│  GitHub API  │
│  Explanation │     │ Claude/GPT/  │     │  PR + Issues │
└──────────────┘     │   Ollama     │     └──────────────┘
                     └──────────────┘

Blame → Runs git blame -L N,N --porcelain to find the commit
Commit → Fetches full commit message and diff via git show
PR Detection → Parses commit message for PR references (#123, merge commits)
GitHub Fetch → Retrieves PR body, review comments, and linked issues
Context Assembly → Builds a structured prompt within token budget
AI Explanation → Your chosen AI provider synthesizes into a clear explanation
Output → Streaming response with source citations

Tech Stack

Component	Technology	Why
Runtime	Bun	Fast CLI startup, native TS, `$\`...`` shell helpers
Language	TypeScript	Type-safe API responses, better DX
AI	Claude / GPT / Ollama	Multi-provider support for flexibility
Credentials	keytar	Secure system keychain storage
CLI Parser	citty	Lightweight, zero deps, TypeScript-first
Colors	picocolors	Zero deps, respects `NO_COLOR`
Testing	bun:test	Built-in, no config overhead
CI/CD	GitHub Actions	Test on push, publish on tag

MVP Scope

In Scope (v1.0)

Git blame on file + line number
GitHub PR + review comment fetching
Linked issue extraction + fetching
Claude API explanation (3-5 sentences)
Function name lookup (--fn)
Plain stdout + JSON output

Added in v2.0

Multi-provider support (Anthropic, OpenAI, Ollama)
Local model support via Ollama
Secure keychain credential storage
Interactive wde auth setup
Streaming AI responses

Planned (v3.0)

Edge Cases Handled

Scenario	Behavior
No PR found	Falls back to commit message + diff (still useful)
Private repo, no token	Clear error with instructions
Rate limit hit	Shows remaining limit, suggests token
Massive PR (500+ comments)	Truncates to most relevant content
Non-GitHub remote	Friendly "GitLab support coming in v2" message

Contributing

# Clone and install
git clone https://github.com/zain534102/why-does-this-exist.git
cd why-does-this-exist
bun install

# Set up authentication
bun run src/cli.ts auth

# Run locally
bun run src/cli.ts src/example.ts:10

# Run tests
bun test

# Type check
bun run typecheck

License

MIT License - see LICENSE for details.

Built by Zain Ali

Stop guessing. Start understanding.

Name		Name	Last commit message	Last commit date
Latest commit History 39 Commits
.agents/skills		.agents/skills
.claude/skills		.claude/skills
.github		.github
docs		docs
scripts		scripts
src		src
tests		tests
.env.example		.env.example
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CLAUDE.md		CLAUDE.md
LICENSE		LICENSE
README.md		README.md
bun.lock		bun.lock
bunfig.toml		bunfig.toml
package.json		package.json
tsconfig.json		tsconfig.json

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

why-does-this-exist

`wde` - Decode the why behind legacy code

The Problem

Why not just use Claude Code CLI?

The Difference

When to use what

How Context is Gathered

Layer 1: Local Git (No auth required)

Layer 2: GitHub API (Auth required for private repos)

Authentication Logic

Flags for Control

Features

Installation

Setup

Where are credentials stored?

Environment Variable Override

Usage

Flags

Subcommands

How It Works

Tech Stack

MVP Scope

In Scope (v1.0)

Added in v2.0

Planned (v3.0)

Edge Cases Handled

Contributing

License

About

Uh oh!

Releases 2

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

why-does-this-exist

wde - Decode the why behind legacy code

The Problem

Why not just use Claude Code CLI?

The Difference

When to use what

How Context is Gathered

Layer 1: Local Git (No auth required)

Layer 2: GitHub API (Auth required for private repos)

Authentication Logic

Flags for Control

Features

Installation

Setup

Where are credentials stored?

Environment Variable Override

Usage

Flags

Subcommands

How It Works

Tech Stack

MVP Scope

In Scope (v1.0)

Added in v2.0

Planned (v3.0)

Edge Cases Handled

Contributing

License

About

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases 2

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`wde` - Decode the why behind legacy code

Packages