Open Source

Ship features
while you sleep

An open-source autonomous coding agent that plugs into any codebase. Scan your stack, generate specs through AI interviews, and ship via Ralph loops — from backlog to merged PR.

Get Started Free or
$ npm i -g wiggum-cli
wiggum — demo

Demo of Wiggum CLI: running wiggum init to scan a codebase, wiggum new to generate a spec through an AI interview, and wiggum run to execute an autonomous Ralph loop that plans, implements, tests, verifies, and creates a pull request.

Wiggum CLI is an open-source command-line tool that plugs into any codebase and runs autonomous coding loops — also known as Ralph loops. It scans your tech stack, generates detailed feature specifications through AI-powered interviews, and executes complete plan-implement-test-review-merge cycles using Claude Code, Codex, or any CLI-based coding agent. Free and self-hosted — install with npm i -g wiggum-cli.

Agent Mode
New in v0.17.0

From backlog to merged PR,
zero intervention

Wiggum reads your GitHub issues, generates specs, runs Ralph loops, reviews the output, and auto-merges — then moves to the next issue.

Run wiggum agent

Point Wiggum at your GitHub backlog and walk away. It prioritizes issues by label and dependency order, assesses feature state, generates specs, runs the full Ralph loop, reviews diffs against the spec, and auto-merges when all checks pass.

  • Prioritizes by label and respects dependency ordering
  • Skips shipped work, resumes partial branches
  • Full Ralph loop per issue: plan, implement, test, verify
  • Reviews diffs against spec, auto-merges when checks pass
  • Runs headless in CI: wiggum agent --headless
wiggum agent
Agent | Running | 3 issues
Model: claude-opus-4 | Elapsed: 42:18
#107 GitHub issue integration merged
#108 Headless wiggum new --auto P1
#112 Headless wiggum sync P2
#108 loop:
90% 11 1
42:15 review verifying diff against spec
41:52 test tests/new.test.ts ✓ pass
#107 reviewed & auto-merged to main
Interactive Mode
Step by step

Three commands, full control

Scan your codebase, generate a spec through an AI interview, and execute the Ralph loop — one feature at a time.

1

Scan with /init

Point Wiggum at any project and it maps your tech stack, structure, and conventions. Zero config — it generates the context that makes every spec and loop reliable.

  • Zero config — works on any project
  • Maps frameworks, languages, and conventions
  • Generates context that feeds into specs and loops
wiggum /init
Write(wiggum.config.cjs)
Created wiggum.config.cjs
Write(.wiggum/guides/AGENTS.md)
Created AGENTS.md
Write(.wiggum/guides/FRONTEND.md)
Created FRONTEND.md
... and 14 more files
Done. Created configuration files.
What's next:
/new <feature> Create a specification
2

Interview with /new

Describe what you want in plain English. Wiggum uses your full codebase context to interview you, then generates a detailed spec any CLI-based coding agent can execute — consistent results, every time.

  • Context-aware AI interviews using your codebase
  • Generates structured, implementation-ready specs
  • Documents key decisions and tradeoffs
wiggum /new
New Spec | Complete (5/5) | user-auth
Summary
- Goal: Add OAuth authentication with Google and GitHub providers
- Outcome: Spec written to .wiggum/specs/user-auth.md (387 lines)
Key decisions
1. Using NextAuth.js for OAuth handling
2. Storing sessions in database, not JWT
3. Adding role-based permissions from start
Done. Specification generated successfully.
Review the spec in your editor
/run user-auth Execute the spec
3

Execute with /run

Hand the spec to your coding agent and walk away. The Ralph loop technique runs autonomous iterations with progress checkpoints, delivering working code without constant intervention.

  • Autonomous iteration loops with checkpoints
  • Real-time progress and completion tracking
  • Works with Claude Code, Codex, or any CLI-based coding agent
wiggum /run user-auth
Run Loop | Running | user-auth
Phase: Running | Iter: 2/10 | Branch: feat/user-auth
Tokens: 12.4k | Elapsed: 7:28
Progress:
35% 14 26
Overall:
35% 14 26
Press Ctrl+C to stop the loop
Capabilities

Everything to go from idea to merged PR

Built for real codebases. No wrappers, no lock-in, no config files.

Core
Plug & play
Zero configuration. Point Wiggum at any project — any language, any framework — and it works. No setup files, no adapters, no boilerplate. Scan once, then generate specs and run loops immediately.
$ cd my-project && wiggum init
Detected Next.js 15 + TypeScript + Prisma
Context saved — ready to go
Core
Codebase scanning
Auto-detects your tech stack, project structure, test setup, linting rules, and coding conventions in seconds.
TypeScript Python Go Rust React Next.js Django Rails +more
AI
Spec-driven development
AI interviews grounded in your codebase generate structured, implementation-ready specs. Every spec includes tasks, test plans, and architectural decisions.
Specs are plain markdown — readable, editable, version-controllable. They work with any CLI-based coding agent.
Integrations
GitHub issue integration
Run /issue to browse your backlog in a navigable table picker. Select an issue and Wiggum takes you straight into the AI interview with full issue context pre-loaded. Or pass --issue #42 directly.
GitHub Issues │ owner/repo (4)
#titlestatelabels #145Optimize loop token usageopenperf #139Merge verification into reviewopenP1 #134GitHub App integrationopenfeature
(↑↓ navigate, Enter select, Esc cancel)
UX
TUI monitor
Real-time terminal dashboard with progress bars, token tracking, activity feed, and live diffs. See exactly what your agent is doing at every phase.
Run Loop | implement | user-auth
Iter: 2/10 | Tokens: 12.4k
████████████████████ 35%
Coming soon
Multi-model support
Works with Claude Code today. Codex CLI, Gemini CLI, and other terminal-based coding agents coming soon. Swap providers without changing your specs.
Specs are agent-agnostic markdown — your loops work the same regardless of which model runs them.
Platform
Headless CLI
Run Wiggum in CI pipelines, cron jobs, or other agents. Generate specs without interaction, sync context, and execute loops — all non-interactive.
$ wiggum new --auto --issue 42
$ wiggum sync
$ wiggum agent --headless

Your backlog won't clear itself

Install Wiggum, scan your codebase, and run your first autonomous Ralph loop in under 5 minutes. Open source, free forever. See the roadmap and changelog.

FAQ

Questions & answers

Everything you need to know about Wiggum CLI.

What is Wiggum CLI? +

Wiggum is an open-source CLI tool that plugs into any codebase and runs autonomous coding loops. It scans your tech stack, generates detailed feature specifications through AI-powered interviews, and executes implementation loops using Claude Code, Codex, or any CLI-based coding agent.

How does Wiggum work? +

Wiggum works in three steps: (1) /init scans your project to map frameworks, languages, and conventions — zero config required. (2) /new interviews you about what you want to build, using your codebase context to generate a detailed, implementation-ready spec. (3) /run hands the spec to your coding agent and runs autonomous iteration loops with progress checkpoints until the feature is complete.

Which coding agents does Wiggum support? +

Wiggum works with any CLI-based coding agent. It has been tested with Claude Code (Anthropic), OpenAI Codex CLI, and other terminal-based AI assistants. The specs it generates are agent-agnostic — they work with any tool that can read a markdown file and execute code.

Is Wiggum free to use? +

Yes. The CLI is open source and free forever. You bring your own API keys (e.g. Anthropic, OpenAI) and run everything locally. Pro plans ($19/mo+) add managed API keys, a web dashboard, push notifications, and priority support.

What is the Ralph loop technique? +

The Ralph loop is an autonomous coding methodology pioneered by Geoffrey Huntley. It breaks feature development into phases — planning, implementation, E2E testing, verification, and PR review — running each as an autonomous agent loop with checkpoints. Wiggum implements this technique as a CLI tool.

What is a Ralph loop in Claude Code? +

A Ralph loop in Claude Code is an autonomous coding cycle where Claude Code executes structured phases — plan, implement, test, verify, and PR — without manual intervention. Wiggum CLI automates this by generating specs and managing the loop lifecycle.

Is the Ralph loop the same as the Ralph Wiggum technique? +

Yes. "Ralph loop," "Ralph Wiggum loop," and "Ralph Wiggum technique" all refer to the same autonomous coding methodology pioneered by Geoffrey Huntley. The technique breaks feature development into structured agent phases with checkpoints. Learn more.

Can I use Ralph loops with AI agents other than Claude? +

Yes. The Ralph loop technique is agent-agnostic. Wiggum CLI generates specs in markdown that work with Claude Code, OpenAI Codex, Gemini CLI, OpenCode, or any CLI-based coding agent.

Does Wiggum work with any programming language or framework? +

Yes. Wiggum's /init command auto-detects your tech stack regardless of language or framework. It has been used with TypeScript, Python, Go, Rust, React, Next.js, Astro, Django, Rails, and many others. The generated specs adapt to your project's patterns and conventions.

What does agent mode do? +

Agent mode (wiggum agent) reads your GitHub backlog, picks issues by priority label (P0 > P1 > P2) and dependency order, generates implementation specs from issue context, runs the full Ralph loop per issue, reviews diffs against the spec, and auto-merges the PR when all checks pass — then moves to the next issue. It can also run headless in CI pipelines.

How does the GitHub issue integration work? +

Run /issue in Wiggum to browse your GitHub issues in a navigable table picker. Select an issue and Wiggum takes you straight into the AI interview with full issue context (title, body, labels) pre-loaded. You can also pass --issue #42 directly to wiggum new. In agent mode, issues are read from your backlog automatically.

How is Wiggum different from GitHub Copilot or Cursor? +

Copilot and Cursor assist with in-editor code completion. Wiggum operates at the feature level — it generates full specifications, then runs autonomous multi-step coding loops that implement, test, and create PRs without manual intervention. It complements editor tools rather than replacing them.

Can I use Wiggum on existing projects? +

Absolutely. Wiggum is designed to plug into existing codebases. The /init command scans your project structure, detects conventions, and generates context that ensures specs and loops respect your existing patterns. No migration or restructuring needed.

Where can I see the product roadmap? +

The full public roadmap is available at wiggum.app/roadmap. It shows what's currently being built, what's coming next, and milestone progress toward each release. All roadmap items link back to their GitHub issues for full transparency.

When will paid features (dashboard, notifications) be available? +

Web dashboard, push notifications, and managed API keys are planned but not yet shipped. They are tracked on the public roadmap. The free CLI — including stack scanning, spec generation, and autonomous loops — is fully functional today. You can join the waitlist on the pricing page to be notified when Pro features launch.