Wiggum

AI orchestration scaffold generator for the Ralph Wiggum loop.

Wiggum generates structured task files, progress trackers, and orchestrator prompts from a TOML plan definition — enabling autonomous AI coding loops where an orchestrator agent drives subagents through dependency-ordered tasks until a project is fully implemented.

Install

cargo install wiggum

Quick start

# Create a plan interactively
wiggum init

# Or bootstrap from an existing project
wiggum bootstrap /path/to/project

# Validate the plan
wiggum validate plan.toml --lint

# Preview output
wiggum generate plan.toml --dry-run

# Generate artifacts
wiggum generate plan.toml

Commands

Command	Description
`init`	Interactively create a new plan
`generate`	Generate task files, progress tracker, and orchestrator prompt
`validate`	Validate plan structure and dependency graph
`add-task`	Add a task to an existing plan
`bootstrap`	Generate a plan from an existing project
`serve --mcp`	Start the MCP server
`report`	Generate a post-execution report
`watch`	Live progress monitoring

Generated artifacts

project/
├── IMPLEMENTATION_PLAN.md
├── PROGRESS.md
├── AGENTS.md
├── orchestrator.prompt.md
└── tasks/
    ├── T01-{slug}.md
    ├── T02-{slug}.md
    └── ...

Running the loop

After generating artifacts, open your AI coding tool in agent mode and load the generated orchestrator.prompt.md as the prompt. The orchestrator will:

Read PROGRESS.md to find the next incomplete task
Open the corresponding tasks/T{NN}-{slug}.md file
Spawn a subagent to implement the task
Run preflight checks (build, test, lint) to verify the work
Mark the task complete in PROGRESS.md and record learnings
Repeat until all tasks are done

In VS Code with Copilot, copy orchestrator.prompt.md into your project as a prompt file:

cp orchestrator.prompt.md .github/orchestrator.prompt.md

Then start Copilot in agent mode and send:

Read .github/orchestrator.prompt.md and follow its instructions. Begin by reading PROGRESS.md to identify the next incomplete task, then execute it. After each task passes preflight, update PROGRESS.md and continue to the next task.

Monitor progress in a separate terminal with:

wiggum watch

Example plan

See reference/example-plan.toml for a fully annotated plan covering all supported fields — project metadata, preflight commands, orchestrator persona and rules, multiple phases with dependency wiring, and per-task hints, test hints, and must-haves.

Language support

Rust, Go, TypeScript, Python, Java, C#, Kotlin, Swift, Ruby, Elixir — each with idiomatic defaults for build, test, and lint commands.

Documentation

Full docs: greysquirr3l.github.io/wiggum

License

Dual-licensed under MIT or Apache-2.0.

wiggum 0.2.0