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
Quick start
# Create a plan interactively
# Or bootstrap from an existing project
# Validate the plan
# Preview output
# Generate artifacts
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 |
diff |
Compare two plan files |
resume |
Recover an interrupted orchestrator loop |
retro |
Generate improvement suggestions from PROGRESS.md |
split |
Split an oversized task into smaller units |
templates |
Manage reusable task templates |
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
├── features.json
└── tasks/
├── T01-{slug}.md
├── T02-{slug}.md
└── ...
.vscode/
├── orchestrator.prompt.md
└── evaluator.prompt.md # only when [evaluator] is configured
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.mdto find the next incomplete task - Open the corresponding
tasks/T{NN}-{slug}.mdfile - Spawn a subagent to implement the task
- Run preflight checks (build, test, lint + security audit) to verify the work
- Independently verify — re-runs preflight before trusting the subagent's completion mark
- Updates
features.jsonwith per-criterion pass/fail results - If an
[evaluator]agent is configured, spawns it to score the task independently - Mark the task complete in
PROGRESS.mdand record learnings - Repeat until all tasks are done
In VS Code with Copilot, copy orchestrator.prompt.md into your project as a prompt file:
Then start Copilot in agent mode and send:
Read
.github/orchestrator.prompt.mdand follow its instructions. Begin by readingPROGRESS.mdto identify the next incomplete task, then execute it. After each task passes preflight, updatePROGRESS.mdand continue to the next task.
Monitor progress in a separate terminal with:
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, must-haves, and gates.
Gates (human-in-the-loop stops)
Add a gate to any task to require human confirmation before the orchestrator proceeds:
[[]]
= "deploy"
= "Confirm staging tests passed before the orchestrator runs this task."
# ... rest of task
The generated task file opens with a ⛔ GATE banner, and the orchestrator prompt instructs the loop to stop and wait for confirmation before marking the task in-progress.
Language support
Rust, Go, TypeScript, Python, Java, C#, Kotlin, Swift, Ruby, Elixir — each with idiomatic defaults for build, test, lint, and security audit commands.
Security
Wiggum bakes security into every generated plan at three levels:
1. Non-negotiable rules in every subagent prompt
Six OWASP-derived rules are injected automatically into the ## Security section of every task file and orchestrator prompt — covering secrets management, parameterised queries, HTTP security headers, rate limiting, file upload validation, and SSRF prevention. You don't have to add them; they're always there.
2. Vulnerability audit in every preflight
The language profile's audit command (cargo audit, govulncheck, npm audit, pip-audit, etc.) is appended to every task's preflight chain and exit criteria. Supply-chain CVEs are checked on every task completion, not just at the end. Override or disable per-plan with preflight.audit.
3. Automatic security hardening task
When your plan contains web-facing surface (detected from task slugs containing http, api, server, webhook, upload, auth, etc.), Wiggum auto-appends a security-hardening task as the final task in your plan. Its must_haves and evaluation_criteria map directly to the six OWASP categories with concrete, verifiable conditions. Suppress with [security] skip_hardening_task = true if you're handling security separately.
# Opt out of the auto-injected security task if desired
[]
= true
Documentation
Full docs: greysquirr3l.github.io/wiggum
License
Dual-licensed under MIT or Apache-2.0.