Roadmap

Where we're headed

Wiggum is the plug-and-play Ralph loop runner. Currently in Beta, latest release v0.18.3. Last updated 2026-04-08.

Milestones

Alpha

Core Ralph loop: init, spec generation, and autonomous execution

100%
10 of 10 complete
Beta

Feature complete: agent mode, GitHub integration, headless CLI, observability, multi-CLI adapters

90%
9 of 10 complete
V1.0

Release ready: native agent, MCP integration, parallel loops, sandbox execution

0%
0 of 6 complete

What's happening

Now 2
#131 In Progress
Agent token consumption reduction
Agent mode uses too many tokens — replace push-based strategic docs with pull-based tool access and enforce constraints.
AI
#145 In Progress
Feature loop optimization
Further reduce token usage and improve loop speed — fix double test runs, sleep overhead, and config loading.
Core
Next 4
#158 Planned
Dedicated worktree per loop
Run each loop in its own git worktree for true isolation — deterministic cleanup and no conflicts between concurrent runs.
Core
#134 Planned
GitHub App integration
Use a GitHub App identity for agent bot operations — proper attribution, higher rate limits, and fine-grained permissions.
Integrations
#92 Planned
Session persistence & crash recovery
Save loop state to disk so crashed or interrupted loops can resume from the last checkpoint.
Core
#71 Planned
Parallel loop scheduler
Run Manager UI for concurrent loops — focus, stop, and view logs across multiple running Ralph loops.
Platform
Later 4
#73 Planned
Native coding agent (hybrid mode)
Built-in coding agent for planning and review phases, with external CLI agents handling implementation — reducing API costs and improving quality.
AI
#95 Planned
Wiggum as MCP server
Expose Wiggum's core capabilities (scan, interview, run) as MCP tools — usable from any MCP-compatible client.
Integrations
#94 Planned
Wiggum as skill.md
Distribute scanning and interview as agent skills — any coding agent can invoke Wiggum's spec generation natively.
Integrations
#96 Planned
Sandbox execution
Run loop processes in isolated sandboxes (bubblewrap on Linux, sandbox-exec on macOS) for security.
Infrastructure

Latest releases

View all releases →
v0.18.3 2026-04-08
Security hardening, npm OIDC publishing & agent scheduler
  • Agent backlog scheduler orchestration — automated issue-to-loop pipeline
  • Security hardening: CodeQL fixes, Dependabot remediation, hardened logging/workflows
v0.17.3 2026-03-18
Loop optimization: --resume, merged phases & trimmed prompts
  • Use --resume for implementation, E2E, and review loop iterations 2+ — massive token savings
  • Merged verification phase into review phase to cut one full agent round
v0.17.2 2026-03-09
Agent --issues flag for targeted issue processing
  • Added --issues flag to target specific GitHub issues by number in agent mode