Anvil
An AI product-management factory that turns a one-line idea into a fully cross-referenced artifact tree — research to strategy to PRD to epics to stories to tests — and self-corrects its own gaps.
Overview
Product docs rot because nothing keeps them consistent with each other. Anvil runs your idea through an 11-stage topological pipeline and emits 19 typed artifacts — research, strategy, PRD, epics, stories, tests — each traceable to its parents, each scored for coverage, freshness, consistency and readiness.
When a stage finds a gap it doesn't shrug. It runs a clarification-and-self-reflection loop with hard convergence caps, and an escalation chain that optimizes context, then upgrades the model, then decomposes into multiple calls before giving up. Edit one artifact and a change-cascade engine walks the dependency graph to flag everything downstream. Built solo in about a week.
Highlights
- A distributed workflow engine driving a self-correcting, multi-model pipeline.
- Structured output end to end — no free-text parsing between stages.
- A composite 0–100 health score gates whether the tree is ready to ship.