Reasonable Forge
Two brand sites live and the first product line in production, with its fits confirmed in print
The check Every part clears a printability rules gate before it's sliced, and a fit stays marked UNVERIFIED until it prints
Work
Each project leads with what happened and the check that makes it believable. Most repos are private; where one is public, it's linked.
Case studies
Reasonable UX
The eval picked the cheaper config: 2 wins, 1 loss
The check A separate eval harness scored each model tier instead of assuming the bigger one wins
Source on GitHub ↗QA Agent
84% fewer tokens per run (~55k down to ~8.8k)
The check Cost telemetry on every run, so the saving is measured, not estimated
Source on GitHub ↗Agent Panel
Its enforcement chain is open source as the reasonable-agents plugin (MIT)
The check The vault agent exposes only typed-JSON intents, so private context can't leak into public output
Source on GitHub ↗More projects
- Groove Empire Building
A rhythm and idle game in Godot 4, built as the first proof for a prompt-to-game pipeline.
- Reasonable Basis Building
Tax research grounded in IRS publications: RAG with mandatory citations, and a graceful refusal when the sources don't cover the question.
Open-source plugins
All plugins →- reasonable-agents v0.1
Harness-enforced adversarial review for Claude Code subagents. Source on GitHub ↗
Submitted to the Claude plugin directory in June 2026; it passed the automated security scan in September and is awaiting review.