BHIL AI-First Development Toolkit
A production-oriented methodology repo: traceable PRD → SPEC → ADR → TASK chains, sprint loops, and AI-native patterns from an external practitioner.
https://github.com/camalus/BHIL-AI-First-Development-Toolkit
What this is
The BHIL AI-First Development Toolkit is a methodology repository for building AI-native apps in iterative sprints, with AI agents as primary implementors and humans as architects, reviewers, and decision-makers. It is not Nimbus software—it is an independent, well-documented system by Barry Hurd Intelligence Lab (BHIL) that we list here as a useful contrast and complement.
Insight into the repo
The core idea is a machine-readable artifact chain: PRD (what) → SPEC (how) → ADR (why) → TASK (steps) → code → review → deploy, with traceability IDs linking parents and children. That matches a philosophy we share: the bottleneck is specification quality, not raw code generation. The repo includes guides (getting started, methodology, sprint workflow, context management, AI-native patterns), templates for PRDs, specs, ADRs, tasks, sprints, eval suites, and guardrails, plus ADR variants for model selection, prompt strategy, and agent orchestration.
It is optimized for Claude Code and related tooling (RuFlo, RuVector are mentioned in the ecosystem story) but the patterns are portable.
Benefits
- End-to-end traceability — Every artifact knows its place in the chain; good for audits and retrospectives.
- Sprint-shaped rhythm — Explicit sprint workflow and templates keep scope and documentation moving together.
- AI-specific decisions as first-class — Model, prompt, and orchestration ADRs are built in, not afterthoughts.
- Solo-friendly — Includes guidance for review cadence and daily workflow when you do not have a large team.
We include it on the Lab as a peer setup: different origin, similar problems—structured intent, clear handoffs, and humans at the approval points.