The Framework
ROME
Requirements-Oriented Multi-agent Engineering
A framework on Claude Code that takes software from written requirements to tested code through four gated phases, each run by a named agent, with every artefact traceable in both directions.
What it is
An orchestrator, Roma, dispatches one specialist agent per phase. A guard blocks the next phase until the QA agent, sarah, has passed the gate. Requirements are captured in AORDL: thirteen mandatory fields, no ambiguity carried forward.
Suited to
Builds where showing why the code is what it is matters as much as the code — regulated, audited or enterprise work. Too heavy for a spike or a prototype, by design.
What's next
[ Roadmap — two or three next capabilities, one line each — TO CONFIRM ]
Built by
William Paulton, developer and architect, London. In independent development since June 2025.
Reading
How ROME works, one piece at a time.
Each article takes a single aspect of the framework and explains it properly — what problem it solves, how it is built, and what it costs.
Processing user requirements into a set of formal designs with AORDL
How a sponsor's raw brief becomes thirteen mandatory fields, then a traced design — walked through on one real example, start to finish.
Read DraftHow quality, reproducibility and accuracy are maintained
Why the same specification run twice should give you the same system twice, and the machinery that makes that true rather than hopeful.
DraftHow ROME iterates, modifies and fixes defects
A bug is rarely a bug in the code. Tracing a failure back to the requirement that permitted it, and re-running the phases it touches.
PlannedMeet the family — sub-agents with names
Eleven named agents, and one rule that binds them: none may approve its own work. Why naming them changed how the framework was designed.
PlannedSoftware audits and quality gates
What a gate actually inspects before it will let a phase close, and what an audit trail looks like when the work was done by agents.