reversa-brainstorm

Run a staged ideation pipeline that separates problem framing, option exploration, risk premortem, and decision convergence into distinct steps. Each stage produces a Markdown artifact, so the reasoning is documented before any specification or code exists.

Progress is tracked in an active-ideation.json file recording the session directory, current stage, and context. The actual stage is derived from which artifact files exist on disk, and disk state wins if the metadata diverges.

Yes. The pipeline detects whether a reverse-engineering extraction exists in the output folder and registers the context as legacy, anchoring ideation in what was already mapped. Greenfield sessions without extraction are equally supported.

When a session is already in progress, the orchestrator presents a menu to continue from where it stopped, open a parallel session, or reopen a specific stage. The previous session files are never deleted or modified without explicit confirmation.

The orchestrator only routes between stages and never writes the pipeline documents itself, so each stage depends on its dedicated agent. It also never produces code during ideation, which is by design but means implementation must wait for downstream workflows.