Download Latest Version flow-next v8.4.0 source code.zip (14.1 MB) Google Add to Preferred Sources
Home / flow-next-v8.2.0
Name Modified Size InfoDownloads / Week
Parent folder
flow-next v8.2.0 source code.tar.gz < 16 hours ago 11.4 MB
flow-next v8.2.0 source code.zip < 16 hours ago 14.1 MB
README.md < 16 hours ago 2.0 kB
Totals: 3 Items   25.5 MB 0

Product owners and business analysts can run their own part of the spec without an engineer driving. One line to /flow-next:flow starts a business interview that keeps to product decisions, in plain words, and hands anything technical to engineering. Reviews that fall back to a smaller Claude model now reach Haiku 5.5, which is faster and costs less than Haiku 4.5.

Changed

  • Product owners and business analysts get a shorter, plainer interview. Tell /flow-next:flow your role ("I'm the product owner, interview me on fn-12", or /flow-next:flow fn-12 --biz) and it runs refine with the business lens, even on a spec already marked ready. Under a business lens, refine leaves data models, migrations, contracts, interfaces and security to engineering, and asks only about a consequence you own, in your words. "Not my call" records the question for engineering instead of counting as a skip, so it adds no skipped-items checkpoint question. The read-back and summary leave out internal terms. Refine without a lens, and the technical lens, ask what they asked before. The teams guide has a four-step path for this role, including starting from a requirements document you already have (for teams).
  • Review fallbacks and the Copilot triage judge reach Claude Haiku 5.5. The opt-in Copilot triage judge (FLOW_TRIAGE_LLM=1 with --backend copilot) now defaults to claude-haiku-5.5 instead of claude-haiku-4.5. Haiku 5.5 is Anthropic's fastest model and costs less than Haiku 4.5. When a stronger model is unavailable, the claude review backend now tries claude-haiku-5-5 before claude-haiku-4-5. The Copilot review ranking also gains claude-opus-5.5, claude-sonnet-5.5, claude-sonnet-5 and claude-haiku-5.5, so a Copilot review can step down to a current Claude model before an older one. A model you name explicitly still wins. A CLI that does not serve an id rejects it, and the ladder steps past it to the next model.
Source: README.md, updated 2026-10-09