Developer Tacsiazuma has released "shall-we," an open-source pre-engineering sanity check that challenges teams to justify their ideas before any code gets written. Hosted on GitHub, the tool implements the Agent Skills open standard and integrates directly into AI coding assistants like Claude Code, Cursor, GitHub Copilot, Gemini CLI, and OpenAI Codex. The project emerged from frustration with an industry-wide trend: features getting shipped because they're technically possible, not because anyone asked if they should exist in the first place.

The Problem With Building Everything

We're deep in another AI hype cycle where shipping has never been easier—and has never been more rampant. Grugbrain.dev (aka "the caveman") has spent years advocating against unnecessary complexity, arguing that sometimes the best code is the code you don't write. Meanwhile, architects who've survived multiple hype cycles keep warning about ideas that aren't bad, just badly timed, badly scoped, or solving problems nobody actually has. Shall-we operationalizes both perspectives at the PM layer, where pivoting away from a bad idea still costs nothing.

How Shall-We Works

When invoked, shall-we restates your proposed feature to confirm understanding, then asks targeted questions one at a time—stopping early if a deal-breaker surfaces. It evaluates ideas through four lenses: business value, market demand, complexity, and risk. The output is a structured verdict of YES, NO, or LATER with supporting reasoning. Strengths and concerns get listed explicitly, along with conditions that would need to be true for a different outcome. It's essentially a rubber duck conversation with actual opinions—something most teams desperately need before that "quick feature" becomes six months of technical debt.

Supported Agents and Platforms

Claude Code users can install via /plugin marketplace add Tacsiazuma/shall-we followed by /plugin install shall-we@shall-we, then invoke with /shall-we. Gemini CLI installation uses gemini skills install with the GitHub URL. Cursor automatically picks up skills cloned into ~/.cursor/skills/—no restart required. VS Code (Copilot) and OpenAI Codex both use .agents/skills directories within projects. Any other compatible agent can reference the skill directly from agentskills.io's standard path at skills/shall-we/SKILL.md.

Key Takeaways

  • Shall-we targets the decision layer before tickets, design docs, or code exist—where changing direction is free
  • The four evaluation lenses (business value, market demand, complexity, risk) provide structured rigor without heavy process overhead
  • Works with major AI coding assistants out of the box, with a standardized skill format for future compatibility
  • Inspired by grugbrain.dev's anti-complexity stance and experienced architects who've seen hype cycles come and go

The Bottom Line

This is the kind of tooling the ecosystem needed yesterday. As AI agents make it trivial to generate code, we desperately need something to slow down the "ship first, think later" reflex. Shall-we won't catch everything, but it's a cheap way to stress-test your next big idea before your sprint backlog owns you. Bookmark this one—your future self will thank you during the post-mortem.