Ouroboros

Every coding agent runs the same loop for everything. Ouroboros doesn't: a Band room of six agents designs the verification loop your specific task needs, checks, critics, a human gate, then executes it on your real repo, fixing the file until your real tests go green. "Loop engineering" is the 2026 shift (Addy Osmani & Boris Cherny): the unit of agentic coding is no longer the prompt, it's the loop — generate, check, revise, until a real exit condition holds. But everyone hand-builds those loops, and every coding-agent tool runs the SAME fixed pipeline for everything. A CSS tweak and a database migration get the identical loop. That's vibe-coding with extra steps. Ouroboros, the snake that eats its own tail, automates loop engineering as a band of agents. Six specialists collaborate THROUGH Band: @LoopArchitect reads the task and synthesizes the loop it needs (which checks gate it, which critics vote, whether a human must sign); @LoopCritic attacks the exit condition and recruits a specialist on demand (a SecurityCritic when the task touches auth) via band_add_participant; @LoopRunner drives it; @CodeAuthor patches the code; @QAEngineer runs the real tests; @RivalReviewer (a different model) bounces failures back. Every step is an @mention handoff on the Band transcript — the loop's design and its execution share one audit trail. It's real, not a story. Point it at your repo and your test command (python band_repo.py --repo ... --test "pytest -q" --file ...). @QAEngineer has a tool that runs your actual suite; @CodeAuthor has tools that read and patch your actual file. The loop runs until your real tests pass — or hands you a verified failure instead of a confident wrong answer. Verified end-to-end: the live room fixed a real pytest bug, patched the file, and only finalized once the suite went green. Cross-framework (Pydantic AI + LangGraph), AI/ML API for reasoning, Featherless for the rival reviewer.