A shared registry so your agent windows stop lying to each other

What it does Squadra turns several coding-agent windows into a team with a paper trail. One window coordinates, others build, and a window that didn’t write the code reviews it. Assignments, decisions, executed checks and review reports live in one shared registry on your machine — not in a chat’s fading memory. A local bilingual (English/Italian) dashboard shows progress, evidence, and where it’s waiting on you.

The interesting bit Proofs are bound to the exact source: if the code changes after a check, the proof and review no longer count. That’s the boring, mechanical part — and it’s the whole reason to trust the rest. It also tracks token spend per window, model and task, and suggests which window should take the next task, computed locally.

Key highlights

• Works across Claude Code, Codex and Cursor, with a plugin store for community MCP plugins
• Tasks start only after the recipient acknowledges them; questions come with options, a recommendation and the exact reply sentence
• Built-in skills for recalling lessons, simplifying code and forecasting work
• Python standard library only at runtime; runs fully offline, no account, MIT licence
• Cisco security scan workflow on the repo, with browsable verified audits

Caveats

• macOS/POSIX is the tested environment; Windows process controls and recovery with submodules, LFS or sparse checkouts are explicitly not qualified
• Node.js is required for selftest and the dashboard’s JS checks — they fail without it, never skip
• Version migration notes are long and fussy: pre-3.10.1 registries can’t go back once updated, and some upgrades require every window to move together

Verdict Worth a look if you routinely run multiple agent windows on one repo and want accountability instead of vibes. If you’re a one-window person, this is coordination machinery you don’t need.