UnodeAi

About

UnodeAI exists because multi-agent coding runs produce unverifiable claims: one agent reports 'done' and a second model is asked to grade it, which proves little. The extension structures delegation instead around observed evidence — commands run, files touched, approvals granted — with verdicts ascending a ladder from 'no evidence' to 'coordinator accepted,' and rejections requiring a recorded reason that visibly amends earlier verdicts. A project-manager agent plans and assigns tasks to specialist role agents, each running with configured tools, provider routes, and folder access that task scoping can narrow but never widen; optional worktree isolation and an approved verify command gate merges. Runs export as client-readable Markdown reports or privacy-stripped JSON. Developers and agencies delegating work to AI crews — and needing receipts for clients or auditors — use it in VS Code and Cursor; it is MIT-licensed, telemetry-free, and runs entirely on the user's own providers and keys.