BeforeDone runs configured checks, creates receipts bound to declared file scopes, rejects stale evidence, and builds incident reports from observable events and artifacts. It is a local guardrail for normal agent mistakes, not a sandbox or trust system.
Open-source protocol for AI agents doing longform industry research: brief gates, state files, source/claim discipline, staged drafting, review loops, hard stops, gotchas, and offline evals with regression fixtures.
