Checkpoints Documentation
Three checkpoint types, one rule: Claude automates everything it can, then stops precisely where human judgment is irreplaceable
Comprehensive checkpoint specification for Claude Code workflows defining three types (human-verify, decision, human-action) with detailed examples and execution protocols for managing human verification points, authentication gates, and the automation-first principle that makes checkpoints effective.
Also connected to
The rules for when an AI agent should stop retrying, stop guessing, and surface the problem to a human - before a fixable issue becomes an architectural mess
Senior frontend code reviewer agent for reviewing diffs with focus on correctness, UX,...
A four-minute explainer script that walks through why autonomous AI coding fails without orchestration - and how human-in-the-loop infrastructure fixes it
The agent that doesn't trust summaries - it reads the actual code, checks the actual wiring, and answers the only question that matters: did we achieve the goal or just complete the tasks?
The read-only auditor that answers one question: does what was built match what was planned - and if not, is the deviation justified or is it scope creep?