Show HN: Stepbook – run long agent tasks as short steps with saved state

Category: ai-ml

Tags: ai-agents, llm, workflow-orchestration, state-management, claude-code

Score: 7.3/10 (Innovation: 7, Technical: 8, Documentation: 8, Utility: 6)

Stepbook runs long agent tasks as a series of short, stateless steps where each step is a fresh agent session that reads and updates a small saved state, enabling pause/resume and handoff between agents. It adds a journal, git commits, and schema-validated state patches on top of the SKILL.state approach, with built-in skills for checklists and eval loops. The project is well-documented and technically ambitious, targeting tasks with verifiable completion criteria.

Target audience: backend devs, AI engineers, devops

Repository: https://github.com/bahdan111/stepbook · Python · Apache-2.0 · 3 stars

View on Hacker News