AgentStep – declarative, reviewable agent workflows

Turn ephemeral code generation into reproducible, reviewable declarative steps

Problem

Heavy coding-agent users find prompts imperative and ephemeral, code becoming a black box, and intent getting lost over long horizons; pull requests, CI/CD, and GitHub itself have become process bottlenecks, which is why projects like Telos exist.

Solution

A declarative file describes goals, constraints, and acceptance steps; the agent produces reviewable artifacts and decision records at every step, with step gates, intent tracking, and replay to make the black-box process traceable and reproducible.

Users

Indie developers and small engineering teams who use Claude Code / Codex daily

MVP

Define the declarative step-file schema; emit step completion records and artifact diffs; enforce step-level gates before proceeding; run locally via CLI; export a review-ready checklist

Revenue

Open-source core plus team subscription for audit logs, collaboration, and CI integration; enterprise self-hosted license

Why now

Coding agents are now daily tools, but workflow tooling is still built for humans writing code by hand; indie developers understand this pain and can ship lightweight alternatives fastest