Developer releases Stageflow, a configurable multi-stage pipeline for checking AI vibecoding slop

Stageflow is a pipeline tool that breaks complex agent jobs into stages, each with a fresh agent session doing one pre-configured task. LLMs power each stage, with clean handoffs keeping behavior scoped. It targets developers who want reliable agent workflows without drift, plus optional human approval gates.

A development tool that runs a fixed 4-stage pipeline (requirements, architecture, implementation, review) for feature work, requiring human approval at each stage. It uses LLMs to generate each stage's output, and if rejected, regenerates with user notes and previous attempt as context. Built for developers who want more control than autonomous agents but more automation than chat copilots.

A Parcha engineer launched AgentRun on Hacker News, an open-source DSL powering their grep.ai harness that converts repeatable parts of agent jobs into inspectable workflows. Steps can mix tool calls, code, LLM-driven routing decisions, and full agents only where investigation is needed, cutting the cost of running a complete agent loop. It ships with a scripted demo, GitHub examples, a Pi extension, and a background video.

Flows is a tool that guides AI builders through a structured workflow (Goal → Plan → Work → Check → Repair → Review → Release) to ensure completeness and correctness. It uses LLMs to generate plans and checklists, and an independent agent can follow the plan to build a working app. It solves the problem of AI-generated apps missing features like authentication or business logic.

A free tool that checks whether a prompt, SOP, workflow idea, article outline, or automation plan is clear enough to build around. It runs entirely in the browser, does not call an API, and does not store input, helping users avoid messy AI builds by catching vague goals and missing review steps early.