Hey Product Hunt, I built FlowGuard because I kept seeing AI coding agents fix a local bug while damaging the larger workflow. The failure pattern is usually not "the function doesn't run." It is something more stateful: the same object gets processed twice, a retry repeats a side effect, cache drifts from the source of truth, a module writes state it should not own, or a workflow has an exit path but no real progress guarantee. FlowGuard is an open-source finite-state workflow simulator for those moments. Before an AI agent changes risky workflow behavior, it can model the boundary as: ```text Input x State -> Set(Output x State) ``` Then FlowGuard enumerates the reachable paths and reports invariant failures, loop/progress issues, contract problems, and counterexample traces. It is not an LLM wrapper, not a prompt tool, and not a replacement for tests. It is meant to sit next to planning/spec/code-review/bug-fix skills as a behavior cross-check before implementation. The easiest way to try it is to give the GitHub repo to a capable coding agent and ask it to install the source checkout, read the included `model-first-function-flow` skill, and use it on a real stateful workflow. Repo: https://github.com/liuyingxuvka/... One important note: this project is source-install only right now. The PyPI package named `flowguard` is a different project, so please use the GitHub repo directly. The feedback I care about most: After trying FlowGuard on a real AI-agent workflow, does it catch a pain point you actually have, and does it work as a useful cross-check alongside your existing bug-fix, code-review, or planning skills rather than duplicating them?

FlowGuard

Simulate risky AI-agent workflows before coding

Simulate risky AI-agent workflows before coding