Runnable configs that put your coding agent on rails — clone one, point Claude Code or Cursor at it, and watch the skipped step stop being a move it can make. The coding-agent examples lead; start there if you're new. New to flowgate? See the three ways to run it — from inside Claude Code to flowgate's own runtime.
examples/tdd/
Your agent loves to write the implementation first and backfill a
test that always passes. Here it can't: the red → green → refactor
cycle is the workflow, so there's no move to write production code
until a test is actually failing. Try to jump the queue and you get
INVALID_TRANSITION — the step simply isn't on the menu.
Dogfooded in CI with a mechanical driver.
examples/deploy-pipeline/
The agent that ships before CI is green can't here — deploy
isn't a move until lint, test, and build have all passed. Those three
steps run back-to-back on their own: the agent calls
flowgate.command once and gets the result at the deploy
decision — three real commands run, zero extra round trips to the model.
examples/content-publish/
Same rails, a non-coding workflow: draft → brand review → human approval → publish. The agent walks the steps on its own, but at the approval gate it stops cold — that move is reserved for a human, and the agent literally can't make it.
examples/expense-approval/
How far the guards stretch: a multi-tenant expense workflow with two-tier approval. Small expenses need one manager; large ones need sign-off from several approvers before they clear. Payment is idempotent — the executor won't double-pay on a retry.
examples/
Small, single-pattern configs to copy from: simple-proxy.yaml
(pass a tool straight through), governed-change.yaml (a single
approval gate), import-and-discovery.yaml (pull tools in from
other MCP servers). Good starting points when you already know the shape you want.