graph-looplisted

# /graph-loop — draw the graph, then run it A graph is **a plan for your AI work, drawn so you can see it**: which jobs happen, and which job waits for which. Two parts only. - **Node** = one job. One agent, one bounded task, one input, one output. - **Edge** = one job needs what another produced. **An edge only counts when real data passes along it.** > Nodes do the thinking. Edges carry the results. That's the whole vocabulary. A node is only wire-able if it has a **contract**: ```text ▸ NODE CONTRACT JOB: one bounded job, nothing else IN: { ...defined fields } ← passed in, never assumed OUT: { ...defined fields } ← schema'd, not a wall of free text SCHEMA: enforced. free text → rejected and retried WHY: a defined output is what lets the NEXT node consume this one without a human in the middle. ``` ## The loop ### 1. DRAW — list the nodes Take the workflow the user already runs (or the goal they want) and write it out as numbered jobs. Most workflows arrive as a **chain** — "do A, then B, then C" — which is technically already a graph, just the saddest one: every node with one arrow in and one out. A chain has no redundancy; if C stalls, D never happens and A's work is stranded. ### 2. THE FAKE-EDGE TEST — the core move Walk the steps in order. At each one ask exactly one question: > **Does this step actually need the result of the step before it?** - **Yes** → real edge. Keep the order. - **No** → **there is no edge. The