Ship the Agent

assembling the full plan-execute-track-stop loop

Part of: Multi-Step Task Agent

Every piece is built: a planner that turns a goal into ordered steps, a tool registry, dependency ordering, result references, a bounded state log, retries with required/optional handling, and stop conditions. This lesson wires all of it into one TaskAgent and ships it. The shape of the finished agent Every line of run traces back to a lesson you've already built: topo order (lesson 3), resolve args (lesson 4), run step safe with required/optional handling (lesson 6). The step budget and loop detection from lesson 7 slot into a replanning variant of this same loop, checked once per iteration before deciding whether to plan another step. What "shipped" means here The same three tests as every product in this track. It runs from a clean start with one command. It survives an empty or malformed plan without crashing. And someone else could hand it a goal working from your instructions alone. An agent that plans, runs steps in dependency order, resolves references, tracks bounded state, retries while telling required failures from optional ones, and stops itself has crossed the line from demo to deliverable. Into your Portfolio Finishing this lesson records the Multi-Step Task Agent in

Challenge: Run the Full Agent Loop