BuildSeniorData / MLAuto-graded

Build a pipeline that hands state between agents

Multi-step automation only works if each step sees the last one's output and one failure doesn't halt everything.Real agent systems are rarely one giant prompt. They are a chain of specialists — planner, retriever, writer, critic — where each stage reads what the previous ones produced and passes an updated shared state forward. The code that threads that state, stops the chain early when a stage says the work is done, and keeps going when a stage blows up is the pipeline runner. The one that ships here does none of those three things reliably. Make run_pipeline(agents, state) carry state forward, honour an early finish, and survive a raising agent.

35 min

python

Start lab

You're done when

The pipeline runs the agents in the order it was given, and completes without crashing.
Each agent sees the state the previous agents wrote — state is threaded forward, not reset between stages.
The final state contains every write made along the chain, not just the last agent's.
When an agent signals it is done, the remaining agents are skipped rather than run anyway.
That done signal is still visible in the final state the caller receives.
An agent that raises does not kill the run — the pipeline continues to the next stage.
The failure is recorded as structured data, and the following agent still sees the state as it was before the failure.

What it measures

Agent engineeringComposing specialists into a chain that behaves predictably, rather than hoping a single large prompt covers every case.

Context engineeringDeciding what each stage needs to see and making state flow forward intact, which is what separates a working handoff from a chain that silently drops work.

Designing for partial failureTreating a flaky stage as normal rather than exceptional — isolating it, recording it in a form the next stage can use, and continuing.

Your environment

Terminal + editorA full shell in the browser

Claude CodePre-authenticated, no key setup

Isolated microVMYours alone, wiped afterwards

Start this lab

35 min

Start labYou'll sign in to Anthropos to start. The lab opens in a private cloud workspace with Claude Code ready to use.

AI Labs is part of Anthropos Workforce. Not on your plan yet? Book a demo