Cianplan

Level 5: AI / LLM systems · L5.5 Agents, tool use & MCP

Safe side effects: idempotency and approvals

Make an agent's actions safe with idempotency, allow-lists and human approval.

12 min · Pro

What you take away

  • An agent cannot distinguish a failed call from a lost response, so every side-effecting tool needs an idempotency key derived from the run and the step.

And 3 more inside the class, in Pro.

How the class is laid out

  1. The problem
  2. "Ask before anything risky"
  3. Three layers, in the order you build them
  4. Workflow or agent
  5. The loop
  6. Tools
  7. MCP
  8. Safe side effects
  9. Check your understanding