Act as a senior AI infrastructure engineer. Build an interactive application called The AI Tool-Call Replay & Failure Lab that helps developers investigate failed tool calls made by AI agents.
Let users import or manually create execution traces containing model messages, tool names, arguments, results, timestamps, retries, and errors.
The application must:
Reconstruct the sequence of agent decisions and tool executions.
Detect invalid arguments, missing outputs, unexpected retries, and broken handoffs.
Highlight the earliest observable point where execution diverged from the expected workflow.
Allow users to edit a tool result and replay the remaining simulated steps.
Compare successful and failed execution traces.
Track retry counts, latency, and estimated tool-call costs.
Generate regression scenarios from previously observed failures.
Export traces and test cases as JSON.
Include realistic examples involving a failed database query, an invalid calendar argument, and a tool returning an unexpected response shape. Use deterministic simulation rather than claiming to execute a real agent. Build a working, responsive interface with filters, a timeline, and clear evidence-based explanations.
0 Comments