1
Ask your agent to run a tiny debugger flow
2
Check the expected behavior
You should see the session created, the binary loaded, execution stop at
main, a full context snapshot returned, and one successful step.3
Fall back to manual tool calls if needed
If your agent cannot complete the flow, run the sequence below through any MCP inspector or compatible client.
./workspace/chal and the in-container tool path is /workspace/chal.
Common stepping commands are
c, n, s, ni, and si.