Bound an Agent Loop

Create the agent-lab directory

mkdir makes a new directory. Your agent's configuration will live here — one place to look when you ask "why did the agent do that?".

Cap the loop length

echo with the > redirect writes the setting into agent.yaml. max_iterations is the hard stop on the think-act-observe loop — after 10 rounds the agent gives up instead of looping forever on a problem it cannot solve.

Read the config back

cat prints the file. A one-line config is normal — the value is not its size but the fact that the limit is written down and versioned, not hardcoded somewhere in the code.

Declare the error policy

This setting goes into its own file, fallback.txt — one file per decision keeps each rule easy to find and swap. on_error=retry_once means a failed tool call gets exactly one second chance before the agent reports failure, instead of retrying silently forever.

Review the full setup

ls lists the directory so you can see both config files in place. Together they bound the agent in time (max_iterations) and in behavior (retry policy).

Submit for validation

The validator checks that agent.yaml and fallback.txt exist, that agent.yaml contains max_iterations=10, and that fallback.txt declares the retry policy.