← Back to labs index

Lab 11 — Loop engineering

Capability: Codex has an internal agent loop. --output-schema constrains the final response shape when the chosen model/provider supports it; it does not implement a validated, bounded retry/poll/refinement scheduler for you.

Install Codex CLI; create a private temporary CODEX_HOME, copy this lab’s config.toml.example to $CODEX_HOME/config.toml and fill in your endpoint (ending in /openai/v1) and deployment. Provide AZURE_OPENAI_API_KEY through your resource key from a secret manager. The Microsoft guide says Entra ID is not supported for Codex; never save the key in files. Workspace: start in lab11-loops/ and follow the common preflight. Keep the schema in the lab directory; only the generated report belongs in workspace/:

"service: demo-order`nport: 3000" |
  Set-Content (Join-Path $workdir 'service.txt') -Encoding utf8
codex exec --skip-git-repo-check -C "$workdir" --sandbox read-only `
  --output-schema (Join-Path $labdir 'service-map.schema.json') `
  --output-last-message (Join-Path $workdir 'report.json') `
  "Read service.txt. Return a JSON service map. Cite service.txt:1-2 as evidence."
Get-Content (Join-Path $workdir 'report.json') -Raw | ConvertFrom-Json

printf 'service: demo-order\nport: 3000\n' > "$workdir/service.txt"
codex exec --skip-git-repo-check -C "$workdir" --sandbox read-only \
  --output-schema "$labdir/service-map.schema.json" \
  --output-last-message "$workdir/report.json" \
  "Read service.txt. Return a JSON service map. Cite service.txt:1-2 as evidence."
python -m json.tool "$workdir/report.json"

The schema path is absolute so changing working directories cannot break it; if --output-schema is unsupported by the Foundry deployment, mark this part unavailable rather than asserting schema validity. Check that every name, port and citation matches the source: a JSON shape constraint does not validate facts.

For a bounded manual validation loop, if an answer is wrong, send the specific errors through codex exec resume --last "Correct only these errors: ..." from inside workspace/, or pass the explicit session ID from the first run, and recheck. Stop on success, repeated identical errors, or three attempts. Write down the exit reason. For retry, polling and refinement, draw the trigger/evaluator/state change and all three exits; do not execute side-effecting retries or unbounded polls against live resources. This is a manual exercise, not a native programmable loop API.

Follow reset and cleanup afterward.