| name | foundry-agent-setup |
| description | Use when: changing Azure AI Foundry setup, model deployment validation, .agent_ids.json, agent staleness, create-agent, validate_setup.py, or supported GPT-5 model policy. |
Foundry Agent Setup Skill
Use this workflow for setup wizard, CLI validation, model policy, agent creation, and persisted agent metadata changes.
Current Policy
- Supported non-chat base models:
gpt-5.1, gpt-5.2, gpt-5.4, gpt-5.5.
- Block
chat, mini, nano, pro, and codex variants, including versioned siblings.
- Unknown underlying model is non-blocking only when Azure cannot report
model_name for an existing deployment.
.agent_ids.json is v2-only and requires role objects with name, version, and id.- Persist and compare
model_deployment_name, underlying_model_name, and agent_definition_signature.
- Missing, blank, mismatched, or stale agent metadata blocks generation until agents are recreated.
Workflow
- Start with
model_policy.py, error_utils.py, web_app.py, and validate_setup.py.
- Keep policy centralized; do not duplicate allow/block logic in routes.
- If agent definitions change, update the signature inputs and staleness tests.
- Keep setup UI copy concise and accurate.
Validation
Run:
.venv/bin/pytest tests/test_model_policy.py tests/test_agent_staleness.py tests/test_web_endpoints.py::TestModelDeploymentValidation -q
make test-quick
