Problem
When a workflow specifies a retired model (e.g., gpt-5-codex), gh-aw accepts it, compiles the workflow, and passes the literal string to the Copilot CLI. The resulting startup failure looks like a provider auth error, not a model configuration error, confusing workflow authors.
Context
Upstream report: github/gh-aw#35032
Root Cause
gh-aw's model validation happens after AWF provider initialization. When the model alias resolves to a retired model, the Copilot CLI returns an auth-style error because it cannot load the model endpoint. The error message does not mention the model name.
Proposed Solution
- In the gh-aw compiler/engine initialization, resolve model aliases to canonical names before starting the AWF provider.
- Query the Copilot model catalogue at compile time; if the resolved model is absent or retired, emit a clear error:
model 'gpt-5-codex' is retired — use one of: [list].
- Fail fast before AWF containers are started, so users see the actionable error immediately.
- Add a test asserting that a retired model name produces the new error message.
Generated by Firewall Issue Dispatcher · sonnet46 1.2M · ◷
Problem
When a workflow specifies a retired model (e.g.,
gpt-5-codex), gh-aw accepts it, compiles the workflow, and passes the literal string to the Copilot CLI. The resulting startup failure looks like a provider auth error, not a model configuration error, confusing workflow authors.Context
Upstream report: github/gh-aw#35032
Root Cause
gh-aw's model validation happens after AWF provider initialization. When the model alias resolves to a retired model, the Copilot CLI returns an auth-style error because it cannot load the model endpoint. The error message does not mention the model name.
Proposed Solution
model 'gpt-5-codex' is retired — use one of: [list].