[awf] API proxy: validate Copilot model before agent launch to prevent misleading auth errors

[lpcox]

Problem

When a workflow sets engine.model: gpt-5-codex (a retired model), AWF passes the literal string as COPILOT_MODEL to the Copilot CLI without validation. The agent fails with a provider authentication-style error rather than a clear "model unavailable" message, misleading operators into thinking their API keys are wrong.

Context

Source issue: github/gh-aw#35032

Affected config:

engine:
  id: copilot
  model: gpt-5-codex

Root Cause

pkg/workflow/engine.go:285-289 copies the raw YAML engine.model string into EngineConfig.Model without resolving aliases or checking availability. AWF then surfaces an opaque auth failure from the Copilot process instead of a model-resolution error.

Proposed Solution

• In src/cli.ts (or a new src/model-validator.ts), add pre-flight model validation against a known list of supported/retired models before starting containers.
• When --enable-api-proxy is active (containers/api-proxy/), the API proxy should return a 400 Bad Request with a descriptive body when the requested model is retired or unknown, rather than forwarding a request that produces an auth-flavored error.
• Log a clear warning: "Model 'gpt-5-codex' is retired; did you mean 'gpt-5.3-codex'?" before exiting with a non-zero code.

Generated by Firewall Issue Dispatcher · sonnet46 2.5M · ◷

[github-actions]

👋 Hello! I noticed this issue might be a duplicate of an existing issue:

• [awf] engine: Validate and resolve COPILOT_MODEL aliases before launch to avoid misleading auth errors #4270 — [awf] engine: Validate and resolve COPILOT_MODEL aliases before launch to avoid misleading auth errors (closed as completed on 2026-06-04)

Both issues describe the same problem: when engine.model is set to a retired model (e.g. gpt-5-codex), AWF passes it unvalidated to the Copilot CLI, causing an opaque auth-style error instead of a clear model-resolution failure. Both reference the same upstream source issue and propose the same fix (pre-flight model validation + a descriptive error message).

If #4270's resolution already addressed this, this issue may be redundant. If there are aspects not covered — particularly the API proxy returning a 400 Bad Request for retired models — please clarify how this issue differs!

This is an automated message from the issue duplication detector.

Generated by Issue Duplication Detector for issue #4566 · sonnet46 552.3K · ◷