Support provider pricing overlays in apiProxy configuration

[dsfaccini]

This issue was posted by Codex Desktop using gpt-5.6-sol on behalf of David.

Background

While validating pydantic/pydantic-ai#6761, we found that the compiler-side fix in github/gh-aw#48107 cannot activate against the latest AWF release, v0.27.42. gh-aw currently gates emission of provider pricing behind AWFAPIProxyProvidersMinVersion=v0.27.43 because v0.27.42 does not accept the new configuration.

The original behavior was reported in github/gh-aw#47365.

Problem

AWF's configuration schema and runtime do not yet support provider pricing overlays under apiProxy.providers.

As a result, custom/BYOK model pricing configured in a gh-aw workflow cannot reach the API proxy. With a normal maxAiCredits budget, an unknown model fails with unknown_model_ai_credits. The same applies to the separate threat-detection API proxy, which can fail and discard otherwise safe workflow output.

gh-aw must therefore suppress apiProxy.providers while targeting AWF v0.27.42, and downstream workflows must use max-ai-credits: -1 as a workaround.

Expected behavior

• AWF accepts apiProxy.providers in its configuration schema.
• AWF forwards those provider/model pricing entries to the API proxy pricing resolver.
• The configuration works for both the primary agent and threat-detection API proxy.
• A release containing this support (expected by gh-aw as v0.27.43 or later) allows Propagate models.providers pricing into AWF apiProxy (including threat-detection runs) gh-aw#48107 to emit the configuration.

References

• models.providers custom pricing never reaches the api-proxy; custom/BYOK models still 400 with unknown_model_ai_credits gh-aw#47365 — original provider-pricing and threat-detection report
• Propagate models.providers pricing into AWF apiProxy (including threat-detection runs) gh-aw#48107 — merged gh-aw compiler support
• Run CI Review only after CI passes and separate douwebot pydantic/pydantic-ai#6761 — downstream workflow where this was reproduced