Files

3.8 KiB

1. Proposal And Validation

  • 1.1 Review add-gpt-image-profile-compatibility and add-gpt-image-edit-support for overlap and migration impact.
  • 1.2 Validate update-image-api-compatibility-modes with OpenSpec tooling when the CLI is available.
  • 1.3 Approve the internal compatibility model before implementation starts.

2. Settings And Migration

  • 2.1 Extend ImageApiCompatibility to include tuzi-gpt-image.
  • 2.2 Accept legacy tuzi-compatible values and normalize them to tuzi-gpt-image.
  • 2.3 Preserve stored auto values and confirm they continue to resolve correctly for historical OpenAI profiles.
  • 2.4 Preserve openai-compatible-basic for rollback and generic gateway scenarios.
  • 2.5 Ensure snapshot, copy, import, and export flows preserve the compatibility field.

3. Routing Contract

  • 3.1 Update compatibility resolution so auto resolves to openai-gpt-image, tuzi-gpt-image, or openai-compatible-basic.
  • 3.2 Keep non-GPT image model routing unchanged.
  • 3.3 Add a dedicated Tuzi GPT generation request schema such as tuzi.image.gpt-generation-json.
  • 3.4 Route official GPT edit bindings only for the official GPT compatibility mode.
  • 3.5 Ensure resolved compatibility and request schema stay aligned in metadata and logs.

4. Adapter Ownership Cleanup

  • 4.1 Add tuzi-gpt-image-adapter.ts.
  • 4.2 Move GPT-specific Tuzi translation logic out of default-adapters.ts.
  • 4.3 Keep gpt-image-adapter responsible only for official GPT request schemas.
  • 4.4 Keep the default/basic adapter as the generic fallback for openai-compatible-basic.
  • 4.5 Reuse the shared size/quality resolver where helpful, but keep official and Tuzi contract semantics separate.

5. UI And Diagnostics

  • 5.1 Simplify the main profile UI to emphasize OpenAI GPT Image and Tuzi GPT 兼容.
  • 5.2 Show a resolved summary when a profile remains on auto.
  • 5.3 Keep hidden or advanced fallback modes round-trippable for legacy profiles.
  • 5.4 Expose enough debug information to understand stored mode, resolved mode, request schema, and adapter selection without leaking secrets.

6. Defaults And Migration Guardrails

  • 6.1 Default newly created provider profiles to openai-gpt-image.
  • 6.2 Default built-in managed provider profiles to openai-gpt-image when no stored override exists.
  • 6.3 Preserve explicit stored compatibility values when managed profiles are rebuilt or reopened.
  • 6.4 Upgrade only missing compatibility fields to openai-gpt-image, while preserving explicit custom-profile auto.
  • 6.5 Update settings UI hints or labels so OpenAI GPT Image is clearly the recommended default without removing auto.

7. Phase 1 Verification

  • 7.1 Add routing tests proving the same gpt-image-* model can route to official GPT, Tuzi GPT, or generic fallback depending on profile.
  • 7.2 Add adapter-selection tests proving tuzi-gpt-image no longer lands on the generic default adapter.
  • 7.3 Add request serialization tests for official GPT generation/edit and Tuzi GPT generation.
  • 7.4 Add migration tests for auto, tuzi-compatible, and openai-compatible-basic.
  • 7.5 Run targeted Vitest coverage for provider routing, settings normalization, adapters, task persistence, and MCP image generation.
  • 7.6 Add regression tests for new-profile defaults, managed-profile defaults, managed-profile override persistence, and explicit custom-profile auto preservation.

8. Phase 2 Follow-Up

  • 8.1 Define the Tuzi GPT edit contract and dedicated request schema.
  • 8.2 Route Tuzi GPT edit requests through tuzi-gpt-image-adapter.
  • 8.3 Extend edit-path persistence, planner preference, and regression tests once the Tuzi edit contract is finalized.