Implement the hosted image provider behind the image capability contract, targeting the practical baseline of the OpenAI gpt-image-2 API.
Cover text-to-image and image editing where supported by the API, prompt and image inputs, size/quality/background/output-format/compression-style options exposed by the baseline, response decoding, and provider error mapping. Do not claim unsupported parity: document exact API version/endpoint assumptions and any deliberate differences.
Acceptance criteria
Uses the existing HTTP/auth/configuration patterns rather than a one-off client.
Request serialization and response decoding are covered with httptest fixtures.
Provider failures, malformed responses, timeouts, oversized inputs, and cancellation are handled without panics or leaked secrets.
No live API key is needed for tests.
A small runnable example shows the request shape and configuration.
Design
See docs/prds/image-capability-contract.md#hosted-provider-498 for the settled design: generation via ai-sdk/image, edit via a direct HTTP client (ai-sdk has no edit/mask support), plus the full provider-error-mapping table. Do not re-derive this from the acceptance criteria alone.
Suggested /council lenses
lens-contract, lens-operator -- run /council --lenses contract,operator once gate-clean, before opening the PR.
Notes
Leverage github.com/samcharles93/ai-sdk's existing image.Provider/core.GenerateImage/registry.RegisterImage for the generate path (azure/togetherai/xai already implement it; OpenAI does not yet) instead of writing a raw HTTP client -- same adapter pattern curatorLLMRunner uses for chat. Gap: ai-sdk's GenerateImageRequest has no input-image/mask fields, so editing needs either an upstream ai-sdk change or a separate direct client for the edit endpoint. Decide which at this bead's own design step. See docs/prds/image-capability-contract.md's closing note.
Close reason
Closed
Imported from beads archie-core-1786748942169-3-ae7cfaea · GitHub #498 · type task · priority P2 · created 2026-08-14 · closed 2026-10-03
## Scope
Implement the hosted image provider behind the image capability contract, targeting the practical baseline of the OpenAI `gpt-image-2` API.
Cover text-to-image and image editing where supported by the API, prompt and image inputs, size/quality/background/output-format/compression-style options exposed by the baseline, response decoding, and provider error mapping. Do not claim unsupported parity: document exact API version/endpoint assumptions and any deliberate differences.
## Acceptance criteria
- Uses the existing HTTP/auth/configuration patterns rather than a one-off client.
- Request serialization and response decoding are covered with httptest fixtures.
- Provider failures, malformed responses, timeouts, oversized inputs, and cancellation are handled without panics or leaked secrets.
- No live API key is needed for tests.
- A small runnable example shows the request shape and configuration.
## Design
See docs/prds/image-capability-contract.md#hosted-provider-498 for the settled design: generation via ai-sdk/image, edit via a direct HTTP client (ai-sdk has no edit/mask support), plus the full provider-error-mapping table. Do not re-derive this from the acceptance criteria alone.
## Suggested /council lenses
lens-contract, lens-operator -- run `/council --lenses contract,operator` once gate-clean, before opening the PR.
### Notes
Leverage github.com/samcharles93/ai-sdk's existing image.Provider/core.GenerateImage/registry.RegisterImage for the generate path (azure/togetherai/xai already implement it; OpenAI does not yet) instead of writing a raw HTTP client -- same adapter pattern curatorLLMRunner uses for chat. Gap: ai-sdk's GenerateImageRequest has no input-image/mask fields, so editing needs either an upstream ai-sdk change or a separate direct client for the edit endpoint. Decide which at this bead's own design step. See docs/prds/image-capability-contract.md's closing note.
### Close reason
Closed
---
_Imported from beads `archie-core-1786748942169-3-ae7cfaea` · GitHub #498 · type task · priority P2 · created 2026-08-14 · closed 2026-10-03_
sam
added the mediumtask labels 2026-10-27 13:34:37 +00:00
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Scope
Implement the hosted image provider behind the image capability contract, targeting the practical baseline of the OpenAI
gpt-image-2API.Cover text-to-image and image editing where supported by the API, prompt and image inputs, size/quality/background/output-format/compression-style options exposed by the baseline, response decoding, and provider error mapping. Do not claim unsupported parity: document exact API version/endpoint assumptions and any deliberate differences.
Acceptance criteria
Design
See docs/prds/image-capability-contract.md#hosted-provider-498 for the settled design: generation via ai-sdk/image, edit via a direct HTTP client (ai-sdk has no edit/mask support), plus the full provider-error-mapping table. Do not re-derive this from the acceptance criteria alone.
Suggested /council lenses
lens-contract, lens-operator -- run
/council --lenses contract,operatoronce gate-clean, before opening the PR.Notes
Leverage github.com/samcharles93/ai-sdk's existing image.Provider/core.GenerateImage/registry.RegisterImage for the generate path (azure/togetherai/xai already implement it; OpenAI does not yet) instead of writing a raw HTTP client -- same adapter pattern curatorLLMRunner uses for chat. Gap: ai-sdk's GenerateImageRequest has no input-image/mask fields, so editing needs either an upstream ai-sdk change or a separate direct client for the edit endpoint. Decide which at this bead's own design step. See docs/prds/image-capability-contract.md's closing note.
Close reason
Closed
Imported from beads
archie-core-1786748942169-3-ae7cfaea· GitHub #498 · type task · priority P2 · created 2026-08-14 · closed 2026-10-03