Task: Merge AI Design image-to-image into current client main

Identity

Integrate the already reviewed AI Design single-reference image-to-image client change (809364e) onto the current client main baseline a03043d.
Preserve the already integrated local release-build and first-conversation latency work while resolving any overlapping README, page, test, or canonical documentation edits semantically.
Re-run the focused page suite, typecheck, scoped lint and Renderer build on the resulting integration head, then obtain an independent read-only review.
Land the verified result on local main only if its existing updater task can be left intact; never overwrite, stash, reset, or absorb that task's uncommitted files.

The picker continues to submit one real Workspace Asset ID through the existing attachmentAssetIds transport; no parallel upload or local-file description path is introduced.
Only brief.medium === 'image' selects image-reference semantics. video, null and missing medium retain the historical video-first-frame behavior.
Preserve Workspace / Conversation / Task ownership from ADR-001 and keep image/video tasks Workspace-scoped.
Preserve the deployed exact quick reply 从作品列表选择图片 for this compatible increment; a structured action/purpose remains future protocol work.
Keep all Git and file mutations rooted in this isolated integration worktree until the original main worktree is clean and ownership-safe.

Merged reviewed feature commit 809364e into the current a03043d client baseline with merge commit 88281b8; Git resolved the overlapping ImageCanvas and page-test history without source conflicts and preserved the existing first-conversation work.
Promoted the generalized single-image picker, one-Asset attachment boundary, image/video/null-medium behavior, exact-string compatibility risk and production-smoke obligation into the current canonical project memory.
README now describes text-to-image, single-reference image-to-image, project work selection and local upload on top of the latest client product state.
Independent review found that legacy snapshots with an omitted Brief medium were rejected by the cloud adapter before the page fallback. Commit 7a807a2 now accepts only an omitted/null/image/video medium at the boundary, normalizes omission to null, and continues to reject other non-null values.

Source merge preflight used git merge-tree against current main; only the README paragraph needed an overlapping semantic merge and the final merge completed automatically.
pnpm install --offline --frozen-lockfile: passed with the pinned pnpm; dependencies were restored entirely from the local store.
pnpm exec vitest run tests/unit/image-canvas-page.test.tsx: 25 passed.
pnpm exec vitest run tests/unit/works-square-design-workspace.test.ts: 22 passed, including REST and WebSocket missing-medium normalization and invalid-medium rejection.
pnpm run typecheck: passed.
pnpm exec eslint src/pages/ImageCanvas/index.tsx tests/unit/image-canvas-page.test.tsx: passed.
pnpm run build:vite: passed for Renderer, Main and Preload; emitted only the repository's existing dynamic-import and chunk-size warnings.
git diff --check: passed.
check_project_docs.py: passed.
check_doc_drift.py --task-id 20260813-integrate-i2i-client-main-5d9e: passed after keeping the source task record read-only on its source branch, matching the repository's established Integration Gate pattern.
Packaged desktop and live Works Square/Bailian provider smoke were not run.
Initial independent Sol review: FAIL with one P2 missing-medium adapter gap and one P3 stale verification record; both were corrected. Final independent Sol review is rerun against the corrected head before promotion.

Replace exact localized quick-reply matching with a structured asset-picker action/purpose before adding more input modes or changing that label.
Run packaged-desktop and live Works Square/Bailian image-to-image smoke after the matching client and server releases are deployed.