Merged origin/main into the PR branch to pick up the Z.AI provider work and
resolved the shared baoyu-imagine docs/code touches. While reconciling the
branch, fixed the outstanding review findings by preserving explicit
--imageSize validation for Replicate while ignoring inherited config defaults,
and by classifying Replicate multi-output extraction failures as
non-retryable so paid requests are not repeated.
Constraint: Merge had to preserve incoming Z.AI provider support from origin/main
Constraint: Replicate should still reject explicit --imageSize usage while tolerating inherited default_image_size config
Rejected: Silently ignore every Replicate imageSize value | would hide explicit CLI/task misuse
Rejected: Leave multi-output extraction under generic retry logic | repeats deterministic paid failures
Confidence: high
Scope-risk: moderate
Reversibility: clean
Directive: Keep imageSize source tracking aligned with CLI/config/batch task paths before tightening provider validation
Tested: node --import tsx --test skills/baoyu-imagine/scripts/main.test.ts skills/baoyu-imagine/scripts/providers/replicate.test.ts skills/baoyu-imagine/scripts/providers/zai.test.ts
Tested: npm test
Not-tested: Live GitHub thread resolution/reply actions
Adds the Z.AI (智谱) provider supporting glm-image and cogview-4-250304
models via the Z.AI sync image API. Configure with ZAI_API_KEY (or
BIGMODEL_API_KEY for backward compat). Reference images are not supported yet.
Jim Liu 宝玉