generate_presentation: image embedding in slides via ppt-rs::add_image

[oxoxDev]

Summary

generate_presentation (#2778, PR #3016) currently emits text-only slides — titles, bodies, bullets, speaker notes. Real decks usually include images: charts, screenshots, photo placeholders. ppt-rs already supports image embedding via Image::from_bytes(...).add_to(...) on the slide builder. Wire it through so the agent can dispatch a deck with images alongside text.

Problem

Users asking the agent to "make a deck with the Q3 revenue chart" or "include a screenshot of the dashboard" get a text-only deck back today. The tool spec on SlideSpec has no images field, the Rust engine doesn't accept image payloads, and there's no path from a chat-uploaded image or a Composio-fetched file to a slide-embedded asset.

Solution (optional)

1. Extend SlideSpec (src/openhuman/tools/impl/presentation/types.rs) with an optional images: Vec<SlideImage> field. SlideImage carries either an in-workspace artifact id, a [FILE:path] marker (per Extend multimodal input to accept document and file attachments beyond images #2777), or a remote URL with a size cap.
2. Resolve image references in the engine (engine.rs build_slides): walk each SlideImage, fetch bytes (with a deadline-bounded HTTP client for URLs, tokio::fs::read for FILE markers, artifacts::store::get_artifact_bytes for artifact ids), decode + validate (PNG/JPEG/WebP only, max 5 MB), and hand off to ppt-rs's Image::from_bytes.
3. Layout heuristics: when a slide has both bullets and images, lay them out side-by-side; when only images, full-bleed. Start with a fixed two-column grid for ≤2 images, 2x2 grid for 3-4. Defer arbitrary positioning.
4. Tests: round-trip a PNG, assert image entry present in the [Content_Types].xml and ppt/media/image1.png entry in the resulting zip. Size-cap rejection. MIME rejection.
5. Orchestrator prompt update (agent_registry/agents/orchestrator/prompt.md): document that generate_presentation accepts an images array, and clarify the grounding rule covers image fetches (e.g. agent must research/memory_tree before claiming a chart shows X).

Acceptance criteria

• SlideSpec.images accepted — passing SlideImage entries on a slide attaches the image at generation time.
• Three resolution paths — artifact-id, [FILE:path] marker (depends on Extend multimodal input to accept document and file attachments beyond images #2777), remote URL (deadline-bounded, ≤5 MB, allowed MIME).
• Layout heuristic — single image full-bleed; multiple images grid-arranged.
• Tests — round-trip a PNG with the deck, assert media entry in the zip; reject oversized; reject wrong MIME.
• Diff coverage ≥ 80% — per .github/workflows/pr-ci.yml.

Related

• Tool added in: feat(tools): add generate_presentation tool (native rust engine, ppt-rs) (#2778) #3016 (PR), parent Add PPT generation tool using python-pptx in the Python runtime #2778
• File-marker dependency: Extend multimodal input to accept document and file attachments beyond images #2777 (PR feat(agent/multimodal): support [FILE:…] markers with text extraction (#2777) #2954, MERGED) for [FILE:path] resolution
• Epic: Add multimodal support for files and PPT generation #1535