claudecode-cli-vision

To run OCR on PDF pages using Claude Code CLI, rasterize the pages to images and pass them with a direct prompt. The Skill routes provider requests through the CLI vision-capable flow, leveraging existing normalization to attach images to the prompt.

Multimodal vision fails because the default Claude Code CLI path rejects image inputs. This Skill fixes the failure by redirecting provider requests through the correct vision-capable backend flow for image and PDF-page OCR.

Yes, you can use Claude Code CLI for image OCR with non-blocking execution. The Skill preserves deterministic non-blocking query execution and returns timeout-aware errors for workflows calling ClaudeQueryBg with a timeout budget.

The Claude Code CLI path supports multimodal provider routing through this Skill. It automatically routes requests to the vision-capable flow instead of rejecting multimodal inputs, satisfying the functional requirement for correct provider handling.

The best way to integrate vision OCR into SourceVault Stage 4C is routing the provider to the vision-capable CLI flow. This Skill plugs directly into PDFIndex and SourceVault Stage 4C routines that call ClaudeQueryBg with attached image media files.

Image normalization for Claude Code CLI vision prompts works by exporting Image inputs to temporary PNG files. The existing iNormalizePrompt behavior attaches these normalized media files to the Claude CLI prompt for processing.