ai-wiki-image-ingest - SKILL.md Agent Skill

name: ai-wiki-image-ingest description: Ingest raw image files into this AI Obsidian wiki. Use when processing PNG, JPG, JPEG, WEBP, GIF, HEIC, or other supported image sources under the AI raw source root into source notes, index coverage, log entries, and optional compiled concept, entity, synthesis, or map pages.

AI Wiki Image Ingest

Use this skill whenever ingesting a raw image file for this AI wiki. This skill owns image-specific ingest behavior; AGENTS.md owns global wiki rules, raw boundaries, source mapping, metadata, index/log, asset policy, and question-page policy.

Required Context

Before writing, read AGENTS.md, docs/wiki-templates.md, and index.md.

Fast Path

Pick or receive a supported raw image under the raw source root from AGENTS.md.
Open the image with Agent vision.
Inspect the full image for document/screenshot/photo/diagram type, reliable visible text, layout, visual relationships, knowledge points, and limitations.
Stat the raw file and collect the raw metadata required by AGENTS.md.
Create or update the mirrored source note, update index.md, append one log.md line, then compile only reusable concept/entity/synthesis/map updates.

Image Source Rules

Use the source-note template from docs/wiki-templates.md. Image-specific requirements:

Set source_type: image.
Set status: ingested when the image opens and gives enough signal for a source note.
Use partial when the image is blurry, cropped, low-resolution, mostly unreadable, animated with uninspected frames, or otherwise ambiguous.
Use blocked when the image cannot be opened or viewed.
Record image type and visual coverage in Source or Maintenance Notes; include dimensions when cheap to determine.
Source Digest must describe the visual structure, reliable visible text, relationships, knowledge points, and limits. Do not present uncertain OCR as fact.
In Key Claims, use explicit: for claims directly visible in the image and inferred: for agent interpretation of the visual content.
In External Links, include URLs that are clearly visible in the image; otherwise write No external links found in extracted content.
Mark uncertain visible text with uncertain.

Group Boundary

When multiple images in the same raw directory clearly belong to one set, create or update one source note per image first. Then compile a concept or synthesis page for the group only if the set contributes reusable knowledge beyond individual source notes.

Failure Path

If the image cannot be opened or viewed, still create or update the mirrored source note with status: blocked, raw metadata, the view/open error summary, and a concrete human next step. If the image opens but the content is too blurry, cropped, low-resolution, or otherwise unreliable, use partial and describe the limitation. Do not edit, convert, repair, or write next to the raw image.

Do not create questions/ pages unless the user explicitly asks to save a question or answer.