image-metadata-from-cache

Image Metadata from Cache

Retrieves AI-generated title, keywords, and description for images in a note by reading the ai-image-analyzer plugin cache. Run the script with vault root and note path; it returns a JSON array of cached image metadata.

Instructions

1. Run the script with vault root and note path (relative or absolute). The script extracts image refs from the note, resolves paths (bare filename → Assets/<filename>; path in link → normalized vault-relative), looks up the plugin cache per image, and outputs JSON.
2. From the vault: If skills are symlinked under .cursor/skills/, invoke the script via its path (e.g. .cursor/skills/image-metadata-from-cache/scripts/get_image_metadata.py). No extra dependencies; stdlib only.

Script usage

uv run python .cursor/skills/image-metadata-from-cache/scripts/get_image_metadata.py <vault_path> <note_path>

Example (from vault root):

uv run python .cursor/skills/image-metadata-from-cache/scripts/get_image_metadata.py . "00-Inbox/2026-03-07.md"

Arguments: vault_path — Obsidian vault root (directory containing .obsidian/). note_path — Note file path relative to vault or absolute.

Output: JSON array to stdout, one object per cached image (deduplicated by path). Empty array if no images or no cache entries.

[
  { "path": "Assets/Pasted image 20260307180120.png", "text": "Title: ...\nKeywords: ...\nDescription: ..." },
  { "path": "Assets/other.png", "text": "..." }
]

Path resolution

• Bare filename (e.g. ![[Pasted image 20260307180120.png]]) → Assets/<filename>.
• Path in link (e.g. ![[Assets/foo.png]] or ![](Assets/foo.png)) → normalized vault-relative (no leading /, forward slashes). Markdown paths are URL-decoded.

All image assets live under Assets/; filenames are unique.

Cache

• Dir: vault/.obsidian/plugins/ai-image-analyzer/cache/. Key: MD5 of vault-relative path (no leading slash). Each file: JSON with path, text, libVersion; script returns path and text only.

Integration

• organize-daily-notes: When a note has embedded images, run this script and use the returned JSON as context for fixes. Do not modify the embeds.
• assimilate-knowledge / summarize-note: Optionally run this script to include image descriptions in context.