theramjad/image-cleanup

What This Skill Does

After excalidraw-gen creates 10 images per section and the user picks their favorite(s), this skill:

1. Deletes unreferenced images — scans .md files for ![[...]] embeds, cross-references against actual image files, removes anything not referenced
2. Renames images descriptively — Claude reads each remaining image with vision and gives it a descriptive filename based on what the image actually shows
3. Moves to vault root — all images go to images/ at the vault root (/Users/ray/Desktop/ray-os/images/), organized as images/{note-slug}/descriptive-name.png
4. Updates markdown references — rewrites all ![[...]] embeds to match the new paths
5. Cleans up — removes empty directories and old local images/ folders

Before → After

Before (in socials/youtube/videos/):
images/two-room-office/excalidraw_1.png    ← unreferenced
images/two-room-office/excalidraw_2.png    ← unreferenced
images/two-room-office/excalidraw_6.png    ← referenced
images/thin-waist/excalidraw_9.png         ← referenced

After (at vault root):
images/my-video-name/two-room-office-analogy.png
images/my-video-name/thin-waist-architecture.png

Markdown updates:

![[images/two-room-office/excalidraw_6.png]]  →  ![[images/my-video-name/two-room-office-analogy.png]]

---

Workflow

1. User provides a target folder (or you infer it from context). If none provided, ask.
2. Run the dry run to find unreferenced images:

   python3 /Users/ray/Desktop/ray-os/.claude/skills/image-cleanup/scripts/cleanup.py "/path/to/target/folder" --no-flatten
   

3. Show the user the dry run output so they can review.
4. After confirmation, delete unreferenced images:

   python3 /Users/ray/Desktop/ray-os/.claude/skills/image-cleanup/scripts/cleanup.py "/path/to/target/folder" --no-flatten --execute
   

5. Read each remaining image using the Read tool (vision) and propose descriptive filenames based on what the image actually depicts. Present the rename plan to the user.
6. After confirmation, rename + move images to vault root:
   • Create subfolder images/{note-slug}/ at vault root if it doesn't exist
   • Move and rename each image
   • Update all ![[...]] references in the markdown files
   • Remove the old local images/ directory
   • Remove any empty directories left behind

---

Image Destination Rules

All images live at the vault root: /Users/ray/Desktop/ray-os/images/

• Multi-image notes → images/{note-slug}/descriptive-name.png
  • e.g., images/anthropic-skills-playbook/skill-shifts-distribution.png
• Single-image notes → images/{note-slug}/descriptive-name.png (same pattern)
  • e.g., images/using-cc-in-wsl/claude-code-in-wsl.png
• Note slug = filename → lowercase, non-alphanumeric runs become hyphens
  • Anthropic Skills Playbook → anthropic-skills-playbook
  • 1M Context Window → 1m-context-window

Never create local images/ folders inside project subdirectories. Everything goes to vault root.

---

Naming Images

Use vision, not folder names. Read each image with the Read tool and name it based on what it actually shows:

• Use the image's title/heading if visible (e.g., "How a Skill Shifts the Distribution" → skill-shifts-distribution.png)
• Use the core concept if no title (e.g., image showing a before/after comparison of prescriptive vs flexible prompts → prescriptive-vs-flexible-prompt.png)
• Keep names concise but descriptive (3-5 words slugified)
• Use hyphens, all lowercase

Do NOT use the excalidraw number, the subfolder name alone, or generic names.

---

Important Notes

• Always dry run first. Never delete images without showing the user what will happen.
• Always show rename plan. Present the proposed names before executing.
• Vault root is sacred. images/ at /Users/ray/Desktop/ray-os/images/ is the single source of truth for all images.
• Markdown references use vault-root paths. All refs should look like `` — Obsidian resolves these from the vault root.
• No dependencies. The cleanup.py script is pure Python 3 stdlib. Renaming and moving is done by Claude directly.

---

Helper Scripts

cleanup.py — Delete unreferenced images

Always use with --no-flatten. Only use this for deleting unreferenced images.

# Dry run
python3 /Users/ray/Desktop/ray-os/.claude/skills/image-cleanup/scripts/cleanup.py "/path/to/folder" --no-flatten

# Execute
python3 /Users/ray/Desktop/ray-os/.claude/skills/image-cleanup/scripts/cleanup.py "/path/to/folder" --no-flatten --execute

rename-by-context.py — Fallback rename tool

Only use if you cannot read images with vision. Renames excalidraw files using subfolder names or surrounding markdown context.

python3 /Users/ray/Desktop/ray-os/.claude/skills/image-cleanup/scripts/rename-by-context.py "/path/to/folder"