Obsidian Attachment Janitor

Clean up and organize attachments in the Obsidian vault. Two jobs:

1. Find and delete unreferenced attachments
2. Rename referenced attachments with consistent, searchable names

Naming convention

YYYY-MM-DD description.ext

• Date prefix: from the referencing note's filename. Extract the YYYY-MM-DD portion only (ignore time like 1230 in 2026-03-09 1230 ...)
• Description: short English kebab-case phrase describing what the attachment is, based on the embedding context in the note. Keep it under 5 words.
• Extension: preserved from original

When the same note embeds multiple attachments, differentiate by what each one depicts, not by index number. Read the surrounding text to understand each attachment's role.

Example — a blog post note 2026-03-09 1230 让 AI 接管笔记维护.md embeds 4 images:

Original	New
Gemini_Generated_Image_5tv8tg5tv8tg5tv8.png	2026-03-09 ai-note-maintenance-cover.png
Gemini_Generated_Image_fdvb4qfdvb4qfdvb.png	2026-03-09 agent-system-overview.png
CleanShot 2026-03-09 at 20.40.27@2x.png	2026-03-09 pr-diff-screenshot.png
CleanShot 2026-03-10 at 23.45.40@2x.png	2026-03-10 sentinel-report-screenshot.png

Notice the last one uses 2026-03-10 — the date comes from the attachment's own CleanShot timestamp when it differs from the note date and is clearly identifiable. Prefer the note's date when in doubt.

Workflow

Step 1: Build reference map

1. List all files in Attachments/
2. Search all .md files for ![[filename]] embeds
   • Exclude .agents/skills/ — documentation examples, not real references
   • Strip parameters: ![[file.png|300]] → file.png, ![[file.png|alt text]] → file.png
3. Result: a map of attachment filename → [referencing note paths]

Step 2: Classify

• Unreferenced: zero references → candidate for deletion
• Referenced, datable: referencing note has YYYY-MM-DD in filename → candidate for rename
• Referenced, undatable: referencing note has no date prefix (e.g., Private/Career/Portfolio/ notes) → skip rename, keep as-is

Step 3: Generate rename plan

For each datable attachment:

1. Read the referencing note to understand the embedding context
2. Extract YYYY-MM-DD from the note filename
3. Generate a descriptive English kebab-case name
4. Check for naming collisions — if two files would get the same name, make descriptions more specific rather than adding numbers

Skip if the attachment already matches the convention.

Step 4: Present plan

Show the user three lists:

1. Delete — unreferenced attachments with file sizes
2. Rename — current name → new name with the referencing note for context
3. Skip — attachments kept as-is with reason (undatable, already named well, etc.)

Wait for explicit user confirmation before executing anything.

Step 5: Execute

Use the Obsidian CLI for both operations — it handles reference updates automatically.

For deletions:

obsidian delete path="Attachments/filename.png" permanent

For renames:

obsidian rename path="Attachments/old-name.png" name="new-name.png"

Always use path= instead of file= — the file= parameter resolves like a wikilink and fails on filenames with special characters (spaces, @, underscores, hashes).

Step 6: Verify

Run a quick scan to confirm no broken embed references were introduced. Report summary: X deleted, Y renamed, Z skipped.