agentsclimarketplace

Obsidian clipping repair

Skill vsxd/agent-skills/skills/obsidian-clipping-repair

Portable agent skills for Codex, Claude Code, and other AI coding agents.

Install
npx -y skills add vsxd/agent-skills --skill obsidian-clipping-repair

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Repair and enrich Obsidian clipping notes imported from browser extensions, mobile share sheets, or other Markdown clipping flows. Use when notes in a vault's `Clippings`-like folders contain only a source URL, thin preview text, partial metadata, broken filenames, or incomplete bodies and need their source content, metadata, and note structure normalized without damaging real user-authored notes.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

4.0 KB, as published. Nobody here has run it

Obsidian Clipping Repair

Repair broken or thin clipping notes inside an Obsidian vault.

Use This Skill When

  • Notes were created by a browser clipper, mobile share flow, or clipping extension.
  • The saved Markdown contains only a link, a title plus teaser, partial metadata, or a body that is obviously incomplete.
  • The user wants vault-safe cleanup: better filenames, normalized frontmatter, recovered article content, and optional media import.

Do Not Use

  • The note already contains substantial human writing, annotation, or curation.
  • The intended source cannot be matched confidently.
  • Repair would require fabricating missing article text.

Workflow

  1. Determine scope. If the user names one note, inspect that note directly. If the user points at a vault or folder, inventory likely candidates first. Use references/candidate-detection.md.

  2. Preserve note evidence before rewriting. Keep the current filename, title, source URL, existing frontmatter, and any short preview text. These are disambiguation hints, not noise.

  3. Recover the true source. Resolve redirects, archive pages, hub pages, mirrors, and social-post links to the page the note actually intended. Use references/source-recovery.md.

  4. Extract the article or document with the lightest reliable method. Prefer one clean extraction path over repeated scraping attempts. Use structured page readers when available; fall back to rendered-page navigation only when needed.

  5. Rewrite the note conservatively. Normalize frontmatter, replace only the broken placeholder body, preserve useful user-authored text, and keep media organized predictably. Use references/note-shaping.md.

  6. Rename and verify. Rename the file only after the final title is stable. Stop on collisions, mismatches, or low-confidence recoveries.

  7. Batch safely when working across a vault. Process candidates in small groups, keep a summary of repaired and skipped notes, and do not let one ambiguous clip block the entire batch. Use references/batch-mode.md.

Built-In Helper

For vault-wide work, use the bundled candidate scanner to rank likely broken clipping notes:

python3 skills/obsidian-clipping-repair/scripts/find_candidates.py /path/to/vault --format json --top 50

This scanner is a triage aid, not authority. Review high-scoring candidates before rewriting them.

Ground Rules

  • Prefer canonical URLs without redirect wrappers or tracking parameters.
  • Prefer deterministic note cleanup over clever but fragile scraping.
  • Never overwrite meaningful user writing just because a note looks short.
  • Never replace a note with unrelated page content.
  • If confidence is low, leave the original note intact and report why.

Reference Files

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.