Skill

extract-images

Pulls real paper figures from arXiv source tarballs and embeds them in the source's Wiki/sources/<id>.md file. Use when the user invokes /paperwiki:extract-images, asks to "show me the figures from <paper>", "pull the architecture diagram", "extract images from <id>", or follows up on a digest entry that has an empty Figures section.

Install

npx claudepluginhub kaneyxx/paper-wiki --plugin paper-wiki

Tool Access

This skill uses the workspace's default tool permissions.

Preview

Extract Images downloads the arXiv source bundle for a paper, pulls

SKILL.md

Similar Skills

angular-new-app

Creates new Angular apps using Angular CLI with flags for routing, SSR, SCSS, prefixes, and AI config. Follows best practices for modern TypeScript/Angular development. Use when starting Angular projects.

angular-angular-1

100.1k

angular-developer

20 files

Generates Angular code and provides architectural guidance for projects, components, services, reactivity with signals, forms, dependency injection, routing, SSR, ARIA accessibility, animations, Tailwind styling, testing, and CLI tooling.

angular-angular-1

100.1k

context7-cli

3 files

Executes ctx7 CLI to fetch up-to-date library documentation, manage AI coding skills (install/search/generate/remove/suggest), and configure Context7 MCP. Useful for current API refs, skill handling, or agent setup.

upstash-context7-1

53.8k

Stats

Stars0

Forks0

Last CommitApr 26, 2026

Actions

View Source View Plugin View on GitHub View README

paper-wiki Extract Images

Overview

Extract Images downloads the arXiv source bundle for a paper, pulls the real paper figures (architecture diagrams, experimental plots, qualitative results) out of common figure directories (figures/, fig/, pics/, images/, img/), and embeds them into the ## Figures section of Wiki/sources/<id>.md using Obsidian wikilink-with-width syntax.

This SKILL is the visual half of the per-paper note story. The text half (frontmatter + Core Information + Abstract + Key Takeaways + Notes) is filled by /paperwiki:digest, /paperwiki:analyze, and /paperwiki:wiki-ingest.

When to Use

The user types /paperwiki:extract-images <canonical-id>.
The user says "show me the figures for ", "pull the architecture diagram from ", "what does the method look like".
The user is reviewing a fresh digest and wants the figures from a ranked paper before deciding whether to deep-dive.
The ## Figures section in a Wiki/sources/<id>.md file is empty (still has the placeholder text) — that's the explicit prompt to run this SKILL.

Do not use for non-arXiv canonical ids (the source-tarball URL only exists for arxiv: papers — paperclip/PMC papers don't have one). Surface that to the user instead of running the runner.

Process

Validate canonical id. It must start with arxiv:. If the user handed an arXiv URL or a fuzzy title, normalize first via the paperwiki._internal.normalize helpers (or just ask).
Run the runner. Invoke ${CLAUDE_PLUGIN_ROOT}/.venv/bin/python -m paperwiki.runners.extract_paper_images <vault> <canonical-id>. Read the JSON output.
Surface the result. Confirm image_count to the user. If image_count == 0, tell them the paper is probably PDF-only on arXiv (legitimate, no failure) and offer next steps: /paperwiki:analyze for a Claude-written deep dive based on the abstract, or download the PDF themselves.
Optionally preview. If the user asks "show me the teaser" or similar, render one or two of the embedded figures in a chat message via Obsidian wikilink — paper-wiki has no PDF→preview capability so this is just the wikilink, not actual image rendering.
Hand off to wiki-ingest. Suggest the user run /paperwiki:wiki-ingest <id> next so the freshly imaged source folds into concept articles with the figure references intact.

Common Rationalizations

Excuse	Why it's wrong
"The user wants images, I'll grab the PDF and convert pages."	PDF page renders include logos, equations, and full-text noise. The arXiv source tarball has the AUTHOR'S OWN figure files — that's what users want. Always go through the runner.
"Tarball is missing — I'll just say 'failed' and move on."	A 0-figures result is normal for older PDF-only papers. Tell the user explicitly so they don't think it's a bug.
"I'll add my own commentary into the `## Figures` section."	The runner overwrites that section verbatim each call. Put commentary in `## Notes`, which the runner never touches.
"Force a re-download every time to make sure it's fresh."	The cache exists for a reason — arXiv rate-limits source downloads more aggressively than the API. Only use `--force` when the user explicitly asks.

Red Flags

The user gave a paperclip: or s2: canonical id — there is no arXiv source URL. Surface this and offer alternatives instead of running the runner.
Wiki/sources/<id>.md doesn't exist — the runner errors with UserError. Tell the user to run /paperwiki:digest or /paperwiki:analyze first to create the source stub.
image_count is suspiciously high (> 30) — usually means the source bundle has glyph rasters or figure subsets we accidentally caught. Eyeball the list before suggesting wiki-ingest.
arXiv returns 503 — they throttle source-tarball downloads more than the API. Wait an hour, retry. Do NOT keep hammering.

Verification

Wiki/sources/<id>/images/ exists and contains at least the count the runner reported.
The source .md file's ## Figures section now has ![[<id>/images/<filename>|800]] embeds (or the "no figures found" note when image_count == 0).
The source .md file's ## Notes section is unchanged from before the run.
Wiki/.cache/sources/<arxiv-id>.tar.gz exists for future re-extracts without HTTP fetches.