figure-extractorlisted

# Figure Extractor Skill Use this skill when the user needs complete figures or images extracted from PDFs, arXiv papers, or HTML article pages. ## What this skill actually is (read first) `figure-extractor` is a **command-line tool** (Python + PyMuPDF). "Using it" means running its CLI in a shell — e.g. `figure-extractor extract paper.pdf`. It is **not** an in-agent callable, a hosted service, or a "page" you activate, and it cannot run inside a runtime that has no shell or cannot install Python packages (e.g. Notion Agent, browser-only agents). **Invocation contract** - **Mechanism:** shell command `figure-extractor <subcommand> ...` (installed via `pip install -e .`). - **Input:** a local PDF/HTML path or a PDF/arXiv/HTML URL. - **Output:** PNG files + `manifest.json` + `contact_sheet.jpg` (+ optional `figures.zip`) written to `--out`. ## Step 0 — precondition check (always do this first) Before you promise figure extraction, verify the tool is actually runnable in this environment: ```bash figure-extractor --help # if this errors, or there is no shell, the tool is UNAVAILABLE ``` - **Available** (command runs) → follow *Preferred algorithm*. - **Unavailable** (no shell / not installed / sandboxed) → go to *Degraded mode*. Not having the tool is **not** a failure and **not** a rule violation — degrade gracefully and tell the user what you did. ## Core rule (when the extractor is available) Do **not** rely on embedded PDF image extraction alone. Many figur