extractor

# Extractor This skill is a thin, zero-choice wrapper over the Extractor project. Give it one supported file and it delegates route selection, PDF handling, provider choice, artifact validation, and terminal status to Extractor. ## Quick Start ```bash .pi/skills/extractor/run.sh paper.pdf .pi/skills/extractor/run.sh paper.pdf --out ./results .pi/skills/extractor/run.sh paper.pdf --offline .pi/skills/extractor/run.sh paper.pdf --format markdown ``` The command writes the canonical `extractor.result.v1` result to stdout for JSON output. Human progress and errors belong on stderr. ## Normal Flags | Flag | Purpose | | --- | --- | | `input` | Required local file path. | | `--out PATH` | Output directory for run artifacts. | | `--offline` | Disable network/model enrichment. | | `--format json` | Emit canonical JSON, the default. | | `--format markdown` | Emit Markdown presentation when available. | | `--json` | Compatibility alias for `--format json`. | | `--markdown` | Compatibility alias for `--format markdown`. | Do not choose engines, presets, providers, models, OCR strategy, or pipeline stages from this skill. Extractor owns those decisions behind its application facade and returns `extractor.result.v1` for JSON extraction results. ## Maintainer Commands ```bash .pi/skills/extractor/run.sh doctor .pi/skills/extractor/run.sh debug-routing paper.pdf .pi/skills/extractor/sanity.sh ``` `doctor` checks the declared Extractor project and installed command. `debug-routing` prints the canonical Extractor help and supported wrapper arguments; it does not perform extraction. Set `EXTRACTOR_ROOT` only when the Extractor checkout is not at `${HOME}/workspace/experiments/extractor`. Set `EXTRACTOR_COMMAND` to an installed executable path when testing a clean wheel install, for example `EXTRACTOR_COMMAND=/tmp/extractor-clean/bin/extractor`.