Rules for agents (read first)

• Best for tables, invoices, columnar/financial PDFs — layout and cell values survive.
• Parse once, then default to bounded grep: grep -n -i -C2 "term" file | head.
• Use the query skill only when grep would flood — a common term over a corpus too large to scan; then small -k, and --language <lang> for non-English.
• Don't read the source PDF as an image to get its text — this extractor is faster and more accurate for extractable text. (For a scanned/image-only PDF with no text layer, a vision tool is the right fallback — see Troubleshooting.)

PDF to Text

Convert PDFs into layout-preserving plain text. Each word is placed on a character grid that mirrors its on-page position, so columns, indentation, and tabular alignment survive the conversion. This is significantly higher quality than reading a PDF directly with the read tool, which only extracts loose text without spatial fidelity.

When to use this vs. pdf-to-markdown

• Use pdf-to-text when the downstream consumer is plain-text only (a non-Markdown LLM, a grep/awk pipeline, a CSV-style table extractor that cares about column alignment).
• Use pdf-to-markdown when the consumer benefits from semantic structure (headings, lists, tables, reading order). Most RAG and LLM-context pipelines fall here.

Related Nutrient skills

• query — once a file is extracted, search it instead of reading a large output back into context: ranked BM-25 search that returns only the top line windows ("parse once, query many"). Add it the way your agent installs skills.