name: pdf_tools description: Read PDF documents page by page and build new PDF files from text or structured answers with Unicode-capable fonts. version: 0.1.0 type: script runtime: python3 permissions: [fs] when_to_use: The user attached a PDF and asks to read, summarize, search, or compare it, or asks to produce or export a PDF. timeout_sec: 120 scripts:

• name: read_pdf.py

description: Extract the readable text layer per page and report honestly when a document has no text layer.

• name: make_pdf.py

description: Build a PDF from plain or lightly structured text with Unicode-capable fonts. install_specs:

• kind: pip

package: pypdf

• kind: pip

package: reportlab

PDF tools

This skill can read PDFs and build new PDFs.

Read a document

skill_exec(skill="pdf_tools", script="scripts/read_pdf.py", args=["<file path>"])

Useful arguments include --pages 1-3,7 to select pages, --max-chars N to bound extracted text, and --meta-only to return only the filename and page count.

The script returns JSON containing the file name, page count, page-level text, total extracted characters, and truncation status. Briefly confirm what was read before answering from the document contents.

If a PDF has no text layer, the script returns status: "no_text_layer". Do not guess at missing contents. Explain that the document is probably a scan and offer image inspection (view_image or vlm_query) or a dedicated OCR skill if one is installed.

Build a document

skill_exec(skill="pdf_tools", script="scripts/make_pdf.py", args=["--in", "<text file>", "--out", "report.pdf", "--title", "Report title"])

Use --text instead of --in to pass text directly. Supported light markup includes #/##/### headings, blank-line paragraphs, - bullets, 1. numbered items, --- rules, **bold**, and *italic*. Unicode TrueType fonts support both Cyrillic and Latin text.

Ask a clarifying question only when the requested structure or scope is ambiguous; otherwise create the file directly.

Output and delivery convention

Generated files are always written inside the skill state directory under outputs/. Pass a bare file name as --out (e.g. --out report.pdf); a path outside the state directory is refused with status: "output_outside_state_dir". The returned JSON carries the absolute path of the created file in file. After the file is created, deliver it to the user yourself with send_file (host tooling) — the skill only creates the file. NEVER construct download URLs.

Boundaries

• The skill reads and creates PDFs. It does not edit existing PDF forms,

annotations, complex layouts, or embedded content in place.

• Reading is allowed from Ouroboros working directories: uploads, task results

and drives, Deliverables, the skill state directory, and the system temporary directory. Writing is confined to the skill state directory (outputs/).

• If no Unicode font is available, the script reports that explicitly instead

of producing missing-glyph boxes. Use --font to provide a font explicitly.