pdf-extract

pdf-extract is a Sentosa skill for building staged, resumable extraction evidence packs from born-digital PDFs. It includes deterministic extraction scripts, profile/reference documentation, rendered-page OCR support, and vision-task handoff files for host agents.

Install

From a local clone:

./install.sh --codex
./install.sh --all

From a hosted repo once this is published:

curl -fsSL https://raw.githubusercontent.com/<org>/pdf-extract/main/install.sh \
  | bash -s -- --repo-url https://github.com/<org>/pdf-extract.git --all

Until the final remote URL is chosen, --repo-url is required for pipe-based installs. The installer links the skill by default, so updates to the cloned repo are immediately visible to Codex or Claude.

Usage

uv run scripts/pdf_extract.py run --profile <corpus>/extract-profile.toml
uv run scripts/pdf_extract.py status --profile <corpus>/extract-profile.toml
uv run scripts/pdf_extract.py redo <stage> --profile <corpus>/extract-profile.toml

Requirements and output contracts are documented in SKILL.md and references/.