The shared docmetry-eval skill teaches Codex and Claude Code how to use the
local docmetry workflow. It generates the runtime question file, runs
extraction and deterministic evaluation, and writes tabular summary.md plus
machine-readable summary.json results. It does not
host extraction or evaluation and does not make API calls unless a semantic
answer adapter is explicitly selected.
Install the one shared skill into Codex:
mkdir -p ~/.codex/skills
cp -R integrations/shared/docmetry-eval ~/.codex/skills/The skill runs when Codex is working in a docmetry checkout. It is not a standalone PDF engine.
Install the same shared skill into Claude Code:
mkdir -p .claude/skills
cp -R integrations/shared/docmetry-eval .claude/skills/The same folder can be copied into a user-scoped Claude Code skills directory when supported by the installed Claude Code version.
Install extraction providers locally from the docmetry extras. Install only
the semantic adapter needed:
pip install -e ".[all-providers]"
pip install -e ".[gemini]" # optional
pip install -e ".[openai]" # optional
pip install -e ".[anthropic]" # optional