compact_prompt: Instructions file for Claude Code

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

compactprompt is a Python library implementing the four prompt/data compression strategies from CompactPrompt: A Unified Pipeline for Prompt and Data Compression in LLM Workflows (Choi et al., 2025, arXiv:2510.18043). It is an independent, distilled reimplementation — not a port of the authors' reference notebook (which lives one directory up and is a 35-task, config-driven research harness; this package deliberately collapses that into a one-line API).

Commands

# Tests (full suite; optional-dep tests skip when their extras are absent)
pytest
pytest tests/test_ngram.py                  # one file
pytest tests/test_ngram.py::test_round_trip_is_lossless   # one test
pytest -rs                                  # show skip reasons

# Lint (CI gate; must stay at 10.00/10). CI runs over all tracked Python files:
pylint $(git ls-files '*.py')

# Install for development
pip install -e .                            # zero-dep core only
pip install -e '.[all,dev]'                 # everything + pytest, faithful to paper
python -m spacy download en_core_web_sm     # needed for phrase-level pruning

# Runnable tour of every strategy
python examples/quickstart.py

# CLI: review / compact markdown files and Claude Code skills (entry point in cli.py)
compactprompt review ./skills
compactprompt compact FILE.md --engine builtin            # dry run
compactprompt compact FILE.md --engine caveman --apply    # writes + .bak

# Interactive demo (Prompt + Files & Skills tabs)
streamlit run compactprompt_app.py

# Build the docs site (MkDocs); CI/Read the Docs build must stay clean
pip install -e '.[docs]' && mkdocs build --strict

CI runs two workflows (.github/workflows/): pylint.yml and tests.yml. The tests.yml core job installs the package with no extras across Python 3.9–3.13; the full job installs .[all,dev] + the spaCy model.