Stats
Actions
Available In
Tags
Plugin
sisyphus
Two-stage label → extract pipeline for mining structured data from scientific papers. Run /sisyphus:build to be guided through labeling rules, extraction schemas, and a runnable pipeline. The /sisyphus:reference skill is auto-invoked when working on sisyphus code and supplies the API surface.
What's Inside
Skills2
build
/build
Guide the user through building a sisyphus label → extract pipeline for scientific papers. User-invoked via /sisyphus:build.
reference
/reference
API reference for the sisyphus label → extract pipeline. Use when reading or writing code that imports `sisyphus`, `sisyphus.chain`, or related modules — covers the 6 core classes, data contracts between stages, strategy selection (isolated vs merged), and the Python-dependency setup check.
README
sisyphus — MatSci Extraction Plugin for Claude Code
A two-stage label → extract pipeline for mining structured data from scientific papers, packaged as a Claude Code plugin.
The plugin gives you a /sisyphus:build slash command that walks you through labeling rules, an extraction schema, and a complete runnable pipeline — generated for your specific material domain.
Install the plugin
Option A — local development (this directory)
claude --plugin-dir /path/to/sisyphus-skill
Useful for testing changes; the plugin is loaded only for the current Claude Code session.
Option B — from a marketplace
If this plugin is published to a marketplace you've added:
/plugin install sisyphus
See Discover and install plugins for marketplace setup.
Install the Python package (required)
The plugin generates Python code that imports sisyphus, so the package must be importable in your active project environment — not in the plugin's own venv.
The /sisyphus:build command will detect your environment and suggest the right command. Or you can install it manually now:
| Your project uses | Command |
|---|---|
| uv (uv.lock present) | uv add "sisyphus @ git+https://github.com/sukiluvcode/skills.git" |
| poetry (poetry.lock) | poetry add "git+https://github.com/sukiluvcode/skills.git" |
| pipenv (Pipfile) | pipenv install "git+https://github.com/sukiluvcode/skills.git#egg=sisyphus" |
| conda (env activated) | pip install "git+https://github.com/sukiluvcode/skills.git" |
| plain pip / venv | pip install "git+https://github.com/sukiluvcode/skills.git" |
| no env yet (recommended) | uv init && uv add "sisyphus @ git+https://github.com/sukiluvcode/skills.git" |
The package is heavy (langchain, chromadb, faiss, dspy …) — expect a few minutes for the first install.
Common install issues
| Symptom | Fix |
|---|---|
error: externally-managed-environment (macOS / Linux PEP 668) | Create a venv: python3 -m venv .venv && source .venv/bin/activate, then retry |
faiss-cpu wheel build fails on Windows | Pin Python to 3.10–3.12; or pip install faiss-cpu --only-binary=:all: |
chromadb install hangs | pip install -U pip and retry; chromadb >= 0.4.24 ships wheels for common platforms |
| Import works in shell but not in notebook | Notebook is using a different kernel — install ipykernel in the same env and register it |
Set your API key
The pipeline calls OpenAI-compatible LLMs. In your project root:
echo "OPENAI_API_KEY=sk-..." >> .env
Use it
In your project directory, launch Claude Code and run:
/sisyphus:build
Claude Code will:
- Verify
import sisyphusworks in your active environment - Ask what material domain and properties you want to extract
- Suggest regex / semantic-search filters per property
- Generate a complete labeling script
- Ask for the extraction schema
- Generate the complete extraction script
You run the generated scripts yourself — the plugin does code generation, not execution.
What's inside
| Path | Purpose |
|---|---|
.claude-plugin/plugin.json | Plugin manifest |
skills/build/SKILL.md | The /sisyphus:build workflow (user-invoked) |
skills/reference/SKILL.md | The /sisyphus:reference API skill (auto-invoked); contains the dependency setup check |
references/GUIDE.md | Template decision tree |
references/single_prop.py | One-property template |
references/multi_props_isolated.py | Multiple independent properties |
references/multi_props.py | Coupled properties + synthesis context (HEAs) |
references/processing_template.py | Synthesis-process templates |
sisyphus/ | The Python package (chain, index, patch, utils) |
pyproject.toml | Build config — hatchling, dependencies |
Pipeline shape
Stage 1 — Label Filter(source_db) + Labeling(labeler, ...) + Saver('labeled_db')
Stage 2 — Extract Filter(labeled_db) + load + Extraction(extractor, ...) + Writer(result_db)
See references/GUIDE.md for which template fits your use case.
License
MIT
Stats
Version0.1.0
LanguagePython
Stars0
MaintenanceExcellent
LicenseMIT
Last Commit
Added
Tags
