Most AI writing helpers focus on drafting text; this project treats the paper-creation workflow as its product. Its core insight is that reliable academic work needs modular AI skills plus explicit integrity gates so humans stay responsible while AI handles search, formatting, verification, and reviewer-style critique.
What Sets It Apart
- Human-in-the-loop pipeline: orchestrates research → drafting → multi-perspective review → finalize, but requires user checkpoints at each critical stage so the human keeps authorship control. This reduces common failure modes like frame‑lock and overreliance on model memory.
- Integrity-first design: built-in reference verification, cross-model audit hooks, and an append-only Material Passport record aim to catch fabricated citations and metadata errors early, rather than relying on post-hoc checks.
- Skillized for Claude Code: packaged as Claude Code skills and (as of v3.7.0) a one-line plugin install path, with per-skill metadata (data_access_level, task_type) and optional cross-model verification flags for higher-assurance workflows.
- Practical trade-offs surfaced: the project optimizes for reproducible human+AI collaboration (style calibration, reviewer sprint contracts, PRISMA-like SR support) at the cost of requiring Claude Code and an Anthropic API key and carrying a CC BY‑NC license (non-commercial constraint).
Who it's for — and when to look elsewhere
Great fit if you are an academic or research team that wants to accelerate literature search, drafting, and structured peer-review with explicit integrity checks and audit trails, and you can run Claude Code with an Anthropic key. The repo has substantial community adoption (several thousand stars) and extensive docs for pipeline operators.
Look elsewhere if you need a vendor-agnostic LLM workflow (this is Claude Code–centered), require a permissive commercial license (this repo is CC BY‑NC), or want a lightweight single-command summarizer rather than a multi-stage research orchestrator.
Where it fits
Pairs well with experiment orchestration tools (the companion experiment-agent) and citation databases for automated verification; best used when reproducibility, traceable verification, and accountable AI-assisted authorship matter more than minimal setup friction.
