A Python-based suite of Claude Code skills for academic research, covering research to publication with human-in-the-loop safeguards. It includes a 10-stage pipeline, cross-model verification, and auditing features; latest release is v3.19.0 with multiple integrity and gating enhancements.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
A comprehensive suite of Claude Code skills for academic research, covering the full pipeline from research to publication. It provides components for Deep Research, Academic Paper, Academic Paper Reviewer, and Academic Pipeline, along with data provenance, citation handling, and integrity gates. The project emphasizes human-in-the-loop governance and auditability in scientific writing and research workflows.
How it works
- The suite comprises multiple coordinated skills: a 10-stage pipeline orchestrator with claim verification, Material Passport, optional repro_lock, cross-model integrity verification, and score trajectory tracking.
- It includes modules for research, writing, peer review, and revision, with features such as Style Calibration, Writing Quality Check, LaTeX hardening, and VLM figure verification.
- It introduces mechanisms like data_access_level metadata, task_type annotation, and a Benchmark Report Schema for transparent benchmarking.
- Canonical cross-model handoff envelopes and a three-layer integrity framework are used to manage machine-human collaboration and auditability.
Getting started
- Install instructions for Claude Code and the ARS plugin are provided:
/plugin marketplace add Imbad0202/academic-research-skills
/plugin install academic-research-skills
- Verify functionality by invoking
/ars-planto map a paper structure via Socratic dialogue, or use/ars-lit-review "your topic"for a quick test. - Prerequisites include Claude Code, an exported ANTHROPIC_API_KEY, optional Pandoc for DOCX, tectonic + Source Han Serif TC for APA 7.0 PDF, and a real Python interpreter for certain features.
Getting started (continued)
- The README mentions six installation methods and a setup guide at
docs/SETUP.mdfor configuring API keys and cross-model verification viaARS_CROSS_MODEL. - If using Claude Science, import the GitHub URL to pull ARS skills; the process notes that imports are snapshots tied to a specific repo version.
Recent releases
- v3.19.0 (2026-07-22): Revision-round claim-drift guards, PDF read-integrity preflight, read-scope attestation; includes three advisory-or-opt-in integrity layers plus a launcher fix, bundling every release-worthy commit since v3.18.0.
- v3.18.0 (2026-07-18): Self-improvement survey integration referencing Ren et al. (2026); addresses issues #539–#542, #547–#550.
- v3.17.0 (2026-07-16): Pipeline boundary semantics and canonical cross-model handoff envelope; security: tools allowlist for top-level plugin agents.
- v3.16.0 (2026-07-12): Model tiering introduction with ARS_MODEL_TIERING switch.
- v3.15.0 (2026-07-04): Release-gate hardening with a discipline around changes.
Traction
- Stars: 40859
- Forks: 3262
- Open issues: 19
Behind the repo
- The project is linked with Claude Code plugin ecosystems and supports multiple installation methods and cross-model verification, with references to external studies and audits cited in the README.
Caveats
- License: none listed
- Created: 2026-02-26
- Last push: 2026-08-04
- Language: Python
- Topics include academic-pipeline, academic-writing, ai-research, claude, claude-code, literature-review, peer-review, prompt-engineering