wanshuiyin/Auto-claude-code-research-in-sleep

Auto‑claude‑code‑research‑in‑sleep (ARIS) – Quick Overview

What it is – ARIS is a skill‑based workflow and CLI that lets you run autonomous research loops with large‑language‑model agents (Claude, Codex, OpenAI, DeepSeek, etc.). It is not a library of static prompts; it ships a full‑featured command‑line tool (ARIS‑Code) plus a collection of skills (small, reusable agent programs) that can be plugged into many existing AI‑coding environments such as Claude Code, Codex CLI, Cursor, GitHub Copilot CLI, DeepSeek Harness, and others.

Core idea – Treat a research task as a five‑step loop:

1. Plan – generate a brief outline.
2. Draft – write code / text.
3. Adversarial audit – an internal “integrity‑forensics” agent checks for hallucinations, self‑consistency, and fabrication.
4. Iterate – the original agent revises based on the audit.
5. Persist – results are stored in a research‑wiki for later reference.

The loop is fully automated (“research in sleep”) but always surfaces a review‑ready report that can be inspected by a human.

---

Key Features (as described in the README)

• Standalone CLI (ARIS‑Code) – download a single binary, run aris commands, and pick the LLM model (Claude Opus 5, Claude Fable 5, Sonnet 5, etc.).
• 82 bundled skills – ready‑made agents for tasks like code generation, paper summarisation, multimodal movie‑director pipelines, integrity‑forensics, web‑search, etc.
• Self‑audit – the Anti‑Autoresearch module provides 61 signals (46 integrity hack‑patterns, 13 AI‑style impressions, 2 advisory) to produce deterministic, reviewer‑ready reports.
• Cross‑model review – can use Claude as the primary model and fall back to OpenAI’s Codex MCP without needing an extra API key.
• Multimodal extension – the companion ARIS‑Movie‑Director project turns a story outline into a storyboard of still frames, each frame audited by a separate model.
• Adaptors for many ecosystems – plug‑in commands for Claude Code, Codex CLI, Cursor, Trae, Antigravity, Copilot CLI, OpenClaw, DeepSeek Harness, etc.
• Monitoring UI – a tiny macOS widget (aris‑monitor) and the more feature‑rich Claude Fleet dashboard show which agent windows are awaiting user approval.
• ARIS‑Anything – a generic wrapper that applies the same five‑step loop to non‑academic research (due‑diligence, legal research, market analysis, etc.).
• Release cadence – the repo tracks detailed changelogs; recent v0.4.24 adds Claude 5 model support, output‑folding to keep terminal noise low, and improved model‑fallback logic.

---

Getting Started (quick‑start commands from the README)

# Install the standalone CLI (download the latest release binary)
curl -L https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep/releases/latest/download/aris-cli-linux.tar.gz | tar xz
./aris --help

# Run the built‑in monitor (no clone, no pip)
cd aris-monitor && ./run.sh   # pops a tiny always‑on‑top widget

# Example: start a research session with Claude Opus 5
./aris setup                 # wizard to store your Claude credentials
./aris new "Summarise the latest LLM safety paper"
# The agent will plan, draft, audit, iterate and finally write a markdown report.

---

Who Might Use It

• Researchers who want reproducible, auditable LLM‑driven literature reviews.
• Developers building AI‑assistant tools who need a ready‑made “research loop” to plug into their product.
• Students / job‑seekers – the linked ARIS‑in‑AI‑Offer repo generates bilingual cheat‑sheets for ML/LLM interview preparation.
• Teams that already use Claude Code, Codex CLI, or DeepSeek Harness and want a unified workflow.

---

Community & Resources

• Documentation – Intro HTML page, detailed skill definitions (AGENT_GUIDE.md), and a full CLI README (ARIS‑Code‑README_EN.md).
• Demo videos – the README links to a 19‑scene movie generated by ARIS‑Movie‑Director, showing the audit scores.
• External mentions – featured on PaperWeekly, listed in awesome‑agent‑skills, highlighted by AI Digital Crew.
• Support – a “Join Community” link (not expanded in the README) and issue tracker for bug reports and feature requests.

---

Citation

If you use ARIS in a publication, the authors provide a BibTeX entry under the 📖 Cite Us badge in the README.

---

Bottom line – ARIS is a genuine, actively maintained open‑source project that provides a reproducible, self‑auditing research workflow for LLM agents, packaged as a CLI and a set of plug‑ins for many popular AI‑coding environments.