ARIS Auto-Research-In-Sleep

wanshuiyin/Auto-claude-code-research-in-sleep publishes ARIS, a Markdown-first research-agent workflow system. The repository describes ARIS as a methodology rather than a platform: its core behavior is stored in SKILL.md files that can be copied into Claude Code, Codex, Cursor, OpenClaw, Antigravity, GitHub Copilot CLI, Trae, or other Markdown-compatible agent workflows.

Use this listing for the ARIS skill workflow pack itself. Use narrower entries when reviewing a single ARIS skill, a separate MCP reviewer server, or a platform-specific adaptation.

Knowledge Freshness

The repository is active, with current release and GitHub metadata verified on 2026-06-18. The README and AGENT_GUIDE.md document Claude Code usage, Codex skill mirrors, cross-model reviewer routes, OpenClaw adaptation notes, and a large catalog of research skills.

Research tooling, model behavior, MCP reviewer setup, OpenClaw support, GPU environments, paper venues, citation sources, and ML baselines can change quickly. Treat ARIS as a structured research automation harness, then verify commands, dependencies, venue rules, citations, datasets, reviewer configuration, and experiment evidence in the target project before relying on generated results.

Retrieval Sources

This listing is grounded in:

• The upstream repository README.
• The upstream AGENT_GUIDE.md.
• Representative ARIS SKILL.md files for research pipeline, idea discovery, experiment bridge, auto-review loop, paper writing, and rebuttal workflows.
• The shared assurance contract reference.
• The OpenClaw adaptation guide.
• The MIT license file.
• Current GitHub repository metadata.

Core Workflow

Clone the repository:

git clone https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep.git

Review the relevant skill files, then copy only the skills you want into your agent host. Codex users should review the skills/skills-codex/ mirror before copying, while OpenClaw users should read the OpenClaw adaptation guide and map skills to file-first task phases.

Common ARIS entry points include:

/research-pipeline "factorized gap in discrete diffusion LMs"
/idea-discovery "improve method X"
/experiment-bridge "turn this idea into an experiment plan"
/auto-review-loop "review paper/ and experiments/"
/paper-writing "results/ + notes/"
/rebuttal "paper/ + reviews" -- venue: ICML

Capability Scope

The upstream AGENT_GUIDE.md organizes ARIS around research lifecycle workflows:

The guide also documents assurance and audit skills such as /experiment-audit, /result-to-claim, /paper-claim-audit, /citation-audit, and /kill-argument. Those checks are important because research-agent output can look polished while still being weak, stale, or unsupported.

Platform Fit

ARIS is built around portable Markdown skills rather than a single hosted runtime. The repository documents several usage patterns:

• Claude Code skills under skills/<name>/SKILL.md.
• Codex-specific mirrors under skills/skills-codex/.
• Codex plus Claude-review or Gemini-review overlays under the corresponding reviewer skill folders.
• Cursor, Trae, Antigravity, GitHub Copilot CLI, and manual Markdown skill usage.
• OpenClaw adaptation through a file-first phase map that writes outputs such as lit_scan.md, idea_report.md, experiment_plan.md, runbook.md, and review_loop.md.

This makes ARIS useful for teams comparing "Claude Code research pipeline", "Codex research skills", "OpenClaw research workflow", and "cross-model review loop" approaches without committing to one agent runtime immediately.

Production Rules

• Start with a narrow research objective and define success criteria before invoking an autonomous pipeline.
• Keep repository changes, shell execution, GPU/cloud jobs, and long-running loops behind checkpoints until the workflow is trusted.
• Use the assurance and audit skills before turning results into paper claims.
• Verify citations, datasets, baselines, venue rules, and license constraints against primary sources.
• Keep unpublished reviews, internal results, private datasets, and proprietary code out of prompts sent to providers that are not approved for that data.
• Treat cross-model reviewer output as adversarial feedback, not a replacement for human ML review.

Source Review

• The repository README describes ARIS as "lightweight Markdown-only skills" for autonomous ML research, cross-model review loops, idea discovery, and experiment automation.
• The README states that ARIS works with Claude Code, Codex, Cursor, Trae, Antigravity, GitHub Copilot CLI, OpenClaw, and standalone/manual workflows.
• AGENT_GUIDE.md describes ARIS as a research harness whose source of truth is skills/<name>/SKILL.md plus shared reference contracts.
• The guide documents 79 skills and maps common workflows across idea discovery, experiment bridging, review loops, paper writing, rebuttal, resubmission, talks, slides, posters, audits, and Research Wiki usage.
• skills/research-pipeline/SKILL.md chains idea discovery, experiment bridge, auto-review loop, and paper writing, with parameters for auto-proceed, human checkpoints, code review, base repositories, venues, HTML rendering, and resumable execution.
• docs/OPENCLAW_ADAPTATION.md maps ARIS workflows into OpenClaw-friendly, file-first phases with explicit output artifacts.
• The repository license is MIT.

Duplicate Check

No existing HeyClaude content entry or open PR was found for ARIS, Auto-Research-In-Sleep, or wanshuiyin/Auto-claude-code-research-in-sleep at the time this listing was created.

Disclosure

This is a community project listing based on public source material. It is not an endorsement that autonomous research results, generated experiments, paper claims, or rebuttal drafts are correct, novel, reproducible, or venue-ready without independent review.