-orange?style=flat){kind=icon}
What is this?
ARIS (Auto-Research-In-Sleep) is a pack of lightweight Markdown-only skills that turn coding agents like Claude Code into autonomous ML research assistants — covering idea discovery, cross-model review loops, and experiment automation. A standalone ARIS-Code CLI is also available.
ARIS (Auto-Research-In-Sleep) is a pack of lightweight Markdown-only skills that turn coding agents like Claude Code into autonomous ML research assistants — covering idea discovery, cross-model review loops, and experiment automation. A standalone ARIS-Code CLI is also available.Is this a framework I install, or a pack of skills for the agents I already use?Neither framework nor platform — it's Markdown-only skills with no lock-in that load into agents you already use; the README calls it a methodology you can take anywhere. For a bundled experience there's the standalone ARIS-Code CLI.
Neither framework nor platform — it's Markdown-only skills with no lock-in that load into agents you already use; the README calls it a methodology you can take anywhere. For a bundled experience there's the standalone ARIS-Code CLI.How do I load it into my agent?Drop the skills into your agent of choice — Claude Code, Codex CLI, Cursor, Trae, Antigravity, GitHub Copilot CLI, OpenClaw, or DeepSeek Harness. On DeepSeek Harness it's one command: dsh plugin --profile web add dsh-aris; otherwise grab the standalone ARIS-Code CLI.
Drop the skills into your agent of choice — Claude Code, Codex CLI, Cursor, Trae, Antigravity, GitHub Copilot CLI, OpenClaw, or DeepSeek Harness. On DeepSeek Harness it's one command: dsh plugin --profile web add dsh-aris; otherwise grab the standalone ARIS-Code CLI.Do I need an OpenAI API key for the cross-model review?No. Since the v0.4.17 MCP release, aris setup wires your ChatGPT subscription in as the reviewer through Codex MCP, so cross-model review works without a separate OpenAI API key.
No. Since the v0.4.17 MCP release, aris setup wires your ChatGPT subscription in as the reviewer through Codex MCP, so cross-model review works without a separate OpenAI API key.Any gotchas worth knowing before trying it?A few: the dsh-aris plugin needs pnpm on your PATH, the built-in ARIS-Monitor widget is macOS-only, and the CLI walks a fallback chain (Opus 5 → 4.8 → 4.7) so accounts without Claude 5 access keep working untouched.
A few: the dsh-aris plugin needs pnpm on your PATH, the built-in ARIS-Monitor widget is macOS-only, and the CLI walks a fallback chain (Opus 5 → 4.8 → 4.7) so accounts without Claude 5 access keep working untouched.Is it built only for academic ML research?No — ARIS-Anything extends the same five-step loop (plan, draft, adversarial review, iterate, persist) to any structured research: investment due diligence, legal and market research, self-driven learning, investigative journalism, engineering retrospectives.
No — ARIS-Anything extends the same five-step loop (plan, draft, adversarial review, iterate, persist) to any structured research: investment due diligence, legal and market research, self-driven learning, investigative journalism, engineering retrospectives.What does the cross-model review actually look like?One model does the work and a different one reviews it independently — typically Claude runs the research while Codex or GPT acts as reviewer. The README admits its own reviewer once proposed hashes nobody reads, which led to HERO: a ~550-token block for CLAUDE.md / AGENTS.md that bounds what the agent proposes (hashing, edge cases, rubrics, overbuild).
One model does the work and a different one reviews it independently — typically Claude runs the research while Codex or GPT acts as reviewer. The README admits its own reviewer once proposed hashes nobody reads, which led to HERO: a ~550-token block for CLAUDE.md / AGENTS.md that bounds what the agent proposes (hashing, edge cases, rubrics, overbuild).Can it catch an agent faking its research results?That's the Anti-Autoresearch audit: 61 signals — 46 integrity hack-patterns in 8 families, 13 AI-style impressions, and 2 advisory — compiled into a deterministic, reviewer-ready report. It relies on self-consistency and fabrication forensics rather than AI-text detection.
That's the Anti-Autoresearch audit: 61 signals — 46 integrity hack-patterns in 8 families, 13 AI-style impressions, and 2 advisory — compiled into a deterministic, reviewer-ready report. It relies on self-consistency and fabrication forensics rather than AI-text detection.Besides papers, what else can it produce?ARIS-Movie-Director turns a rough story into a movie told in still frames, checked scene by scene (the reference run has 19 scenes), and the same loop draws clean method or flow diagrams. The /render-html skill has also generated 23 bilingual ML interview cheat sheets and three long-form blogs.
ARIS-Movie-Director turns a rough story into a movie told in still frames, checked scene by scene (the reference run has 19 scenes), and the same loop draws clean method or flow diagrams. The /render-html skill has also generated 23 bilingual ML interview cheat sheets and three long-form blogs.How do I keep an eye on long research runs?ARIS-Monitor is a built-in always-on-top macOS widget that lights up red when a session waits for your approval — it starts in about 5 seconds with no clone, pip, or browser. For more, Claude Fleet is a local read-only web dashboard for many parallel Claude Code / Codex windows with full-text transcript search.
ARIS-Monitor is a built-in always-on-top macOS widget that lights up red when a session waits for your approval — it starts in about 5 seconds with no clone, pip, or browser. For more, Claude Fleet is a local read-only web dashboard for many parallel Claude Code / Codex windows with full-text transcript search.What are similar projects?graphify, superpowers-zh, CowAgent, agentic-awesome-skills, claude-scholar, token-monitor, evolver, skills
graphify, superpowers-zh, CowAgent, agentic-awesome-skills, claude-scholar, token-monitor, evolver, skillsWhich topics is it in?paper-writing, autonomous-agent, ai-research, claude-code-skills, antigravity, ai-tools, gpt, openclaw, dsh-plugin, deepseek-harness, cursor, skills
paper-writing, autonomous-agent, ai-research, claude-code-skills, antigravity, ai-tools, gpt, openclaw, dsh-plugin, deepseek-harness, cursor, skills