[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-70543":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":9,"totalLinesOfCode":9,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":9,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":9,"rankLanguage":9,"license":9,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":22,"hasPages":22,"topics":24,"createdAt":9,"pushedAt":9,"updatedAt":45,"readmeContent":46,"aiSummary":47,"trendingCount":16,"starSnapshotCount":16,"syncStatus":48,"lastSyncTime":49,"discoverSource":50},70543,"Auto-claude-code-research-in-sleep","wanshuiyin\u002FAuto-claude-code-research-in-sleep","wanshuiyin","ARIS ⚔️ (Auto-Research-In-Sleep) — Lightweight Markdown-only skills for autonomous ML research: cross-model review loops, idea discovery, and experiment automation. No framework, no lock-in — works with Claude Code, Codex, OpenClaw, or any LLM agent.",null,"https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep","Python",11949,1099,18,43,0,81,562,3065,440,44.12,false,"main",[25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44],"ai-research","autonomous-agent","claude-code","codex","mcp","ml-research","paper-review","research-automation","ai-tools","claude","gpt","llm","machine-learning","openai","paper-writing","claude-code-skills","deep-learning","idea-generation","mcp-server","aris","2026-06-12 02:02:34","# Auto-claude-code-research-in-sleep (ARIS ⚔️🌙)\n\n💡 *Use ARIS in Claude Code \u002F Cursor \u002F Trae as a skill-based workflow, or get the full experience with the standalone CLI — enjoy any way you like!*\n\n🤖 **AI agents:** Read [`AGENT_GUIDE.md`](AGENT_GUIDE.md) instead — structured for LLM consumption, not human browsing.\n\n🔥 [**ARIS-Code CLI — 独立安装版**](docs\u002FARIS-Code-README_CN.md) · [English](docs\u002FARIS-Code-README_EN.md) | [⬇️ Download](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Freleases\u002Flatest)\n\n> 📰 **ARIS-Code v0.4.5 → v0.4.11** (2026-05) — Seven-release polish sequence. **New providers**: DeepSeek V4 Pro \u002F Xiaomi MiMo \u002F Qwen 3.6 \u002F Doubao \u002F Custom OpenAI-compatible \u002F DashScope Coding Plan (via `native-tls` switch). **Reasoning + tool-use first class**: `reasoning_effort='xhigh'` on the wire for o-series \u002F gpt-5.5 \u002F DeepSeek-thinking, `reasoning_content` cache+replay for Kimi \u002F Moonshot \u002F Xiaomi MiMo \u002F DeepSeek-R1, thinking content blocks end-to-end. **Skills bundle catches up to main** (v0.4.11): 65→74 user-facing skills (10 new: `\u002Fcitation-audit`, `\u002Fexperiment-queue`, `\u002Fkill-argument`, `\u002Fresubmit-pipeline`, `\u002Fpaper-talk`, `\u002Fslides-polish`, `\u002Foverleaf-sync`, `\u002Fgemini-search`, `\u002Fopenalex`, `\u002Fqzcli`), 46 SKILL.md refreshed with canonical resolver chain + submission assurance gate; tools\u002F 9→18 helpers (including `research_wiki.py` 315→767 lines with canonical `ingest_paper` API); new `tools\u002Fsync_main_skills.sh` + 3 CI drift tests prevent future drift. **Stream + MCP reliability** (v0.4.10): both Anthropic and OpenAI streaming whole-stream-restart on chunk decode failure (closes `#228` \"error decoding response body\" loop); MCP stdio gains 300s default timeout, `response.id ↔ request.id` correlation, dead-process respawn (closes `#151`\u002F`#172` \"Calling codex...\" stalls). **Skill helper subsystem rewrite**: bundled helpers materialise into `~\u002F.config\u002Faris\u002Fcache\u002F\u003Cversion>\u002F` at startup (no more cwd pollution), `SkillOutput.helperReport` JSON + 4-layer fallback chain, new `integration-contract.md` with 6 failure policies, inventory test + smoke regression guard. **Multi-provider pricing**: GPT-5.5\u002F5.4\u002Fo-series + Gemini 2.5\u002F2.0 + DeepSeek V3\u002FV4\u002FR1 + GLM\u002FMiniMax\u002FKimi\u002FMiMo\u002FQwen\u002FDoubao all priced correctly (OpenAI cache_read = input × 0.1, fixing 5× cost overstatement). **Critical bug fixes**: `PermissionMode::Prompt` was silently allowing every tool (derived-`Ord` bug, every version pre-v0.4.6); hardcoded `current_date = \"2026-03-31\"` made models reject real post-cutoff data; Custom reviewer reset to gpt-5.5 every restart (setup menu option 9 vs 8 typo); third-party Anthropic-compat proxies hit `missing field signature` ([#228](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F228)); gpt-5.5 \u002F o3 \u002F o4 + tools 400 on OpenAI. Credits: [@GetIT-Sunday](https:\u002F\u002Fgithub.com\u002FGetIT-Sunday), [@Anduin9527](https:\u002F\u002Fgithub.com\u002FAnduin9527), [@GO-player-hhy](https:\u002F\u002Fgithub.com\u002FGO-player-hhy), [@Jxy-yxJ](https:\u002F\u002Fgithub.com\u002FJxy-yxJ), [@screw-44](https:\u002F\u002Fgithub.com\u002Fscrew-44), [@StevenUST](https:\u002F\u002Fgithub.com\u002FStevenUST).\n>\n> \u003Cdetails>\u003Csummary>Per-release details (v0.4.5 → v0.4.11)\u003C\u002Fsummary>\n>\n> **v0.4.11** (2026-05-18) — Skills bundle refresh + sync infrastructure. The embedded skills set in the v0.4.10 binary had fallen behind main (~6 of 56 main `skills\u002F` commits had been cherry-picked); v0.4.11 syncs the full set and ships sync infrastructure so the gap can't silently reopen. Bundle: 65→74 user-facing skills, 34→49 helper resources. 10 new skills bundled: `\u002Fcitation-audit` (fourth-layer bibliography audit), `\u002Fexperiment-queue` (SSH multi-seed job queue with OOM retry), `\u002Fkill-argument` (two-thread adversarial review for theory papers), `\u002Fresubmit-pipeline` (W5: text-only port to a new venue), `\u002Fpaper-talk` (end-to-end conference talk pipeline), `\u002Fslides-polish` (per-page Codex layout review), `\u002Foverleaf-sync` (two-way Overleaf Git-bridge), `\u002Fgemini-search` + `\u002Fopenalex` (broader literature sources), `\u002Fqzcli` (Qizhi GPU jobs). 46 existing SKILL.md refreshed — most critically the canonical resolver chain rollout (closes real user incident where `\u002Fresearch-wiki` was empty for a week from hardcoded `tools\u002Fresearch_wiki.py`), submission assurance gate + external verifier (`\u002Fpaper-writing` Phase 6 now functions). tools\u002F goes 9→18: 9 baseline helpers refreshed (`research_wiki.py` 315→767 lines with canonical `ingest_paper` API), 9 new helpers (`extract_paper_style.py`, `figure_renderer.py`, `paper_illustration_image2.py`, `overleaf_{setup,audit}.sh`, `verify_wiki_coverage.sh`, `watchdog.py`, `experiment_queue\u002F{build_manifest,queue_manager}.py`). New `tools\u002Fsync_main_skills.sh` automates main → bundle rsync with symlink pre-flight + codex-mirror prune + `SKILLS_SOURCE_COMMIT` pinning. 3 new CI drift tests in `crates\u002Fruntime\u002Fsrc\u002Fcache.rs` cover all 4 resolver layer patterns. Gemini MCP calls in `\u002Fresearch-lit` and `\u002Fgemini-search` now pass `model: 'auto-gemini-3'` (avoids silent downgrade to 2.5-pro on OAuth-personal capacity exhaustion). CLI runtime unchanged — codex-audit P1 follow-ups remain on v0.4.12 backlog. Cross-reviewed by Codex MCP (gpt-5.5 xhigh) across 5 rounds (REQUEST CHANGES → APPROVE WITH NITS → NO-GO → GO → final GO).\n>\n> **v0.4.10** (2026-05-17) — Stream + MCP reliability + multi-provider pricing. C6 whole-stream restart in Anthropic `MessageStream` + OpenAI SSE loop on chunk decode failure \u002F premature EOF (`ARIS_STREAM_RETRY`, default 2, clamp 0..=5, fires only when nothing emitted yet — closes [#228](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F228)-style \"error decoding response body\" loop). M3 MCP stdio gains 300s default `tokio::time::timeout` over send+read (override `MCP_REQUEST_TIMEOUT_SECS`, clamp 1..=1800); `response.id ↔ request.id` correlation; `ensure_server_ready()` `try_wait()` dead-process respawn; `kill().await` on all failure paths so the next call starts clean (closes [#151](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F151) \u002F [#172](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F172) \"Calling codex...\" stalls). C8\u002FP4 OpenAI streaming requests now send `stream_options.include_usage:true` + parse `cached_tokens`; Anthropic streaming merges `MessageStart.usage` (input\u002Fcache) with `MessageDelta.usage` (output). C9 multi-provider pricing registry (15+ models, OpenAI cache_read = input × 0.1 corrects 5× generic overstatement, DeepSeek cache_hit\u002Fcache_miss tiers, `has_word()` boundary matcher for `provider\u002F\u003Cmodel>` slugs). 9 dead-code warnings cleared; `aris setup` help text synced with actual behaviour.\n>\n> **v0.4.9** (2026-05-17) — Closes Codex v0.4.7 audit residuals (L1 TLS double-stack, L3 reasoning_cache compaction misalign, L4 reasoning replay unbounded). 2 new skills bundled (`\u002Ffigure-spec` + `\u002Fpaper-illustration-image2` with `scripts\u002F` subdirs, new Layer 0b = `$ARIS_CACHE_DIR\u002Fskills\u002F\u003Cname>\u002Fscripts\u002F`); `research_wiki.py` promoted to shared `tools\u002F` (9+ callers); 5 more SKILL.md migrated to fallback chain.\n>\n> **v0.4.8** (2026-05-17) — Skill helper subsystem rewrite. Bundled helpers extract to `~\u002F.config\u002Faris\u002Fcache\u002F\u003Cversion>\u002F` at startup; every Skill invocation surfaces `helperReport` JSON + 4-layer resolver preamble; `\u002Fskills export` copies helpers; new `integration-contract.md` with 6 failure policies; 8 shared helpers (arxiv\u002Fdeepxiv\u002Fexa\u002FS2\u002Fopenalex\u002Fsave_trace\u002Fverify_papers\u002Fverify_paper_audits) bundled; `\u002Fresearch-lit` + `\u002Fdeepxiv` migrated. Plus 4 bug fixes: gpt-5.5+tools 400 on OpenAI; Custom reviewer reset; missing `signature` field ([#228](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F228)); `--version` Build date hardcoded.\n>\n> **v0.4.7** (2026-05-16) — DashScope Coding Plan 405 fixed ([#159](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F159)) via `native-tls` switch ([#225](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F225)); `reasoning_content` replay for all reasoning models (OpenAI o1\u002Fo3\u002Fo4 \u002F DeepSeek-R1 etc.), not just Kimi ([#226](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F226)); 600+ lines dead code cleanup + `rustyline` dep removed + \"Claw Code\" → \"ARIS-Code\" rebrand.\n>\n> **v0.4.6** (2026-05-14) — 🚨 Two long-standing silent bugs fixed: `PermissionMode::Prompt` silently allowed every tool (derived-`Ord` bug); system prompt hardcoded `current_date = \"2026-03-31\"` made models reject post-cutoff data as future\u002Fprompt-injection. Plus Custom OpenAI-compatible provider (`\u002Fsetup` option 11) with dynamic `\u002Fmodels` discovery ([@Anduin9527](https:\u002F\u002Fgithub.com\u002FAnduin9527) [#221](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F221) + [#222](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F222)).\n>\n> **v0.4.5** (2026-05-13) — First-class reasoning-model support: thinking content blocks end-to-end (fixes [#161](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F161)) + `reasoning_effort='xhigh'` for GPT-5.5 \u002F o1 \u002F o3 \u002F o4 \u002F DeepSeek-thinking. DeepSeek V4 Pro + Xiaomi MiMo + Qwen 3.6 + Doubao in `\u002Fsetup` (options 7-10). Object-style hooks parser. Default model bumped to Claude Opus 4.7 + GPT-5.5. REPL input hardening (multi-line wrap \u002F Cmd+V paste \u002F CJK boundary). GitHub Actions CI. Credits: [@GO-player-hhy](https:\u002F\u002Fgithub.com\u002FGO-player-hhy) ([#186](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F186)), [@Jxy-yxJ](https:\u002F\u002Fgithub.com\u002FJxy-yxJ) ([#171](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F171)), [@GetIT-Sunday](https:\u002F\u002Fgithub.com\u002FGetIT-Sunday) ([#216](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F216) partial).\n>\n> \u003C\u002Fdetails>\n>\n> \u003Cdetails>\u003Csummary>Older versions\u003C\u002Fsummary>\n>\n> **v0.4.4** (2026-04-20) — **Setup UX + reviewer routing fixes** (resolves [#158](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F158), [#162](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F162)) | `\u002Fsetup` no longer forces Bearer for Anthropic + custom URL | Provider-aware proxy URL hints | Stale state no longer leaks across provider switches | LlmReview smart fallback\n>\n> **v0.4.3** (2026-04-17) — **Third-party Anthropic-compat proxy support** (Bedrock etc.) | Skip beta flags that proxies reject | Propagate custom base URL for `anthropic` provider | Credit [@screw-44](https:\u002F\u002Fgithub.com\u002Fscrew-44)\n>\n> **v0.4.2** (2026-04-17) — **Auto-compaction corruption fix** | Compaction summary preserved on OpenAI-compat executors | Shell-provided API keys no longer erased on launch\n>\n> **v0.4.1** (2026-04-15) — **Plan mode** (`\u002Fplan`) | Cooperative Ctrl+C interrupt | Auto-retry (429\u002F5xx\u002Fnetwork) | **Research Wiki** 📚 (persistent knowledge base) | **Self-Evolution** 🧬 (`\u002Fmeta-optimize`) | Local models (LM Studio\u002FOllama) | 62 skills synced\n>\n> **v0.3.11** (2026-04-13) — Reviewer Anthropic-compatible mode (Claude via proxy)\n>\n> **v0.3.9** (2026-04-11) — Proxy\u002Fcustom base URL (CCSwitch) | Local models (LM Studio\u002FOllama) | Windows (experimental)\n>\n> **v0.3.5** (2026-04-08) — **Research Wiki** (persistent papers\u002Fideas\u002Fexperiments\u002Fclaims + relationship graph) | **Meta-Optimize** self-evolution (analyze logs → propose SKILL.md patches)\n>\n> **v0.3.0** (2026-04-03) — Multi-file memory index | Rich task system (TodoWrite) | `\u002Fplan` | Security hardening\n>\n> **v0.2.2** (2026-04-03) — `\u002Fplan` step-by-step planning | `\u002Ftasks` persistent tracking\n>\n> **v0.2.1** (2026-04-03) — Persistent Memory | Kimi K2.5 multi-turn fix | CJK cursor fix\n>\n> **v0.2.0** (2026-04-02) — Open source | Kimi + MiniMax + GLM support | Smart LlmReview routing | CI\u002FCD\n>\n> **v0.1.0** (2026-04-02) — Initial release | Multi-executor & reviewer | 42 bundled skills\n>\n> \u003C\u002Fdetails>\n\n\u003Cimg src=\"docs\u002Faris-code-banner.png\" width=\"600\" alt=\"ARIS-Code CLI\">\n\n![ARIS Logo](docs\u002Faris_logo.svg)\n\n![Hero](docs\u002Fhero_combined.svg)\n\n[中文版 README](README_CN.md) | English\n\n> 🌙 **Let Claude Code do research while you sleep.** Wake up to find your paper scored, weaknesses identified, experiments run, and narrative rewritten — autonomously.\n>\n> 🪶 **Radically lightweight — zero dependencies, zero lock-in.** The entire system is plain Markdown files. No framework to learn, no database to maintain, no Docker to configure, no daemon to babysit. Every skill is a single `SKILL.md` readable by any LLM — swap Claude Code for [Codex CLI](skills\u002Fskills-codex\u002F), [OpenClaw](docs\u002FOPENCLAW_ADAPTATION.md), [Cursor](docs\u002FCURSOR_ADAPTATION.md), [Trae](docs\u002FTRAE_ARIS_RUNBOOK_EN.md), [Antigravity](docs\u002FANTIGRAVITY_ADAPTATION.md), [Copilot CLI](docs\u002FCOPILOT_CLI_ADAPTATION.md), Windsurf, or your own agent and the workflows still work. Fork it, rewrite it, adapt it to your stack.\n>\n> *💡 ARIS is a methodology, not a platform. What matters is the research workflow — take it wherever you go. 🌱*\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2605.03042\">\n    \u003Cimg src=\"docs\u002Fhf_daily_paper_1.svg\" alt=\"Hugging Face Daily Paper · #1 Paper of the Day\" width=\"360\">\n  \u003C\u002Fa>\n\u003C\u002Fp>\n\n[![Technical Report](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FTechnical%20Report-arXiv%3A2605.03042-b31b1b?style=flat&logo=arxiv)](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2605.03042) · [![ARIS Intro Slides](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FARIS%20Intro%20Slides-PDF-EC1C24?style=flat&logo=adobeacrobatreader&logoColor=white)](docs\u002Faris_intro_slides.pdf) · [![Featured on PaperWeekly](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFeatured%20on-PaperWeekly-red?style=flat)](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FtDniVryVGjDkkkWl-5sTkQ) · [![PaperWeekly — MiniMax-M2.7](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaperWeekly-MiniMax--M2.7-red?style=flat)](https:\u002F\u002Fmp.weixin.qq.com\u002Fs\u002FKLFU74lAL2FAIc9K6i1Kqg) · [![Featured in awesome-agent-skills](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFeatured%20in-awesome--agent--skills-blue?style=flat&logo=github)](https:\u002F\u002Fgithub.com\u002FVoltAgent\u002Fawesome-agent-skills) · [![AI Digital Crew - Project of the Day](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FAI%20Digital%20Crew-Project%20of%20the%20Day%20(2026.03.14)-orange?style=flat)](https:\u002F\u002Faidigitalcrew.com) · [💬 Join Community](#-community) · [![Cite](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F📖_Cite_Us-BibTeX-green?style=flat)](#-citation)\n\nCustom [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code) skills for autonomous ML research workflows. These skills orchestrate **cross-model collaboration** — Claude Code drives the research while an external LLM (via [Codex MCP](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex)) acts as a critical reviewer. 🔀 **Also supports [alternative model combinations](#-alternative-model-combinations) (Kimi, LongCat, DeepSeek, etc.) — no Claude or OpenAI API required.** For example, [MiniMax-M2.7 + GLM-5 or GLM-5 + MiniMax-M2.7](docs\u002FMiniMax-GLM-Configuration.md). 🤖 **[Codex CLI native](skills\u002Fskills-codex\u002F)** — full skill set also available for OpenAI Codex. 🖱️ **[Cursor](docs\u002FCURSOR_ADAPTATION.md)** — works in Cursor too. 🖥️ **[Trae](docs\u002FTRAE_ARIS_RUNBOOK_EN.md)** — ByteDance AI IDE. 🚀 **[Antigravity](docs\u002FANTIGRAVITY_ADAPTATION.md)** — Google's agent-first IDE. 🐙 **[Copilot CLI](docs\u002FCOPILOT_CLI_ADAPTATION.md)** — GitHub's terminal agent (native SKILL.md + MCP). 🆓 **[Free tier via ModelScope](docs\u002FMODELSCOPE_GUIDE.md) — zero cost, zero lock-in.**\n\n> 💭 **Why not self-play with a single model?** Using Claude Code subagents or agent teams for both execution and review is technically possible, but tends to fall into **local minima** — the same model reviewing its own patterns creates blind spots.\n>\n> *Think of it like adversarial vs. stochastic bandits: a single model self-reviewing is the stochastic case (predictable reward noise), while cross-model review is adversarial (the reviewer actively probes weaknesses the executor didn't anticipate) — and adversarial bandits are fundamentally harder to game.*\n>\n> 💭 **Why two models, not more?** Two is the minimum needed to break self-play blind spots, and 2-player games converge to Nash equilibrium far more efficiently than n-player ones. Adding more reviewers increases API cost and coordination overhead with diminishing returns — the biggest gain is going from 1→2, not 2→4.\n>\n> Claude Code's strength is fast, fluid execution; Codex (GPT-5.4 xhigh) is slower but more deliberate and rigorous in critique. These complementary styles — **speed × rigor** — produce better outcomes than either model talking to itself.\n>\n> 🧿 **Want the strongest possible reviewer?** Add `— reviewer: oracle-pro` to any skill to route reviews through **GPT-5.4 Pro** via [Oracle MCP](https:\u002F\u002Fgithub.com\u002Fsteipete\u002Foracle). Pro-level reasoning for proof verification, experiment auditing, and final stress tests. Works with API key or free browser mode. [Setup →](#-optional-gpt-54-pro-via-oracle)\n\n## 🎯 More Than Just a Prompt\n\n> These are full pipelines — you can also use each workflow independently. Already have an idea? Skip to Workflow 1.5. Have results? Jump to Workflow 3. Got reviews? Jump to Workflow 4. Want persistent memory? Enable [Research Wiki](#-research-wiki--persistent-research-memory). See [Quick Start](#-quick-start) for all commands and [Workflows](#-workflows) for the full breakdown.\n\n**Basic mode** — give ARIS a research direction, it handles everything:\n\n```\n\u002Fresearch-pipeline \"factorized gap in discrete diffusion LMs\"\n```\n\n**🔥 Targeted mode** — got a paper you want to improve? Give ARIS the paper + the code:\n\n```\n\u002Fresearch-pipeline \"improve method X\" — ref paper: https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.04329, base repo: https:\u002F\u002Fgithub.com\u002Forg\u002Fproject\n```\n\nARIS reads the paper → finds its weaknesses → clones the codebase → generates ideas that specifically fix *those* weaknesses with *that* code → runs experiments → writes your paper. Like telling a research assistant: *\"read this paper, use this repo, find what's missing, and fix it.\"*\n\n> Mix and match: `ref paper` only = \"what can be improved?\", `base repo` only = \"what can I build with this code?\", both = \"improve *this* paper using *this* code.\"\n\n**🔥 Rebuttal mode** — reviews just dropped? Don't panic. ARIS reads every concern, builds a strategy, and drafts a rebuttal that's grounded, structured, and under the character limit:\n\n```\n\u002Frebuttal \"paper\u002F + reviews\" — venue: ICML, character limit: 5000\n```\n\nThree safety gates — rebuttal will NOT finalize if any fails:\n- 🔒 **No fabrication** — every claim maps to paper\u002Freview\u002Fuser-confirmed result\n- 🔒 **No overpromise** — every promise is user-approved\n- 🔒 **Full coverage** — every reviewer concern is tracked\n\nTwo outputs: `PASTE_READY.txt` (exact char count, paste to venue) + `REBUTTAL_DRAFT_rich.md` (extended version for manual editing).\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Show rebuttal parameters\u003C\u002Fb> — venue, character limit (required), quick mode, auto experiment, stress test rounds, followup rounds\u003C\u002Fsummary>\n\n| Parameter | Default | What it does |\n|-----------|---------|-------------|\n| `venue` | `ICML` | Target venue (ICML\u002FNeurIPS\u002FICLR\u002FCVPR\u002FACL\u002FAAAI\u002FACM) |\n| `character limit` | — | **Required.** Hard character limit for rebuttal text |\n| `quick mode` | `false` | Stop after parsing + strategy (Phase 0-3). See what reviewers want before drafting |\n| `auto experiment` | `false` | Auto-run supplementary experiments via `\u002Fexperiment-bridge` when reviewers ask for new evidence |\n| `max stress test rounds` | `1` | How many times GPT-5.4 xhigh stress-tests the draft |\n| `max followup rounds` | `3` | Per-reviewer follow-up round limit |\n\n\u003C\u002Fdetails>\n\n**After acceptance** — your paper is in, now prepare the presentation:\n\n```\n\u002Fpaper-slides \"paper\u002F\"     # → Beamer PDF + PPTX + speaker notes + Q&A prep\n\u002Fpaper-poster \"paper\u002F\"     # → A0\u002FA1 poster PDF + editable PPTX + SVG\n```\n\n> *💡 From idea to paper to podium — one toolchain. 🌱*\n\n## 📢 What's New\n\n- **2026-05-17** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🐙 **[GitHub Copilot CLI adaptation](docs\u002FCOPILOT_CLI_ADAPTATION.md)** — native `SKILL.md` + MCP support, no skill mirror needed. Installer (`install_aris_copilot.sh`) + smart-updater + 13-test suite. Community contribution by [@EarendelH](https:\u002F\u002Fgithub.com\u002FEarendelH) ([#229](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F229), closes [#214](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F214) \u002F [#227](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F227) \u002F [#203](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F203)).\n- **2026-05-17** — ![FIX](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFIX-orange?style=flat-square) 🛠 **Tools-stability roadmap (Phase 1+2+3) complete** (closes [#176](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F176) \u002F [#177](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F177) \u002F [#178](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F178)). Community reported that helper scripts weren't propagating into user projects after `install_aris.sh`. **Phase 1** — every SKILL.md caller of the 10 canonical helpers now resolves via the strict-safe 3-layer chain `.aris\u002Ftools\u002F` → `tools\u002F` → `$ARIS_REPO\u002Ftools\u002F` documented in [`integration-contract.md`](skills\u002Fshared-references\u002Fintegration-contract.md) §2 (which also defines 5 failure policies A\u002FB\u002FC\u002FD1\u002FD2\u002FE). **Phase 2** — new [advisory CI lint](.github\u002Fworkflows\u002Flint-skills-helpers.yml) catches hardcoded `python3 tools\u002Ffoo.py` patterns in PR-modified SKILL.md (advisory only, never fails CI). **Phase 3** — three single-owner helpers (`figure-spec`, `paper-illustration-image2`, `experiment-queue`) moved into their SKILL's `scripts\u002F` subdirectory; owner SKILLs use Layer 0 `${CLAUDE_SKILL_DIR}\u002Fscripts\u002F` ahead of the canonical chain; legacy `tools\u002F` paths retained as `os.execv` Python forwarding shims. **⚠️ Existing users**: no action needed — legacy `tools\u002F` entries are now shims. If you haven't run `install_aris.sh` since 2026-04-30, one idempotent rerun catches everything up.\n- **2026-05-14** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🩹 **`\u002Fpaper-plan` + `\u002Fpaper-write` learn `GAP_REPORT.md` + `\u003C!-- DATA_NEEDED -->` discipline** ([#217](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F217)). When `— style-ref:` is set and the user's project has structural assets (`figures\u002F`, `results\u002F`, `NARRATIVE_REPORT.md`, etc.), `\u002Fpaper-plan` emits a **Gap Report** mapping the exemplar's section topology + density (from `style_profile.md`) against your actual assets — surfacing slots you have **no evidence to fill** (e.g., \"exemplar has 3×4 ablation table, you have no ablation data\"). Then `\u002Fpaper-write` writes `\u003C!-- DATA_NEEDED: \u003CSlot ID> — \u003Cdescription> -->` HTML comments **instead of fabricating content** at missing slots — invisible in the compiled PDF, `grep`-friendly for human triage \u002F `\u002Fexperiment-bridge` follow-up. Narrow carve-out from the \"no placeholders\" rule, scoped to GAP_REPORT-listed slots only. Original idea by [@zhangpelf](https:\u002F\u002Fgithub.com\u002Fzhangpelf).\n\u003Cdetails>\n\u003Csummary>Earlier updates (2026-03-12 — 2026-05-14, 63 entries)\u003C\u002Fsummary>\n\n- **2026-05-14** — ![BREAKING](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FBREAKING-purple?style=flat-square) ⚙️ **Default reviewer model: `gpt-5.4` → `gpt-5.5`** across ~30 SKILL.md `REVIEWER_MODEL` defaults. Codex MCP has routed `gpt-5.5` as the default since 2026-04-24; this catches the docs up to runtime. **⚠️ Behavior changes**: (a) `.aris\u002Ftraces\u002F*` JSONs from prior runs are **not reproducible** — re-runs on 5.5 may emit different `WARN\u002FFAIL` verdicts on borderline cases (reviewer-quality lift, not regression). (b) ChatGPT Plus\u002FPro monthly quotas drain faster under heavy use. **Fallback**: pass `— reviewer-model: gpt-5.4` per invocation, or pin `REVIEWER_MODEL = gpt-5.4` per skill. Oracle Pro tier (routed via `— reviewer: oracle-pro`) is a separate path and unaffected.\n- **2026-05-13** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🔍 **[`tools\u002Fverify_papers.py`](tools\u002Fverify_papers.py) + Pre-Search Verification Protocol — anti-hallucination filter for literature-facing skills**. New helper does 3-layer fallback verification (arXiv batch API up to 40 IDs\u002Frequest → CrossRef DOI lookup → Semantic Scholar fuzzy title match, default 0.6 word-overlap) and emits 4-state per-paper status (`verified` \u002F `unverified` \u002F `verify_pending` \u002F `error`) plus a top-level verdict aligning with `assurance-contract.md` (`PASS` \u002F `WARN` \u002F `BLOCKED` \u002F `ERROR`). Transient failures (5xx, timeouts, 429) are tagged `verify_pending` and **excluded from the hallucination rate** so network blips don't get conflated with fabricated references. Per-project cache at `\u003Cproject>\u002F.aris\u002Fcache\u002Fverify_papers.json` with 30-day TTL; canonical key priority `arxiv:{id_without_version}` → `doi:{lowercase}` → `title:{sha1[:16]}`. New `Pre-Search Verification Protocol` subsection in [`shared-references\u002Fcitation-discipline.md`](skills\u002Fshared-references\u002Fcitation-discipline.md) makes the split explicit: this protocol is the **fast filter** between SEARCH (Step 1) and full VERIFY (Step 2); `\u002Fcitation-audit` and `\u002Fpaper-claim-audit` remain the submission-time audit gates and are not replaced. [`\u002Fresearch-lit`](skills\u002Fresearch-lit\u002FSKILL.md) gets a mandatory `Step 1.5: Verify Candidate Papers` calling the helper; [`\u002Fidea-creator`](skills\u002Fidea-creator\u002FSKILL.md) and [`\u002Fnovelty-check`](skills\u002Fnovelty-check\u002FSKILL.md) add a Key Rule reference for cited Closest Prior Work \u002F landscape entries. Unverified papers are **retained** in output tagged `[UNVERIFIED]` (retention-over-silent-removal) so search-quality issues stay visible. Set `ARIS_VERIFY_EMAIL` in your shell to lift CrossRef to the polite-pool rate. Original signal from [@YiwenZhu77](https:\u002F\u002Fgithub.com\u002FYiwenZhu77) in [#120](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F120) — landed via clean reimplementation rather than direct merge (PR was 5 weeks old + scope creep into figure-style).\n- **2026-05-06** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🎤 **[`\u002Fpaper-talk`](skills\u002Fpaper-talk\u002FSKILL.md) workflow + [`\u002Fslides-polish`](skills\u002Fslides-polish\u002FSKILL.md) skill — end-to-end conference talk pipeline**. `\u002Fpaper-talk` orchestrates paper → slide outline → Beamer + PPTX → per-page polish → assurance audits → final report (sister to `\u002Fpaper-writing`, `\u002Fpaper-poster`); composes `\u002Fpaper-slides`, `\u002Fslides-polish`, plus `\u002Fpaper-claim-audit` + `\u002Fcitation-audit` when `assurance: conference-ready`. `\u002Fslides-polish` is the post-generation visual pass: per-page Codex review against a reference PDF + a fix-pattern catalog (PPTX font scaling 1.5-1.8× for projector-readable size, text-frame resize after font bump, banner-as-tcolorbox, italic style leak guard, em-dash spacing, Chinese EA font hint via PingFang SC, anonymity placeholder discipline). Assurance ladder `draft \u002F polished (default) \u002F conference-ready` is independent from the effort axis; `effort: lite, assurance: conference-ready` is legal and means \"fast pipeline, every audit must emit verdict before final\". Phase 4 staging adapter materializes slide text + speaker notes + talk script as a synthetic paper directory (`.aris\u002Fpaper-talk\u002Faudit-input\u002Fsections\u002F*.tex` + symlinked `.bib` \u002F `results\u002F` \u002F `figures\u002F`) so the existing audits run with their paper-shaped contracts and emit 6-state JSON verdicts per `shared-references\u002Fassurance-contract.md`.\n- **2026-05-05** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🔁 **`\u002Fresubmit-pipeline` — Workflow 5: text-only resubmit across venues** ([#208](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F208)). Port a polished paper from one venue to another under hard constraints (no new experiments, no bib edits, no framework changes, never overwrite prior submissions). 5 phases: physical isolation → 5-layer anonymity check → audits (proof \u002F claim \u002F citation `--soft-only`) → microedits via `\u002Fauto-paper-improvement-loop --edit-whitelist` with per-round diff gate → adversarial gate via `\u002Fkill-argument` → final compile + Overleaf push via `\u002Foverleaf-sync`. Two prerequisite SKILL upgrades shipped in the same PR: **`\u002Fauto-paper-improvement-loop --edit-whitelist \u003Cpath>`** (YAML schema with allowed\u002Fforbidden paths + `forbidden_operations` like `new_cite` \u002F `new_theorem_env` \u002F `numerical_claim`, `forbidden_deletions`, `requires_user_approval_for`, `max_edits_per_round`) and **`\u002Fcitation-audit --soft-only`** (translates KEEP\u002FFIX\u002FREPLACE\u002FREMOVE verdicts to text-rewrite proposals when bib is frozen; hallucinated citations get `drop_cite_in_body_only` action). Master `RESUBMIT_REPORT.json` ledger per `shared-references\u002Fassurance-contract.md`; 7-verdict failure mode table including `USER_DECISION` runtime state.\n- **2026-05-05** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🗡 **`\u002Fkill-argument` — adversarial Attack-Adjudication review for theory papers** ([#206](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F206)). Two fresh codex 5.5 + xhigh threads: Thread 1 writes the strongest 200-word rejection memo a senior area chair would produce; Thread 2 (independent adjudicator, NOT defender) reads the current paper and classifies each rejection point as `answered_by_current_text` \u002F `partially_answered` \u002F `still_unresolved` with file:line evidence. Output: `KILL_ARGUMENT.{md,json}`, detect-only. Integrated as **Phase 5.6** of `\u002Fpaper-writing` (between claim-audit and citation-audit) and as the canonical implementation called from `\u002Fauto-paper-improvement-loop` Step 5.5 — replaces inline prompt in both places. Mandatory at `assurance: submission` for theory-heavy \u002F scope-heavy papers; emits `NOT_APPLICABLE` for empirical papers without scope claims. Audit JSON is `verify_paper_audits.sh`-compatible (full schema per `shared-references\u002Fassurance-contract.md`, 6-state verdict). Catches the failure mode score-based reviews miss: when every local component is correct (numbers match, cites resolve, theorems prove) but the paper still oversells what it actually establishes.\n- **2026-05-04** — ![FIX](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FFIX-orange?style=flat-square) 🪲 **`\u002Fresearch-wiki` and 8 caller skills now resolve helper via fallback chain** ([#204](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F204)). Bug: after `bash tools\u002Finstall_aris.sh` the helper lives at `.aris\u002Ftools\u002Fresearch_wiki.py` (symlink), but skills hard-coded `tools\u002Fresearch_wiki.py` and silently failed when invoked — `research-wiki\u002F` stayed empty across full W1 runs. Fix: 3-layer chain (`.aris\u002Ftools\u002F` → `tools\u002F` → `$ARIS_REPO\u002Ftools\u002F`) codified in [`shared-references\u002Fwiki-helper-resolution.md`](skills\u002Fshared-references\u002Fwiki-helper-resolution.md). The manual-copy workaround at `\u003Cproject>\u002Ftools\u002Fresearch_wiki.py` is layer 2, so users who `cp`-installed the helper as a temporary fix continue to work. **⚠️ Existing users**: rerun `bash tools\u002Finstall_aris.sh` once — also picks up a separate Python 3.9 `ImportError` fix in the helper.\n- **2026-05-03** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🎨 **Opt-in `— style-ref: \u003Csource>` for writer-side skills** ([#202](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F202)). `\u002Fpaper-{plan,write,writing,illustration,poster,slides}`, `\u002Fgrant-proposal`, and `\u002Fauto-paper-improvement-loop` accept an optional `— style-ref: \u003Csource>` argument that mimics a reference paper's *structural* style (section ordering, theorem\u002Ffigure density, sentence cadence, citation style) **without copying its prose, claims, or terminology**. Sources: local `.tex` dir\u002Ffile, local PDF, arXiv id (`2501.12345` or `arxiv:2501.12345`), HTTP\u002FHTTPS URL. Overleaf URLs\u002FIDs are rejected — clone via `\u002Foverleaf-sync setup \u003Cid>` first. **Default OFF**; existing behavior unchanged when the flag is absent. Reviewer \u002F auditor sub-skills (`\u002Fproof-checker`, `\u002Fpaper-claim-audit`, `\u002Fcitation-audit`, the improvement-loop reviewer) never see the style ref — cross-model review independence preserved. **⚠️ Existing ARIS users**: the helper ships at `tools\u002Fextract_paper_style.py`, distributed via the `.aris\u002Ftools` symlink (`install_aris.sh` Phase 0, added in [#192](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F192)). **Re-run `bash tools\u002Finstall_aris.sh` once** to refresh the symlink and pick up the helper. Manual fallback: `cp \u003CARIS-repo>\u002Ftools\u002Fextract_paper_style.py \u003Cyour-project>\u002Ftools\u002F`. Without either, the writer skill aborts with a clear error pointing here.\n- **2026-05-02** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🪨 **Community spotlight: [rosetta](https:\u002F\u002Fgithub.com\u002FSyntaxSmith\u002Frosetta)** by [@SyntaxSmith](https:\u002F\u002Fgithub.com\u002FSyntaxSmith). Programmatic access to **ChatGPT Pro \u002F `gpt-5.5-pro` \u002F DeepResearch** from Node, via Chrome CDP Fetch interception + WebSocket second-leg streaming; ships an MCP server for Claude Code \u002F Codex \u002F Cline. Alternative implementation path to Oracle MCP for ARIS users invoking `— reviewer: oracle-pro` — same target capability (Pro-tier reviewer), different mechanics. Indexed under [Awesome Community Skills & Extensions](#-awesome-community-skills--extensions). 🌟 if you're using it!\n- **2026-05-02** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 💎🧿 **Model & MCP routing updates**. (a) [`\u002Fgemini-search`](skills\u002Fgemini-search\u002FSKILL.md) default bumped to `gemini-3-pro-preview` (strongest Gemini, out-of-box). ⚠️ **Action required**: requires `gemini-cli` v0.40+ (run `gemini --version`; upgrade with `npm i -g @google\u002Fgemini-cli` if older). Legacy override: `\u002Fgemini-search \"topic\" — model: gemini-2.5-pro`. Other overrides: `gemini-3-flash-preview` (faster), `auto-gemini-3` (load-routed). (b) [`\u002Fidea-discovery`](skills\u002Fidea-discovery\u002FSKILL.md) Phase 1 now includes Gemini in its literature survey by default ([#199](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F199)) — auto-injects `— sources: all, gemini` into `\u002Fresearch-lit` unless the user passed an explicit `— sources:`; graceful skip if `gemini-cli` not installed. (c) Oracle MCP upstream PR queue ([`steipete\u002Foracle\u002Fpulls`](https:\u002F\u002Fgithub.com\u002Fsteipete\u002Foracle\u002Fpulls)) is the first triage stop when invoking `— reviewer: oracle-pro` (especially `o3-deep-research` \u002F `gpt-5.5-pro`) — ARIS does not vendor Oracle MCP; check upstream first if behavior surprises you ([reviewer-routing.md](skills\u002Fshared-references\u002Freviewer-routing.md))\n- **2026-05-02** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🛠️🔗 **Tools-infrastructure migration started**. (a) [`install_aris.sh`](tools\u002Finstall_aris.sh) creates optional `.aris\u002Ftools` symlink ([#192](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F192), closes [#174](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F174)) — Phase 0 of the 4-step tools-stability plan (#174 → #176 → #177 → #178); idempotent, **zero impact until rerun**. (b) [`\u002Fexperiment-queue`](skills\u002Fexperiment-queue\u002FSKILL.md) orchestration paths repaired ([#193](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F193)) — first real user of the symlink; 7 cascading bugs fixed via 3 rounds of Codex MCP `gpt-5.5` xhigh audit. Pure prose + docstring; `queue_manager.py` logic untouched. Windows `install_aris.ps1` parallel update tracked as follow-up\n- **2026-05-02** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🔬 **Three new opt-in audit flags via fast-path delegated-agent workflow** ([#187](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F187), [#188](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F188), [#189](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F189)). [`\u002Fcitation-audit --uncited`](skills\u002Fcitation-audit\u002FSKILL.md) surfaces bib entries with no `\\cite{}` reference (detect-only). [`\u002Fproof-checker --deep-fix`](skills\u002Fproof-checker\u002FSKILL.md) adds a repair-grade plan to the Phase 1 reviewer prompt (corrected statement \u002F patch plan \u002F closure tests + Schur\u002Fquadratic-form algebra sanity). [`\u002Fproof-checker --restatement-check`](skills\u002Fproof-checker\u002FSKILL.md) adds Phase 3.6 cross-location theorem drift detection (6 drift signatures). **Zero behavior change** when flags unset. Plus doc PRs [#190](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F190) (thread-policy) + [#191](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F191) (auto-loop xref). Delegated-agent + maintainer-fixup pattern; Codex MCP `gpt-5.5` xhigh review caught 6+ blockers\n- **2026-05-01** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🔍 **Gemini + OpenAlex literature sources for `\u002Fresearch-lit`** ([#175](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F175), community contribution by [@stdAri](https:\u002F\u002Fgithub.com\u002FstdAri)). Two opt-in sources: [`\u002Fgemini-search`](skills\u002Fgemini-search\u002FSKILL.md) (AI-driven discovery via [`jamubc\u002Fgemini-mcp-tool`](https:\u002F\u002Fgithub.com\u002Fjamubc\u002Fgemini-mcp-tool) MCP) and [`\u002Fopenalex`](skills\u002Fopenalex\u002FSKILL.md) (250M+ work open citation graph, no API key). Triggered via `— sources: gemini` or `— sources: openalex`; **zero behavior change** when default `all` (both excluded). Maintainer fixups: corrected `@google\u002Fgemini-cli` npm name; added `try\u002Fexcept ImportError` + bash preflight for graceful OpenAlex skip when `requests` missing\n- **2026-04-30** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 📝 **`\u002Frebuttal` per-reviewer thread mode + transferable patterns** ([SKILL.md](skills\u002Frebuttal\u002FSKILL.md)). Adds `VENUE_MODE` (`single_document` | `per_reviewer_thread`) for OpenReview-style venues, `reviewer_priority: pivotal` routing, `structural_distinction` response mode, 5 reviewer-defensive heuristics, 2 Phase 5 lints, and severity-scaled stress rounds. Default `VENUE_MODE = single_document` keeps ICML-style behavior — **zero change for existing users**. Three rounds of cross-model review before\u002Fafter merge\n- **2026-04-30** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🪞 **Codex skill mirror rebuilt + dedicated install\u002Fupdate chain** ([#179](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F179), community contribution by [@No-518](https:\u002F\u002Fgithub.com\u002FNo-518)). `skills\u002Fskills-codex\u002F` now mirrors all 67 mainline skills; replaces `mcp__codex__codex` reviewer path with Codex-native `spawn_agent` + `send_input`. New [`tools\u002Finstall_aris_codex.sh`](tools\u002Finstall_aris_codex.sh) + [`tools\u002Fsmart_update_codex.sh`](tools\u002Fsmart_update_codex.sh) handle project-local symlinks with manifest tracking. Anti-drift: `tests\u002Ftest_codex_skill_mirror.py` + `tests\u002Ftest_codex_install_update.py` (26 failure paths). Open discussion in [#184](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F184)\n- **2026-04-24** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🎨 **[`\u002Fpaper-illustration-image2`](skills\u002Fpaper-illustration-image2\u002FSKILL.md)** — Codex-native image generation as Phase 2b illustration backend ([#166](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fpull\u002F166), community contribution by [@kbr19-thu](https:\u002F\u002Fgithub.com\u002Fkbr19-thu) 清华). Uses ChatGPT Plus\u002FPro quota via local [Codex app-server MCP bridge](mcp-servers\u002Fcodex-image2\u002F) — **no `GEMINI_API_KEY` required**. Triggered by `\u002Fpaper-writing — illustration: codex-image2`; default stays `figurespec` (**zero behavior change**). Async-only API, sandboxed writes to `figures\u002Fai_generated\u002F`, [integration-contract](skills\u002Fshared-references\u002Fintegration-contract.md)-compliant helper. Marked **experimental** (Codex debug app-server is unstable upstream)\n- **2026-04-21** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 📚 **Research Wiki ingest actually works now** ([`research_wiki.py`](tools\u002Fresearch_wiki.py), [`\u002Fresearch-wiki`](skills\u002Fresearch-wiki\u002FSKILL.md)). Fixes user-reported bug where `\u002Fresearch-wiki init` left `papers\u002F` empty forever (`ingest` subcommand had no implementation; paper-reading skills had no wiki hook). New canonical `python3 tools\u002Fresearch_wiki.py ingest_paper` helper owns slugging \u002F metadata fetch \u002F dedup \u002F page render; all 6 paper-reading skills wired to it. Manual backfill via `sync --arxiv-ids` or `sync --from-file`. Ships with [`integration-contract.md`](skills\u002Fshared-references\u002Fintegration-contract.md) formalizing the six-component pattern every cross-skill integration must follow\n- **2026-04-21** — ![NEW](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FNEW-red?style=flat-square) 🛡️ **Assurance Gate: `— effort: beast | max` now really runs mandatory audits** ([`assurance-contract.md`](skills\u002Fshared-references\u002Fassurance-contract.md), [`tools\u002Fverify_paper_audits.sh`](tools\u002Fverify_paper_audits.sh)). Fixes silent-skip of `\u002Fproof-checker` \u002F `\u002Fpaper-claim-audit` \u002F `\u002Fcitation-audit` at high effort. New `assurance` axis (`draft` | `submission`) independent from `effort`: `lite` \u002F `balanced` → `draft` (**zero behavior change**), `max` \u002F `beast` → `submission`. At submission the 3 audits emit a JSON artifact with 6-state verdict; `paper-writing` Phase 6 runs the external verifier as source of truth (non-zero exit blocks Final Report). SHA256 input hashing catches stale audits. Escape hatch: `— effort: beast, assurance: draft`\n- **2026-04-20** — 🩹 **Project install: flat layout + manifest tracking** — fixes a real bug where the previous nested install (`.claude\u002Fskills\u002Faris\u002F`) hid skills from Claude Code's slash-command discovery (CC only scans one directory level). Anyone who ran `install_aris.sh` before this date was silently affected. New `install_aris.sh` creates one symlink per skill at `.claude\u002Fskills\u002F\u003Cname>`, writes a versioned manifest to `.aris\u002Finstalled-skills.txt`, and is **re-runnable to reconcile** new\u002Fremoved upstream skills. Defense-in-depth: 13 safety rules (no-symlinked-parents, exact-target revalidation, slug regex, atomic same-dir manifest rename, no-overwrite-real-files, mkdir-based portable lock, ADOPT for crash recovery, …). Granular `--adopt-existing` \u002F `--replace-link` flags replace the all-or-nothing `--force`. Migration paths: `--from-old` for legacy nested symlink, `--migrate-copy keep-user|prefer-upstream` for legacy nested copy. `smart_update.sh --target-subdir .claude\u002Fskills\u002Faris` is now deprecated with a redirect to `install_aris.sh`. Stale-file bug in `cp -r` overlay also fixed (now `rm -rf && cp -r` for safe-update path)\n- **2026-04-19** — 🔗 **[`\u002Foverleaf-sync`](skills\u002Foverleaf-sync\u002FSKILL.md)** — two-way bridge between local ARIS paper directory and an Overleaf project via the official **Overleaf Git bridge** (Premium). Lets collaborators keep editing in the Overleaf web UI while ARIS audit\u002Fedit pipelines (`\u002Fpaper-claim-audit`, `\u002Fcitation-audit`, `\u002Fauto-paper-improvement-loop`) keep running locally. Sub-commands: `setup` (one-time, user-driven so the agent never sees the token) \u002F `pull` (with diff-protocol — flags half-sentences, typos, claim\u002Fcite changes that should re-trigger audits) \u002F `push` (with confirmation gate before writing to shared Overleaf state) \u002F `status` (3-way divergence check). **Token never touches the agent or any file** — primed once into macOS Keychain via the user's terminal, then auth-free for all subsequent agent operations\n- **2026-04-19** — 📚 **[`\u002Fcitation-audit`](skills\u002Fcitation-audit\u002FSKILL.md)** — fourth and final layer of the evidence-and-claim assurance stack (`experiment-audit` → `result-to-claim` → `paper-claim-audit` → `citation-audit`). Fresh cross-family reviewer (gpt-5.4 via Codex MCP) with web\u002FDBLP\u002FarXiv lookup verifies every `\\cite{...}` along three independent axes: **existence** (paper resolves at claimed arXiv ID\u002FDOI\u002Fvenue), **metadata correctness** (authors\u002Fyear\u002Fvenue\u002Ftitle match canonical sources), and **context appropriateness** (the cited paper actually establishes the claim it supports — the most diagnostic check). Per-entry verdicts: KEEP \u002F FIX \u002F REPLACE \u002F REMOVE. Auto-integrated into **Workflow 3 Phase 5.8** as the pre-submission bibliography gate. Empirical motivation: in a real submission run, several real papers were cited in contexts they did not actually support, and at least one entry shipped with `author = \"Anonymous\"` — none caught by metadata-only checks\n\n- **2026-04-17** — 🔀 **`\u002Fexperiment-queue` integrated into Workflow 1.5 + research-pipeline** — `experiment-bridge` Phase 4 Deploy now auto-routes by milestone job count: ≤5 jobs → `\u002Frun-experiment`, ≥10 jobs or phase dependencies → `\u002Fexperiment-queue` (with OOM retry, stale-screen cleanup, wave-transition gating, crash-safe state). New `--- batch: queue` override for global force-queue mode. Large multi-seed sweeps from `EXPERIMENT_PLAN.md` (e.g., 36-cell `N × seed × n_train` grids) now get proper orchestration without manual queue invocation\n- **2026-04-17** — 🔗 **[Project-local symlink install](tools\u002Finstall_aris.sh)** (resolves [#118](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F118)) — new recommended default install. `bash tools\u002Finstall_aris.sh` auto-detects platform (Claude Code \u002F Codex CLI), creates `.claude\u002Fskills\u002Faris` or `.agents\u002Fskills\u002Faris` symlink to the ARIS repo, adds a managed `\u003C!-- ARIS:BEGIN -->` block to `CLAUDE.md` \u002F `AGENTS.md` telling the agent to use only project-local skills, and records install metadata in `.aris\u002Fskill-source.txt`. **Solves the skill collision problem** when ARIS is mixed with Superpowers \u002F OpenHands \u002F other community packs in the same global skill directory. PowerShell version (`install_aris.ps1`) ships with junction support for Windows. **`smart_update.sh --target-subdir`** flag added for `.agents\u002Fskills\u002Faris` (Codex) project-copy installs; symlinked installs now correctly refuse `smart_update` and direct users to `git pull`. Global install remains supported for power users\n- **2026-04-16** — 🎨 **[`\u002Ffigure-spec`](skills\u002Ffigure-spec\u002FSKILL.md)** — deterministic JSON→SVG renderer packaged as a first-class skill. Preferred default for architecture\u002Fworkflow\u002Fpipeline\u002Faudit-cascade figures in papers. Shape-aware edge clipping (rect\u002Fcircle\u002Fellipse\u002Fdiamond), self-loops, curved edges, multi-line labels with CJK width estimation. Editable vector output, reproducible (same spec → same SVG), no external API. **Phase 2b in Workflow 3 restored**: `illustration: figurespec` (new default) \u002F `gemini` \u002F `mermaid` \u002F `false` — 4-way illustration selector with complementary strengths\n- **2026-04-16** — ⚙️ **[`\u002Fexperiment-queue`](skills\u002Fexperiment-queue\u002FSKILL.md)** — SSH job queue for multi-seed\u002Fmulti-config ML experiments. Designed from real 36-cell NeurIPS sweep pain points: OOM-aware retry with backoff, stale-screen cleanup, wave-transition race prevention, teacher→student phase dependencies, crash-safe scheduler that resumes from JSON state. Declarative grid specs expand automatically (e.g., `N × seed × n_train → 36 jobs`). Configurable `conda_hook` + `gpu_free_threshold_mib` for non-standard environments. Use for ≥10 jobs; `\u002Frun-experiment` stays for ad-hoc\n- **2026-04-15** — 🛡️ **Paper Writing Pipeline Hardening** — 10 empirically-motivated patches from a real NeurIPS run. `REVIEWER_BIAS_GUARD=true`: every review round uses a fresh thread (codex-reply inflated 3→8\u002F10). Reviewer Independence Protocol: no fix summaries to reviewer. Step 4.5 Restatement Regression Test: catches theorem drift across fix rounds. Step 5.5 Kill Argument Exercise: final-round adversarial attack\u002Fdefense for theory papers. Location-aware overfull blocking. Theory Paper Consistency Pass in `\u002Fpaper-write`. Enforced Bib Hygiene with DBLP\u002FCrossRef validation. Phase 5.5 Mandatory Final Claim Audit as submission gate. **Review Tracing Protocol**: full prompt\u002Fresponse pairs saved to `.aris\u002Ftraces\u002F` for reviewer-independence audit ([`review-tracing.md`](skills\u002Fshared-references\u002Freview-tracing.md), [`save_trace.sh`](tools\u002Fsave_trace.sh)). Inspired by community contribution from @李傲龍\n- **2026-04-15** — 🎨 **[FigureSpec Renderer v2](tools\u002Ffigure_renderer.py)** — deterministic JSON→SVG figure generation for academic papers. Shape-aware edge clipping (rect\u002Fcircle\u002Fellipse\u002Fdiamond), self-loops, curved edges, multi-line labels with CJK width estimation, comprehensive validation (type checks, structure, palette). Went through 5 rounds of Codex review (3\u002F10→7\u002F10). All architecture and workflow diagrams in the ARIS tech report were generated with this pipeline. New `--- mode: vector` for `\u002Fpaper-illustration` skill\n- **2026-04-14** — 📋 **[`\u002Fpaper-claim-audit`](skills\u002Fpaper-claim-audit\u002FSKILL.md)** — zero-context paper-to-evidence verification. Fresh reviewer with NO prior context compares every number in the paper against raw result files. Catches rounding inflation, best-seed cherry-pick, config mismatch, delta errors, scope overclaim. Auto-integrated into Workflow 3 (Phase 4.7). Completes the 3-layer audit chain: `\u002Fexperiment-audit` (code) → `\u002Fresult-to-claim` (science) → `\u002Fpaper-claim-audit` (reporting). 👁️ **Visual PDF review** also added to improvement loop — reviewer now sees compiled PDF, not just LaTeX source. Inspired by [Hermes Agent](https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent\u002Ftree\u002Fmain\u002Fskills\u002Fresearch\u002Fresearch-paper-writing)\n- **2026-04-13** — 🧿 **[GPT-5.4 Pro via Oracle](skills\u002Fshared-references\u002Freviewer-routing.md)** — `— reviewer: oracle-pro` on any skill for the strongest available reviewer. API mode (fast) or browser mode (free). Supported on: `\u002Fresearch-review`, `\u002Fauto-review-loop`, `\u002Fexperiment-audit`, `\u002Fproof-checker`, `\u002Frebuttal`, `\u002Fidea-creator`, `\u002Fresearch-lit`. Default stays Codex xhigh. Not installed = zero impact. [Setup →](#-optional-gpt-54-pro-via-oracle)\n- **2026-04-13** — 🔬 **[`\u002Fproof-checker`](skills\u002Fproof-checker\u002FSKILL.md)** — rigorous mathematical proof verification via cross-model review. 20-category issue taxonomy, two-axis severity, side-condition checklists (DCT\u002FMCT\u002FFubini\u002FIFT\u002F...), counterexample red team, proof-obligation ledger. Auto-integrated into Workflow 3: detects `\\begin{theorem}` and runs before improvement loop. Complements `\u002Fproof-writer`\n- **2026-04-10** — ⚡ **[Effort Levels](skills\u002Fshared-references\u002Feffort-contract.md)** — `— effort: lite | balanced | max | beast`. Controls work intensity across all skills: papers found, ideas generated, review rounds, writing depth. Codex reasoning stays `xhigh` always. `beast` = every knob to maximum for top-venue sprints. Default `balanced` = zero change for existing users. [Details →](#-effort-levels)\n- **2026-04-10** — 🔎 **[DeepXiv integration](skills\u002Fdeepxiv\u002FSKILL.md)** — progressive paper retrieval via DeepXiv CLI. Opt-in: `— sources: deepxiv` or `— sources: all, deepxiv`. Staged reading: search → brief → head → section. `pip install deepxiv-sdk` to enable. Community contribution by [@DreamEnding](https:\u002F\u002Fgithub.com\u002FDreamEnding)\n- **2026-04-10** — 🛡️ **[`\u002Fexperiment-audit`](skills\u002Fexperiment-audit\u002FSKILL.md)** — cross-model experiment integrity verification. GPT-5.4 reads your eval scripts and results directly, checks for fake ground truth, self-normalized scores, phantom results, and scope inflation ([#131](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F131), [#57](https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep\u002Fissues\u002F57)). Advisory — warns loudly, never blocks. `\u002Fresult-to-claim` auto-reads audit if present. New [experiment-integrity.md](skills\u002Fshared-references\u002Fexperiment-integrity.md) shared reference. **The executor must never judge its own integrity.**\n- **2026-04-10** — 🧠 **[`tools\u002Fsmart_update.sh`](tools\u002Fsmart_update.sh)** — intelligent skill updater. Compares local vs upstream, detects personal customizations (server paths, API keys), only updates safe skills. `bash tools\u002Fsmart_update.sh --apply`\n- **2026-04-10** — 🏆 **Community paper: [UAV-CC](community_papers\u002FUAV-CC.pdf)** — first community paper with full PDF archived. UAV change captioning benchmark for IEEE TGRS by [@wxx827](https:\u002F\u002Fgithub.com\u002Fwxx827). Stack: Claude Opus 4.6 + Codex 5.4 xhigh + Cursor. Papers now archived in `community_papers\u002F`\n- **2026-04-08** — 📚 **[`\u002Fresearch-wiki`](skills\u002Fresearch-wiki\u002FSKILL.md)** — persistent research knowledge base inspired by [Karpathy's LLM Wiki](https:\u002F\u002Fgist.github.com\u002Fkarpathy\u002F442a6bf555914893e9891c11519de94f). Accumulates papers, ideas, experiments, and claims across the entire research lifecycle with typed relationships. Wiki-aware hooks in `\u002Fresearch-lit` (ingest papers), `\u002Fidea-creator` (read wiki + write ideas back), and `\u002Fresult-to-claim` (update claim status + trigger re-ideation). Failed ideas become anti-repetition memory. **ARIS now learns from its mistakes.**\n- **2026-04-05** — 🧬 **[`\u002Fmeta-optimize`](skills\u002Fmeta-optimize\u002FSKILL.md)** — outer-loop harness optimization for ARIS. Passively logs skill invocations, tool calls, failures, and parameter overrides via [Claude Code hooks](templates\u002Fclaude-hooks\u002Fmeta_logging.json). Run `\u002Fmeta-optimize` to analyze accumulated usage data and propose SKILL.md improvements — reviewer-gated, user-approved. Inspired by [Meta-Harness](https:\u002F\u002Farxiv.org\u002Fabs\u002F2603.28052) (Lee et al., 2026). **ARIS now optimizes itself.**\n- **2026-04-04** — 🔧 **Codex Plugin deep integration** — `\u002Fcodex:rescue` now auto-invoked when experiments fail (Workflow 1.5) or LaTeX won't compile (Workflow 3). GPT independently diagnoses the bug before Claude retries — two AI debuggers are better than one. Optional: `codex exec` powers nightmare review, `\u002Fcodex:rescue` powers auto-debug. [Setup →](#optional-codex-plugin-for-code-review)\n- **2026-04-03** — ☁️ **[Modal serverless GPU](skills\u002Fserverless-modal\u002FSKILL.md)** — no GPU? `gpu: modal` in CLAUDE.md, one command (`modal run launcher.py`), no SSH, no Docker, auto scale-to-zero. **$30\u002Fmonth free tier** — enough to try ARIS experiments without any hardware. `pip install modal && modal setup` and go. Community contribution by [@zeyuzhangzyz](https:\u002F\u002Fgithub.com\u002Fzeyuzhangzyz)\n- **2026-04-03** — 🎮 **Reviewer Difficulty Levels** — `medium` (default, unchanged), `hard` (reviewer memory + debate protocol), `nightmare` (GPT reads repo directly via `codex exec` — Claude can't hide anything). `— difficulty: nightmare` for maximum stress test before submission\n\n- **2026-03-30** — 🔥 **Auto-debug & exhaust-before-surrender** — experiment-bridge auto-diagnoses failures (OOM, import, CUDA, NaN) and retries up to 3×. Inspired by [PUA](https:\u002F\u002Fgithub.com\u002Ftanweai\u002Fpua)\n- **2026-03-30** — ☁️ **[Vast.ai GPU rental](skills\u002Fvast-gpu\u002FSKILL.md)** — `gpu: vast` auto-rents cheapest GPU. By [@YIHONG-JIN](https:\u002F\u002Fgithub.com\u002FYIHONG-JIN). 🔧 MiniMax M2.7 upgrade by [@octo-patch](https:\u002F\u002Fgithub.com\u002Focto-patch)\n- **2026-03-27** — 📄 **IEEE venue support** (9 families). 🔎 **[Semantic Scholar](skills\u002Fsemantic-scholar\u002FSKILL.md)**. By [@ypd666](https:\u002F\u002Fgithub.com\u002Fypd666)\n- **2026-03-26** — 📄 **Document-based input** — `RESEARCH_BRIEF.md` auto-detect\n- **2026-03-24** — 📝 **[Workflow 4: `\u002Frebuttal`](skills\u002Frebuttal\u002FSKILL.md)** — 7-phase pipeline, 3 safety gates\n- **2026-03-23** — 🔧 `\u002Ftraining-check`, `\u002Fresult-to-claim`, `\u002Fablation-planner` integrated. 📦 `compact` mode. By [@JingxuanKang](https:\u002F\u002Fgithub.com\u002FJingxuanKang) & [@couragec](https:\u002F\u002Fgithub.com\u002Fcouragec)\n\n- **2026-03-22** — 📋 **[Templates](templates\u002F)** — input templates for every workflow. 📄 **7 venue templates** — CVPR, ACL, AAAI, ACM MM added. 🛡️ **Anti-hallucination fix** — Workflow 2 enforces DBLP → CrossRef → [VERIFY]. 🔗 **`base repo`** — clone a GitHub repo as base codebase (`— base repo: https:\u002F\u002Fgithub.com\u002Forg\u002Fproject`)\n- **2026-03-22** — 🔍 **[Codex + Gemini review guide](docs\u002FCODEX_GEMINI_REVIEW_GUIDE.md)** — Codex executes, Gemini reviews via local `gemini-review` MCP bridge. [CN](docs\u002FCODEX_GEMINI_REVIEW_GUIDE_CN.md)\n- **2026-03-20** — 🚀 **[Antigravity adaptation guide](docs\u002FANTIGRAVITY_ADAPTATION.md)** — use ARIS skills in [Google Antigravity](https:\u002F\u002Fantigravity.google\u002F) (agent-first IDE). Community contribution by [@PeppaPigw](https:\u002F\u002Fgithub.com\u002FPeppaPigw)\n- **2026-03-20** — 🖥️ **[Trae adaptation guide](docs\u002FTRAE_ARIS_RUNBOOK_EN.md)** — use ARIS skills in [Trae](https:\u002F\u002Fwww.trae.ai\u002F) (ByteDance AI IDE). Community contribution by [@Prometheus-cotigo](https:\u002F\u002Fgithub.com\u002FPrometheus-cotigo). 🔢 **[`formula-derivation`](skills\u002Fformula-derivation\u002FSKILL.md)** — Community contribution by [@Falling-Flower](https:\u002F\u002Fgithub.com\u002FFalling-Flower)\n- **2026-03-19** — 🖼️ **[`paper-poster`](skills\u002Fpaper-poster\u002FSKILL.md)** — Conference poster. Community contribution by [@dengzhe-hou](https:\u002F\u002Fgithub.com\u002Fdengzhe-hou)\n- **2026-03-19** — 🔗 **Workflow 1.5 upgraded** — `\u002Fexperiment-bridge` GPT-5.4 code review. 📊 **W&B fix**\n- **2026-03-18** — 🎤 `paper-slides` + 🔁 Codex+Claude bridge + 🖱️ Cursor guide + 🤖 Codex CLI skills + 📝 `grant-proposal` + 🎨 `paper-illustration` (Gemini) + 📊 CitationClaw\n- **2026-03-17** — 🔧 Git code sync + 🆓 ModelScope guide + parameter pass-through\n\n- **2026-03-16** — 🔬 **[`research-refine`](skills\u002Fresearch-refine\u002FSKILL.md)** + [`experiment-plan`](skills\u002Fexperiment-plan\u002FSKILL.md) — turn vague ideas into problem-anchored proposals with claim-driven experiment roadmaps. Now integrated into Workflow 1 (`\u002Fidea-discovery`). Community contribution by [@zjYao36](https:\u002F\u002Fgithub.com\u002FzjYao36)\n- **2026-03-16** — 🇨🇳 **[Alibaba Coding Plan guide](docs\u002FALI_CODING_PLAN_GUIDE.md)** — one API key, 4 models (Kimi-K2.5 + Qwen3.5+ + GLM-5 + MiniMax-M2.7), dual-endpoint setup. Community contribution by [@tianhao909](https:\u002F\u002Fgithub.com\u002Ftianhao909)\n- **2026-03-15** — 🔀 **Bring your own model!** [Any OpenAI-compatible API](#-alternative-model-combinations) now works as reviewer via [`llm-chat`](mcp-servers\u002Fllm-chat\u002F) MCP server. GLM, MiniMax, Kimi, LongCat, DeepSeek all tested — **zero Claude or OpenAI API needed**\n- **2026-03-15** — 🐾 **[OpenClaw adaptation guide](docs\u002FOPENCLAW_ADAPTATION.md)** — use ARIS research workflows in [OpenClaw](https:\u002F\u002Fgithub.com\u002FAll-Hands-AI\u002FOpenHands) without Claude Code slash skills\n- **2026-03-15** — 📐 **[`proof-writer`](skills\u002Fproof-writer\u002FSKILL.md)** — community skill for rigorous theorem proof drafting. 📚 **Anti-hallucination citations** — `\u002Fpaper-write` now fetches real BibTeX from [DBLP](https:\u002F\u002Fdblp.org)\u002F[CrossRef](https:\u002F\u002Fwww.crossref.org) instead of LLM-generated entries — on by default, zero install\n- **2026-03-14** — 📱 [Feishu\u002FLark integration](docs\u002Fintegrations\u002FFEISHU.md): three modes (off\u002Fpush\u002Finteractive), mobile notifications for experiments, reviews, and checkpoints\n- **2026-03-13** — 🛑 Human-in-the-loop: configurable `AUTO_PROCEED` checkpoints across all workflows. Full autopilot or step-by-step approval\n- **2026-03-12** — 🔗 [Zotero](docs\u002Fintegrations\u002FZOTERO.md) + [Obsidian](docs\u002Fintegrations\u002FOBSIDIAN.md) + local PDFs + arXiv\u002FScholar: multi-source literature search with cross-model novelty verification\n- **2026-03-12** — 🚀 Three end-to-end workflows complete: one prompt → top-venue-style paper. `\u002Fresearch-pipeline` chains idea discovery → auto review → paper writing autonomously\n- **2026-03-12** — 📝 `\u002Fpaper-writing` workflow: narrative report → structured outline → figures → LaTeX → compiled PDF → 2-round auto-improvement (4\u002F10 → 8.5\u002F10)\n\n\u003C\u002Fdetails>\n\n## 🚀 Quick Start\n\n```bash\n# 1. Install skills\ngit clone https:\u002F\u002Fgithub.com\u002Fwanshuiyin\u002FAuto-claude-code-research-in-sleep.git\nmkdir -p ~\u002F.claude\u002Fskills\u002F    # create if it doesn't exist (new Claude Code versions)\ncp -r Auto-claude-code-research-in-sleep\u002Fskills\u002F* ~\u002F.claude\u002Fskills\u002F\n\n# 1b. Update skills (when upstream has new versions)\ncd Auto-claude-code-research-in-sleep && git pull\nbash tools\u002Fsmart_update.sh          # dry-run: shows what's new\u002Fchanged\u002Fsafe\nbash tools\u002Fsmart_update.sh --apply  # apply: adds new + updates safe ones\n\n# Optional Codex mirror managed project install\nbash tools\u002Finstall_aris_codex.sh ~\u002Fyour-codex-project\n\n# Managed Codex project update\ncd Auto-claude-code-research-in-sleep && git pull\nbash tools\u002Finstall_aris_codex.sh ~\u002Fyour-codex-project --reconcile\n\n# Copied Codex installs only (not for projects installed by install_aris_codex.sh)\nbash tools\u002Fsmart_update_codex.sh --local ~\u002F.codex\u002Fskills\nbash tools\u002Fsmart_update_codex.sh --local ~\u002F.codex\u002Fskills --apply\n\n# 2. Set up Codex MCP (for review skills)\nnpm install -g @openai\u002Fcodex\ncodex setup                    # set model to gpt-5.5 when prompted\nclaude mcp add codex -s user -- codex mcp-server\n\n# 3. Use in Claude Code\nclaude\n> \u002Fidea-discovery \"your research direction\"  # Workflow 1 — be specific! not \"NLP\" but \"factorized gap in discrete diffusion LMs\"\n> \u002Fexperiment-bridge                         # Workflow 1.5 — have a plan? implement + deploy + collect results\n> \u002Fauto-review-loop \"your paper topic or scope\"  # Workflow 2: review → fix → re-review overnight\n> \u002Fpaper-writing \"NARRATIVE_REPORT.md\"       # Workflow 3: narrative → polished PDF\n> \u002Frebuttal \"paper\u002F + reviews\" — venue: ICML    # Workflow 4: parse reviews → draft rebuttal → follow-up\n> \u002Fresubmit-pipeline \"paper\u002F\" — venue: NeurIPS  # Workflow 5: port a polished paper to a new venue (text-only, no new experiments)\n> \u002Fpaper-talk \"paper\u002F\" — venue: ICLR            # Workflow 6: paper → Beamer + PPTX talk + speaker notes + assurance audits\n> \u002Fresearch-pipeline \"your research direction\"  # Full pipeline: Workflow 1 → 1.5 → 2 → 3 end-to-end\n> \u002Fresearch-wiki init                           # 📚 Enable persistent research memory (one-time)\n> \u002Fmeta-optimize                                # Meta: analyze usage logs → propose skill improvements\n```\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>📚 Research Wiki (optional)\u003C\u002Fb> — one-line init for persistent memory across sessions; see \u003Ca href=\"#-research-wiki--persistent-research-memory\">full Research Wiki section\u003C\u002Fa>\u003C\u002Fsummary>\n\nGive ARIS persistent memory across sessions. Papers, ideas, failed experiments — nothing is forgotten:\n\n```bash\n# In Claude Code:\n> \u002Fresearch-wiki init                         # creates research-wiki\u002F in your project\n# That's it. From now on, \u002Fresearch-lit auto-ingests papers, \u002Fidea-creator reads\n# the wiki before brainstorming (and writes ideas back), \u002Fresult-to-claim updates\n# claim status. Failed ideas become anti-repetition memory for future ideation.\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>🧬 Meta-Optimization (optional)\u003C\u002Fb> — passive usage logging + \u002Fmeta-optimize for data-driven SKILL.md improvements; see \u003Ca href=\"#workflow-m-meta-optimize--aris-optimizes-itself\">full Workflow M section\u003C\u002Fa>\u003C\u002Fsummary>\n\nRun these in your **normal terminal** (not inside Claude Code) to enable passive usage logging:\n\n```bash\n# One-time setup in your project directory\nmkdir -p .claude .aris\u002Fmeta tools\u002Fmeta_opt\ncp Auto-claude-code-research-in-sleep\u002Ftemplates\u002Fclaude-hooks\u002Fmeta_logging.json .claude\u002Fsettings.json\ncp Auto-claude-code-research-in-sleep\u002Ftools\u002Fmeta_opt\u002F*.sh tools\u002Fmeta_opt\u002F\nchmod +x tools\u002Fmeta_opt\u002F*.sh\n# Then start Claude Code — hooks are active immediately\nclaude\n```\n\nEvents are logged to **both** project-level (`.aris\u002Fmeta\u002Fevents.jsonl`) and global (`~\u002F.aris\u002Fmeta\u002Fevents.jsonl`) logs. After 5+ workflow runs, run `\u002Fmeta-optimize` to see data-driven improvement proposals. Use `\u002Fmeta-optimize --global` to analyze trends across all your projects.\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>📝 Templates + 🔎 DeepXiv + 🔎 Exa + 🗑️ Uninstall\u003C\u002Fb> — input templates, two extra literature sources, and the uninstall command\u003C\u002Fsummary>\n\n**📝 Templates available!** See [`templates\u002F`](templates\u002F) for ready-to-use input templates for every workflow — [research brief](templates\u002FRESEARCH_BRIEF_TEMPLATE.md) (Workflow 1), [experiment plan](templates\u002FEXPERIMENT_PLAN_TEMPLATE.md) (Workflow 1.5), [narrative report](templates\u002FNARRATIVE_REPORT_TEMPLATE.md) (Workflow 3), [paper plan](templates\u002FPAPER_PLAN_TEMPLATE.md) (Workflow 3).\n\n**🔎 Optional: DeepXiv progressive retrieval**\n```bash\npip install deepxiv-sdk\n```\nThen use [`\u002Fdeepxiv`](skills\u002Fdeepxiv\u002FSKILL.md) directly or opt into it from `\u002Fresearch-lit` with `— sources: deepxiv` or `— sources: all, deepxiv`.\n\n**🔎 Optional: Exa AI-powered web search**\n```bash\npip install exa-py\nexport EXA_API_KEY=your-key-here\n```\nThen use [`\u002Fexa-search`](skills\u002Fexa-search\u002FSKILL.md) directly or opt into it from `\u002Fresearch-lit` with `— sources: exa` or `— sources: all, exa`. Covers blogs, docs, news, and research papers with built-in content extraction.\n\n**🗑️ Uninstall:** To remove ARIS skills without affecting your own personal skills:\n```bash\ncd Auto-claude-code-research-in-sleep && ls skills\u002F | xargs -I{} rm -rf ~\u002F.claude\u002Fskills\u002F{}\n```\n\n\u003C\u002Fdetails>\n\n\u003Cdetails>\n\u003Csummary>\u003Cb>Show all 16 inline parameters and 12 override examples\u003C\u002Fb> — AUTO_PROCEED \u002F sources \u002F arxiv download \u002F DBLP_BIBTEX \u002F code review \u002F wandb \u002F illustration \u002F venue \u002F base repo \u002F gpu \u002F compact \u002F ref paper \u002F effort \u002F reviewer \u002F difficulty (full per-skill defaults live in \u003Ca href=\"#%EF%B8%8F-customization\">§ Customization\u003C\u002Fa>)\u003C\u002Fsummary>\n\nAll pipeline behaviors are configurable via inline overrides — append `— key: value` to any command:\n\n| Parameter | Default | What it does |\n|-----------|---------|-------------|\n| `AUTO_PROCEED` | `true` | Auto-continue at idea selection gate. Set `false` to manually pick which idea to pursue before committing GPU time |\n| `human checkpoint` | `false` | Pause after each review round so you can read the score, give custom modification instructions, skip specific fixes, or stop early |\n| `sources` | `all` | Which li","ARIS 是一个专为自动化机器学习研究设计的轻量级工具，支持跨模型审查循环、想法发现及实验自动化等功能。项目基于Markdown格式构建，不依赖特定框架，能够与Claude Code、Codex、OpenClaw等大语言模型无缝集成。其核心功能包括提供一套完整的技能集来辅助从论文审阅到实验管理的全流程自动化，并且支持多语言模型提供商如Qwen 3.6、DeepSeek V4 Pro等。适合需要提高研究效率或希望探索利用AI助手进行自主科研工作的研究人员使用。",2,"2026-06-11 03:32:45","trending"]