{"id":51492317,"url":"https://github.com/beita6969/scienceclaw","last_synced_at":"2026-07-26T08:01:30.531Z","repository":{"id":343894761,"uuid":"1178834224","full_name":"beita6969/ScienceClaw","owner":"beita6969","description":"🔬🦞 A self-evolving AI research colleague for scientists. 285 skills, zero hallucination, persistent memory.","archived":false,"fork":false,"pushed_at":"2026-05-22T13:33:29.000Z","size":52552,"stargazers_count":812,"open_issues_count":7,"forks_count":90,"subscribers_count":25,"default_branch":"main","last_synced_at":"2026-05-22T14:43:59.922Z","etag":null,"topics":["ai","ai-agent","bioinformatics","literature-review","llm","mcp","meta-analysis","openclaw","pubmed","research","research-tools","science","scientific-research","self-evolving","zero-hallucination"],"latest_commit_sha":null,"homepage":"http://scienceclaw.science","language":"TypeScript","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/beita6969.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":"docs/security/CONTRIBUTING-THREAT-MODEL.md","support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-03-11T12:17:51.000Z","updated_at":"2026-05-22T02:39:43.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/beita6969/ScienceClaw","commit_stats":null,"previous_names":["beita6969/scienceclaw"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/beita6969/ScienceClaw","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/beita6969%2FScienceClaw","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/beita6969%2FScienceClaw/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/beita6969%2FScienceClaw/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/beita6969%2FScienceClaw/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/beita6969","download_url":"https://codeload.github.com/beita6969/ScienceClaw/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/beita6969%2FScienceClaw/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35905913,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-07-20T02:08:10.276Z","status":"online","status_checked_at":"2026-07-26T02:00:06.503Z","response_time":89,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","ai-agent","bioinformatics","literature-review","llm","mcp","meta-analysis","openclaw","pubmed","research","research-tools","science","scientific-research","self-evolving","zero-hallucination"],"created_at":"2026-07-07T12:02:11.382Z","updated_at":"2026-07-26T08:01:30.519Z","avatar_url":"https://github.com/beita6969.png","language":"TypeScript","funding_links":[],"categories":["Suites, Systems \u0026 Meta"],"sub_categories":["Autonomous Research Systems"],"readme":"\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/banner.png\" alt=\"ScienceClaw — AI Research Gateway\" width=\"800\" /\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003cstrong\u003eA self-evolving AI research colleague for scientists.\u003c/strong\u003e\n\u003c/p\u003e\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"https://img.shields.io/github/stars/beita6969/ScienceClaw?style=flat-square\u0026logo=github\u0026label=Stars\" alt=\"Stars\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/skills-285-8A2BE2?style=flat-square\" alt=\"285 Skills\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/disciplines-28+-2a9d8f?style=flat-square\" alt=\"28+ Disciplines\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/hallucination-zero-e05d44?style=flat-square\" alt=\"Zero Hallucination\"\u003e\n  \u003cimg src=\"https://img.shields.io/github/license/beita6969/ScienceClaw?style=flat-square\" alt=\"License\"\u003e\n\u003c/p\u003e\n\n---\n\n## Why ScienceClaw?\n\nGeneral-purpose AI assistants are built for everyone. ScienceClaw is built for **researchers**.\n\nThe core idea is simple: an AI that does real scientific work — searching literature, querying databases, running analyses — and **gets better at it the more you use it**. It remembers your research context across sessions, adapts its skills to your field, and never fabricates a citation.\n\nScienceClaw is built on the [OpenClaw](https://github.com/openclaw/openclaw) engine, but redesigned from the ground up for academic research.\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/comparison.png\" alt=\"ScienceClaw vs Standard AI\" width=\"720\" /\u003e\n\u003c/p\u003e\n\n---\n\n## 🧬 Core 1: Self-Evolving Skills\n\n**This is ScienceClaw's most important feature.**\n\nMost AI tools ship with a fixed set of capabilities. ScienceClaw's skills **evolve with you**. Every time you complete a research task, the system learns:\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/skill-evolution.png\" alt=\"Skill Self-Evolution Cycle\" width=\"720\" /\u003e\n\u003c/p\u003e\n\n**What this means in practice:**\n\n- **Week 1:** You study immunology. ScienceClaw learns that PubMed + Semantic Scholar works best for your queries, that you prefer forest plots over tables, and that you always need PMID + DOI in citations.\n- **Week 4:** The system has created specialized skills for your subfield — optimized search templates, preferred statistical methods, database priority chains tuned to immunology literature.\n- **Month 3:** ScienceClaw handles your domain like a trained research assistant. It knows which databases to hit first, which journals matter, and how you like your output formatted.\n\n\u003e **Compared to standard OpenClaw:** OpenClaw ships with ~54 general-purpose skills that don't change. ScienceClaw starts with 285 skills and grows from there — the agent writes new `SKILL.md` files at runtime without any redeployment.\n\n---\n\n## 🧠 Core 2: Research Memory That Persists\n\nStandard AI assistants forget everything when the conversation ends. ScienceClaw doesn't.\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/memory-layers.png\" alt=\"Four-Layer Research Memory\" width=\"720\" /\u003e\n\u003c/p\u003e\n\n**What this enables:**\n\n- **\"Continue the literature review we started last Tuesday\"** — it remembers where you left off\n- **\"Use the same search strategy that worked for the BRCA2 project\"** — it retrieves past patterns\n- **Cross-session knowledge accumulation** — findings from project A can inform project B\n- **Smart context pruning** — when the context window fills up, it preserves statistical results, effect sizes, and key citations while compacting intermediate steps\n\n\u003e **Compared to standard OpenClaw:** OpenClaw has a basic memory plugin. ScienceClaw adds temporal decay weighting, LanceDB vector storage, and cross-session research pattern retrieval — specifically designed for long-running academic work.\n\n---\n\n## ⏱️ Core 3: Built for Long-Duration Research\n\nA real literature review takes hours, not seconds. Most AI tools time out after a few minutes. ScienceClaw is engineered for extended research sessions:\n\n| Capability          | Standard OpenClaw      | ScienceClaw                                                       |\n| ------------------- | ---------------------- | ----------------------------------------------------------------- |\n| Agent timeout       | 600s (10 min)          | **3600s (1 hour+)**                                                |\n| Session persistence | Ends with conversation | Heartbeat keeps sessions alive across interruptions               |\n| Research depth      | Single-pass response   | **Multi-phase protocol with mandatory depth thresholds**          |\n| Minimum effort      | No guarantee           | Quick=5, Survey=30, Review=60, Systematic=100+ tool calls         |\n| Early stopping      | Common                 | **Anti-premature-conclusion checklist** blocks shallow answers    |\n| Context management  | Basic truncation       | **Smart compaction** preserves key findings when context fills up |\n\n**The persistence protocol enforces real research depth.** Before ScienceClaw concludes any task, it must verify:\n\n- ✅ Searched at least 3 different databases/sources\n- ✅ Retrieved full metadata (not just titles)\n- ✅ Cross-referenced findings across sources\n- ✅ Checked for contradictory evidence\n- ✅ Verified key statistics against primary sources\n- ✅ Organized results into a structured output file\n- ✅ Met the minimum tool-call threshold for the task type\n\nIf any box is unchecked, it **keeps working** instead of giving you a half-baked answer.\n\n\u003e **Compared to standard OpenClaw:** OpenClaw's default 10-minute timeout is fine for sending messages and setting reminders. ScienceClaw's 1-hour sessions with heartbeat monitoring and mandatory depth enforcement are built for real academic research.\n\n---\n\n## 🚫 Core 4: Zero Hallucination\n\nThis is the highest-priority rule in the entire system. It's non-negotiable.\n\n**The problem:** General AI assistants routinely fabricate citations — inventing DOIs, making up author names, citing papers that don't exist. In scientific work, this is catastrophic.\n\n**ScienceClaw's approach:**\n\n```\nEVERY citation must come from a tool result in the CURRENT conversation.\n\nIf a database didn't return it → you can't cite it.\nIf you're not sure → say \"not verified\" explicitly.\nIf you can't find evidence → say so. Don't guess.\n\nNo \"I think.\" No \"probably.\" No hallucinated PMIDs.\n```\n\nThis is enforced at the protocol level in [`SCIENCE.md`](SCIENCE.md) — the 629-line research protocol that governs all agent behavior. It's not a suggestion. It's a hard rule that applies before any other instruction.\n\n\u003e **Compared to standard OpenClaw:** OpenClaw has no special hallucination controls. ScienceClaw's SCIENCE.md protocol treats every factual claim as requiring evidence — the same standard you'd apply to a manuscript under peer review.\n\n---\n\n## 🌍 Core 5: All of Science, Not Just Biomedicine\n\nScienceClaw covers **natural sciences AND social sciences** across dozens of disciplines:\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/disciplines.png\" alt=\"Scientific Discipline Coverage\" width=\"720\" /\u003e\n\u003c/p\u003e\n\n\u003cdetails\u003e\n\u003csummary\u003e\u003cstrong\u003e📋 Full discipline \u0026 database list\u003c/strong\u003e\u003c/summary\u003e\n\n### Natural Sciences\n\n| Domain                    | Key Skills \u0026 Databases                                                |\n| ------------------------- | --------------------------------------------------------------------- |\n| **Biomedicine**           | PubMed, UniProt, KEGG, PDB, ClinicalTrials, gnomAD, scanpy, biopython |\n| **Chemistry**             | PubChem, ChEMBL, RDKit, drug-discovery, molecular-dynamics            |\n| **Genomics**              | NCBI Entrez, Ensembl, ClinVar, GEO, phylogenetics                     |\n| **Materials Science**     | Materials Project, pymatgen, materials-screening                      |\n| **Physics**               | astropy, quantum-computing, physics-solver, simulation                |\n| **Environmental Science** | Copernicus climate data, geospatial analysis, GIS tools               |\n| **Food Science**          | Specialized analysis pipelines                                        |\n\n### Social Sciences\n\n| Domain                | Key Skills \u0026 Databases                                  |\n| --------------------- | ------------------------------------------------------- |\n| **Economics**         | World Bank, SSRN, census data, econometrics             |\n| **Political Science** | Policy analysis, legislative data                       |\n| **Psychology**        | Experimental design, statistical testing, meta-analysis |\n| **Linguistics**       | spaCy, NLTK, NLP analysis                               |\n| **Education**         | Research methodology, assessment analysis               |\n| **Sociology**         | Network analysis, survey methods                        |\n\n### Cross-Disciplinary Tools\n\n| Category          | Capabilities                                                                                          |\n| ----------------- | ----------------------------------------------------------------------------------------------------- |\n| **Statistics**    | SciPy, statsmodels, scikit-learn, effect sizes, confidence intervals, multiple comparison corrections |\n| **Visualization** | matplotlib, plotly, seaborn, publication-quality figures                                              |\n| **Writing**       | LaTeX papers, systematic reviews (PRISMA), grant proposals, patent drafting                           |\n| **Mathematics**   | SymPy symbolic computation, numerical methods, optimization                                           |\n\n\u003c/details\u003e\n\n**285 skills total** — and growing, because the self-evolution system creates new ones as you work.\n\n\u003e **Compared to standard OpenClaw:** OpenClaw has no scientific database integrations. No PubMed, no UniProt, no arXiv, no World Bank. ScienceClaw connects to 25+ academic databases with structured API query skills across all major scientific disciplines.\n\n---\n\n## Quick Start\n\n```bash\n# Clone\ngit clone https://github.com/beita6969/ScienceClaw.git\ncd ScienceClaw\n\n# One-click setup (installs everything: Node, Python, MCP servers, skills)\nchmod +x setup.sh \u0026\u0026 ./setup.sh\n\n# Or manual install\npnpm install \u0026\u0026 npx openclaw onboard\n```\n\n### Enable Research Features\n\nThe `setup.sh` script automatically configures everything. For manual setup, edit `~/.openclaw/openclaw.json`:\n\n```jsonc\n{\n  \"gateway\": { \"mode\": \"local\" },\n  \"plugins\": {\n    \"slots\": { \"memory\": \"memory-core\" },\n    \"entries\": {\n      \"memory-core\": { \"enabled\": true },\n      \"memory-lancedb\": { \"enabled\": true }\n    }\n  },\n  \"agents\": {\n    \"defaults\": {\n      \"heartbeat\": { \"interval\": 1800 }\n    }\n  }\n}\n```\n\n---\n\n## Project Structure\n\n```\nScienceClaw/\n├── setup.sh                # 🦞 One-click setup (run this first!)\n├── SCIENCE.md              # 629-line research protocol (the brain)\n├── skills/                 # 285 skill definitions (and growing)\n│   ├── skill-evolution/    # Self-improving skill system\n│   ├── research-reflection/# Post-task learning \u0026 evaluation\n│   ├── skill-creator/      # Runtime skill generation\n│   └── ...\n├── src/                    # Core engine\n│   ├── memory/             # 4-layer memory (temporal decay, LanceDB)\n│   ├── agents/             # Agent orchestration \u0026 persistence\n│   └── skills/             # Skill loading \u0026 execution\n├── ui/                     # Web-based research gateway UI\n├── extensions/             # Plugin system\n├── deploy/                 # Docker, Fly.io, Podman configs\n├── config/                 # Vitest, build, lint configs\n└── docs/                   # Documentation\n```\n\n## Contact Us\n\n📧 **mingdazhang@ieee.org**\n\n## License\n\nMIT — see [LICENSE](LICENSE).\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbeita6969%2Fscienceclaw","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fbeita6969%2Fscienceclaw","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fbeita6969%2Fscienceclaw/lists"}