[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-731":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":39,"readmeContent":40,"aiSummary":41,"trendingCount":16,"starSnapshotCount":16,"syncStatus":42,"lastSyncTime":43,"discoverSource":44},731,"SkillClaw","AMAP-ML\u002FSkillClaw","AMAP-ML","Let Skills Evolve Collectively with Agentic Evolver ","",null,"Python",1836,177,30,21,0,207,287,532,621,19.75,"MIT License",false,"main",[26,27,28,29,30,31,32,33,34,35,36,37,38],"agent","agentic-ai","ai-agent","collective-intelligence","continual-learning","hermes","llm","llms","openclaw","self-evolving","skill-evolution","skill-learning","skills","2026-06-12 02:00:17","\u003Cdiv align=\"center\">\n\n\u003Cimg src=\"assets\u002Fskillclaw_logo.png\" alt=\"SkillClaw\" width=\"150\">\n\n# ✨ SkillClaw: Let Skills Evolve Collectively with Agentic Evolver ✨\n\n\u003Ch3>AI agent skills that evolve from every real interaction — just talk.\u003Cbr>Across sessions, agents, devices, and users. Experience compounds. Skills keep growing.\u003C\u002Fh3>\n\n| 🚀 Quick Install | 💬 Just Chat | 🔌 Broad Compatibility | 🧬 Collective Skill Evolution |\n|:-:|:-:|:-:|:-:|\n\n[![Hermes](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FHermes-supported-ff6b6b?style=flat-square&logo=data:image\u002Fsvg+xml;base64,PHN2ZyB4bWxucz0iaHR0cDovL3d3dy53My5vcmcvMjAwMC9zdmciIHZpZXdCb3g9IjAgMCAyNCAyNCI+PHBhdGggZmlsbD0id2hpdGUiIGQ9Ik0xMiAyTDIgN2wxMCA1IDEwLTV6TTIgMTdsOSA1VjEyTDIgN3pNMTMgMjJsOS01VjdMMTMgMTJ6Ii8+PC9zdmc+)](https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent)\n[![OpenClaw](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FOpenClaw-supported-6c5ce7?style=flat-square)](https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw)\n[![Agents](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002F+QwenPaw%20%7C%20IronClaw%20%7C%20PicoClaw%20%7C%20ZeroClaw%20%7C%20...-8A2BE2?style=flat-square)](https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw)\n[![Python](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPython-3.10%2B-3776AB?style=flat-square&logo=python&logoColor=white)](https:\u002F\u002Fwww.python.org\u002F)\n[![License](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-green?style=flat-square)](.\u002FLICENSE)\n[![Paper](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaper-arXiv-b5212f.svg?style=flat-square&logo=arxiv)](https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.08377)\n[![Paper](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaper-PDF-red?style=flat-square&logo=adobeacrobatreader)](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2604.08377)\n[![Paper](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FPaper-Hugging%20Face-yellow?style=flat-square&logo=huggingface)](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2604.08377)\n[![WeChat](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FWeChat-Group-07C160?style=flat-square&logo=wechat&logoColor=white)](assets\u002Fimage.png)\n[![Docs](https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FDocs-中文版-orange?style=flat-square)](assets\u002FREADME_ZH.md)\n\n\u003Cbr>\n\n\u003Ca href=\"https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw\">\u003Cimg src=\"https:\u002F\u002Freadme-typing-svg.demolab.com?font=Fira+Code&weight=600&size=18&duration=2500&pause=1000&color=58A6FF&center=true&vCenter=true&width=750&lines=Collective+Skill+Evolution+for+AI+Agents;Hermes+%C2%B7+OpenClaw+%C2%B7+QwenPaw+%C2%B7+IronClaw+%C2%B7+PicoClaw+%C2%B7+ZeroClaw+and+more\" alt=\"Typing SVG\" \u002F>\u003C\u002Fa>\n\n\u003Cimg src=\"assets\u002Fterminal_cmd.svg\" alt=\"skillclaw setup && skillclaw start --daemon\" width=\"620\">\n\n\u003C\u002Fdiv>\n\n\u003Cbr>\n\n\u003Ctable>\n\u003Ctr>\n\u003Ctd>🚀 \u003Cb>Quick Install\u003C\u002Fb>\u003C\u002Ftd>\n\u003Ctd>Shell installer for macOS\u002FLinux, plus a manual Python install path for Windows. Then run \u003Ccode>skillclaw setup\u003C\u002Fcode> and \u003Ccode>skillclaw start --daemon\u003C\u002Fcode>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>💬 \u003Cb>Just Chat\u003C\u002Fb>\u003C\u002Ftd>\n\u003Ctd>Just talk to your agent as usual — skill evolution happens silently in the background. Zero extra effort.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>🔌 \u003Cb>Broad Compatibility\u003C\u002Fb>\u003C\u002Ftd>\n\u003Ctd>Natively integrates with \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent\">Hermes\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex\">Codex\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code\">Claude Code\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fopenclaw\u002Fopenclaw\">OpenClaw\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fagentscope-ai\u002FQwenPaw\">QwenPaw\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fnearai\u002Fironclaw\">IronClaw\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fsipeed\u002Fpicoclaw\">PicoClaw\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fzeroclaw-labs\u002Fzeroclaw\">ZeroClaw\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fqwibitai\u002FNanoClaw\">NanoClaw\u003C\u002Fa>, \u003Ca href=\"https:\u002F\u002Fgithub.com\u002FNVIDIA\u002FNemoClaw\">NemoClaw\u003C\u002Fa>, and any OpenAI-compatible API.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>🧬 \u003Cb>Collective Skill Evolution\u003C\u002Fb>\u003C\u002Ftd>\n\u003Ctd>Skills evolve from every session, every agent, every context. Solo or team — the loop is the same. Every experience compounds.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftable>\n\n---\n\n\u003Cdiv align=\"center\">\n\n## What SkillClaw Brings to One Hermes User\n\n\u003C\u002Fdiv>\n\nBeen using Hermes for a while — is your skill library still a mess? Duplicates, outdated ones, half-baked ones all piled together like an unsorted loot box. The problem isn't that Hermes doesn't learn enough — it's that nobody helps it **digest**.\n\n**SkillClaw is built for this.** Auto-evolve, auto-deduplicate, auto-improve quality. It won't change how you work or interrupt your flow — it just quietly rewrites your agent's growth curve.\n\nSkillClaw doesn't make Hermes learn more — it makes everything Hermes has learned actually count.\n\n\u003Cdiv align=\"center\">\n\n\u003Cimg src=\"assets\u002Ftwo_loops.svg\" alt=\"Two Loops: Hermes task-time loop + SkillClaw post-task evolution loop\" width=\"860\">\n\n\u003C\u002Fdiv>\n\nThat's just one user's story. One user can also run multiple agents or use multiple devices — SkillClaw unifies them all:\n\n\u003Cdiv align=\"center\">\n\n### Multiple agents? One unified skill library.\n\n\u003Cimg src=\"assets\u002Fmultiplier_effect.svg\" alt=\"The Multiplier Effect: Multiple Hermes agents sharing skills through SkillClaw\" width=\"860\">\n\n\u003C\u002Fdiv>\n\nRunning multiple Hermes agents for different tasks? Without SkillClaw, each builds its own isolated skill silo. With SkillClaw, skills are **merged, deduplicated, and cross-pollinated** into a unified library, then distributed back to all agents. Your Frontend agent's React patterns make the Backend agent's API design better — and vice versa.\n\n\u003Cdiv align=\"center\">\n\n### Multiple devices? Skills follow you, not your machine.\n\n\u003Cimg src=\"assets\u002Fcross_context.svg\" alt=\"Seamless Context: Home, School, and Company Hermes instances unified by SkillClaw\" width=\"860\">\n\n\u003C\u002Fdiv>\n\nSame user, different machines. Your Home Hermes learns React; your School Hermes learns ML; your Work Hermes learns K8s. Without SkillClaw, each starts from scratch. With it, **skills unify across all environments** — every Hermes instance benefits from every other's experience, regardless of where you are.\n\n---\n\n\u003Cdiv align=\"center\">\n\n## Collective Skill Evolution\n\n\u003C\u002Fdiv>\n\nEverything above is what one user gets. Now scale it up: when you join a shared group, **every team member's real-world experience feeds into the same evolution loop**. User A debugs a database issue — the skill evolves. User B, C, D benefit instantly without ever hitting the same problem. N users, one Skill, continuous evolution.\n\n\u003Cdiv align=\"center\">\n\n\u003Cimg src=\"assets\u002Fshift_contrast.svg\" alt=\"The Shift: From Experience Silos to Collective Evolution\" width=\"860\">\n\n\u003Cbr>\n\n\u003Cimg src=\"assets\u002Fskill_evolution.svg\" alt=\"Skill Evolution Flow: How a skill evolves across multiple users\" width=\"860\">\n\n\u003C\u002Fdiv>\n\n---\n\n\u003Cdiv align=\"center\">\n\n\u003Cimg src=\"assets\u002Fskillclaw_main.png\" alt=\"SkillClaw Architecture\" width=\"680\">\n\n\u003C\u002Fdiv>\n\n---\n\n## News\n\n- **2026\u002F04\u002F22** — Added a bilingual dashboard with `skillclaw dashboard sync` and `skillclaw dashboard serve` for inspecting local\u002Fshared skills, validation progress, version history, and session traces.\n- **2026\u002F04\u002F20** — Added [Codex](https:\u002F\u002Fgithub.com\u002Fopenai\u002Fcodex) and [Claude Code](https:\u002F\u002Fdocs.anthropic.com\u002Fen\u002Fdocs\u002Fclaude-code) integration with proxy auto-configuration, native skills-directory defaults, and `doctor` \u002F `restore` commands.\n- **2026\u002F04\u002F17** — Added [QwenPaw](https:\u002F\u002Fgithub.com\u002Fagentscope-ai\u002FQwenPaw) integration and updated the docs for broader multi-agent compatibility.\n- **2026\u002F04\u002F17** — Added full [Hermes](https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent) integration, per-turn skill tracking, `doctor hermes`, `skillclaw skills *` management commands, and a major docs overhaul.\n- **2026\u002F04\u002F14** — WeChat discussion group is live! [Join the group](assets\u002Fimage.png) to chat with us.\n- **2026\u002F04\u002F14** — Initial [Hermes](https:\u002F\u002Fgithub.com\u002FNousResearch\u002Fhermes-agent) support landed together with the first README refresh.\n- **2026\u002F04\u002F12** — Active discussion with [Deer-Flow](https:\u002F\u002Fgithub.com\u002Fbytedance\u002Fdeer-flow\u002Fdiscussions\u002F2133) on cross-framework skill sharing.\n- **2026\u002F04\u002F11** — SkillClaw ranked **#2 Paper of the Day** on [Hugging Face Daily Papers](https:\u002F\u002Fhuggingface.co\u002Fpapers\u002F2604.08377)!\n- **2026\u002F04\u002F10** — SkillClaw is now open source! Code released on [GitHub](https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw).\n\n---\n\n## Overview\n\nSkillClaw makes LLM agents progressively better by **evolving reusable skills** from real session data. A single user already benefits — skills are automatically deduplicated, improved, and verified across sessions. Scale up when you're ready: multiple agents, multiple devices, or multiple users can all feed the same evolution loop.\n\nThe system has two components:\n\n1. **Client Proxy** — A local API proxy (`\u002Fv1\u002Fchat\u002Fcompletions`, `\u002Fv1\u002Fmessages`) that intercepts agent requests, records session artifacts, and manages your local skill library. This is all you need to get started.\n\n2. **Evolve Server** (`evolve_server`) — An optional service that reads session data from shared storage, evolves or creates skills, and writes them back. Add it when you want automatic evolution or team-wide sharing. It supports two engines:\n   - `workflow`: fixed 3-stage LLM pipeline (Summarize → Aggregate → Execute)\n   - `agent`: OpenClaw-driven agent workspace with direct skill editing\n\nBoth components share the same storage layer (Alibaba OSS \u002F S3 \u002F local filesystem) and skill format (`SKILL.md`).\n\n---\n\n## Deployment Model\n\nStart with just the client. Add the server when you need it.\n\n1. **Single user + auto-evolution**: Install the client proxy, then add an evolve server on the same machine (or anywhere that can reach your storage) to automatically refine skills in the background.\n2. **Team \u002F shared group**: Point multiple clients at the same shared storage and run one `skillclaw-evolve-server` for the group. Everyone's experience feeds the same evolution loop.\n\nThe client and server only meet through shared storage (`local`, `oss`, or `s3`). This means:\n\n- If you only want to use SkillClaw yourself, install the client first. You can add an evolve server later.\n- If you want to join an existing team, you still install only the client. You do not run the evolve server unless you are operating the shared group.\n- The evolve server can run on the same laptop, a remote VM, or any machine that can access the shared storage and upstream LLM endpoint.\n\n\n## User Guide\n\nIf this is your first time, start with Path A. It proves the client-side install and usage first, without mixing in shared deployment concerns.\n\n### Prerequisites\n\n- macOS, Linux, or Windows\n- Python >= 3.10\n- A provider account that exposes an OpenAI-compatible API, or AWS Bedrock\n- Install `openclaw` only if you intentionally choose the `openclaw` CLI integration or the server `agent` engine\n\nThe beginner path below is locally smoke-tested on macOS.\n\n### Path A: Run SkillClaw for yourself on one machine\n\n1. Install SkillClaw from this repository. If you already have the source checkout, skip `git clone`.\n\nmacOS \u002F Linux:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw.git && cd SkillClaw\nbash scripts\u002Finstall_skillclaw.sh\nsource .venv\u002Fbin\u002Factivate\n```\n\nWindows PowerShell (manual install because the repository does not currently ship a native `.ps1` installer):\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw.git\nSet-Location SkillClaw\npython -m venv .venv\n.\\.venv\\Scripts\\Activate.ps1\npython -m pip install -U pip\npython -m pip install -e \".[evolve,sharing,server]\"\n```\n\n2. Generate a local config.\n\n```bash\nskillclaw setup\n```\n\nThe setup wizard prompts for the provider, model, local skills directory, PRM settings, optional CLI agent integration, and optional shared storage.\n\nFor a minimal first run:\n\n- choose `none` for the CLI agent if you do not want SkillClaw to auto-configure an external agent yet\n- local skills at `~\u002F.skillclaw\u002Fskills` for the generic setup path; if you choose Hermes, Codex, or Claude Code, the default local library becomes `~\u002F.hermes\u002Fskills`, `~\u002F.codex\u002Fskills`, or `~\u002F.claude\u002Fskills`\n- disable shared storage if you only want to use the local proxy first\n- enable local shared storage only if you want to add the evolve server later on the same machine, and use a dedicated root such as `~\u002F.skillclaw\u002Flocal-share`\n- disable PRM if you want the cheapest first pass\n\n3. Start the client proxy and verify that it is healthy.\n\n```bash\nskillclaw start --daemon\nskillclaw status\nPROXY_PORT=\"$(skillclaw config proxy.port | awk '{print $2}')\"\ncurl \"http:\u002F\u002F127.0.0.1:${PROXY_PORT}\u002Fhealthz\"\n```\n\nThe default proxy port is `30000`, but the health check should follow your configured `proxy.port`. Use `skillclaw config show` to inspect the active upstream model, proxy port, and sharing target.\n\nAt this point SkillClaw is already usable as a single-user local proxy. You do not need to run an evolve server just to use the client.\n\nIf you later want automatic skill evolution for yourself, keep the same client install and continue with [Server Guide](#server-guide).\n\n### Hermes Integration\n\nIf you already use Hermes, the client-side path is:\n\n1. Install Hermes first.\n2. Run `skillclaw setup` and choose `hermes` for `CLI agent to configure`.\n3. Keep `Proxy model name exposed to agents` as `skillclaw-model` unless you have a specific reason to change it.\n4. Start SkillClaw. On startup, SkillClaw rewrites `~\u002F.hermes\u002Fconfig.yaml` to point Hermes at the local proxy.\n5. Hermes uses `~\u002F.hermes\u002Fskills` as the default local skill library. SkillClaw prepares that directory automatically and copies in any missing legacy skills from `~\u002F.skillclaw\u002Fskills`.\n6. If you want to inspect or undo the integration, use `skillclaw doctor hermes` and `skillclaw restore hermes`.\n\nMinimal verification:\n\n```bash\nskillclaw start --daemon\nhermes chat -Q -m skillclaw-model -q \"Reply with exactly HERMES_SKILLCLAW_OK and nothing else.\"\n```\n\nOptional diagnostics:\n\n```bash\nskillclaw doctor hermes\nskillclaw restore hermes\n```\n\n`skillclaw doctor hermes` reports whether Hermes is pointed at the local proxy, whether the Hermes skills directory exists, whether legacy skills are still present, and that session boundaries still fall back to proxy-side heuristics unless Hermes sends explicit session headers.\n\n### Path B: Join an existing shared group\n\nInstall the same client as in Path A, then point your local client at the group's shared storage. The easiest beginner route is to rerun `skillclaw setup`, enable shared storage, and fill in the values your server operator gives you.\n\nYou can also set the keys directly. Example for OSS:\n\n```bash\nskillclaw config sharing.enabled true\nskillclaw config sharing.backend oss\nskillclaw config sharing.endpoint https:\u002F\u002Foss-cn-hangzhou.aliyuncs.com\nskillclaw config sharing.bucket my-skillclaw-bucket\nskillclaw config sharing.access_key_id \"$OSS_ACCESS_KEY_ID\"\nskillclaw config sharing.secret_access_key \"$OSS_ACCESS_KEY_SECRET\"\nskillclaw config sharing.group_id my-group\nskillclaw config sharing.user_alias alice\nskillclaw config sharing.auto_pull_on_start true\n\nskillclaw start --daemon\nskillclaw skills pull\n```\n\nIf your team uses a mounted local shared directory instead of OSS\u002FS3, use `sharing.backend local` plus `sharing.local_root \u002Fpath\u002Fto\u002Fshared\u002Froot` instead of the remote storage keys.\n\nWhen you join a shared group:\n\n- you still run only the local client proxy on your machine\n- you do not run `skillclaw-evolve-server` unless you are also operating the shared group\n- moving from single-user to multi-user is mostly a sharing-config change on the client side\n\n### Optional: turn one client into a background validation worker\n\nThis mode is optional and disabled by default. It is meant for groups that want a second review step before a workflow-generated skill is published.\n\nWhat it does:\n\n- the server stages a candidate skill as a validation job instead of publishing it immediately\n- an opted-in client picks up jobs only when its local proxy is idle\n- the client validates the candidate skill in the background and writes back a result\n- a later evolve cycle publishes the candidate only after the configured thresholds are met\n\nIf `validation.enabled` stays `false`, normal client usage is unchanged.\n\nMinimal client-side setup:\n\n```bash\nskillclaw config validation.enabled true\nskillclaw config validation.idle_after_seconds 300\nskillclaw config validation.poll_interval_seconds 60\nskillclaw config validation.max_jobs_per_day 5\n\nskillclaw validation status\nskillclaw validation run-once --force\n```\n\n`skillclaw start --daemon` will automatically run the background validator afterward. `run-once --force` is the quickest way to test the path without waiting for the idle timer.\n\n### Optional: inspect skills and sessions with the dashboard\n\nThe dashboard is a local visualization layer for the current SkillClaw snapshot. It is useful when you want to inspect:\n\n- local skills and whether they match the shared official version\n- candidate validation jobs and their current status\n- published shared skills and version history\n- local and shared sessions behind skill updates\n\nThe dashboard commands are available from the same `skillclaw` install:\n\n```bash\nskillclaw dashboard sync\nskillclaw dashboard serve\n```\n\nIf you want to point the dashboard at a local shared root and a specific group:\n\n```bash\nskillclaw dashboard sync \\\n  --sharing-local-root \u002Fpath\u002Fto\u002Fshared\u002Froot \\\n  --sharing-group-id my-group \\\n  --sharing-user-alias alice\n\nskillclaw dashboard serve \\\n  --host 127.0.0.1 \\\n  --port 3791 \\\n  --sharing-local-root \u002Fpath\u002Fto\u002Fshared\u002Froot \\\n  --sharing-group-id my-group \\\n  --sharing-user-alias alice\n```\n\nThen open:\n\n```text\nhttp:\u002F\u002F127.0.0.1:3791\n```\n\nBy default, `serve` rebuilds the snapshot on startup. If you already ran `skillclaw dashboard sync`, you can start faster with `--no-sync-on-start`.\n\n## Server Guide\n\nThe evolve server is the shared backend for one user or many users. It can run locally for a personal setup, or remotely for a team setup.\n\n### Run one evolve server for a single-user local loop\n\nThis is the smallest full closed loop: one user, one machine, one local shared root, one evolve server.\n\nThis step assumes your client config already enabled `local` shared storage during `skillclaw setup`.\n\n```bash\nskillclaw-evolve-server --use-skillclaw-config --interval 300 --port 8787\n```\n\nOptional: inspect the shared skill store afterward.\n\n```bash\nskillclaw skills list-remote\n```\n\nWhat this gives you:\n\n- one user\n- one machine\n- no OSS\u002FS3 account\n- the full loop: session capture -> skill evolution -> local skill reuse\n\n### Run one evolve server for a shared group on OSS or S3\n\nRun this on any machine that can reach the shared storage and the upstream LLM API. It does not need to be an end-user laptop.\n\nmacOS \u002F Linux:\n\n```bash\ngit clone https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw.git && cd SkillClaw\nbash scripts\u002Finstall_skillclaw_server.sh\nsource .venv-server\u002Fbin\u002Factivate\ncp evolve_server\u002Fevolve_server.env.example evolve_server\u002F.env\n```\n\nWindows PowerShell (manual install because the repository does not currently ship a native `.ps1` installer):\n\n```powershell\ngit clone https:\u002F\u002Fgithub.com\u002FAMAP-ML\u002FSkillClaw.git\nSet-Location SkillClaw\npython -m venv .venv-server\n.\\.venv-server\\Scripts\\Activate.ps1\npython -m pip install -U pip\npython -m pip install -e \".[server]\"\nCopy-Item .\\evolve_server\\evolve_server.env.example .\\evolve_server\\.env\n```\n\nThen fill in `evolve_server\u002F.env`, or pass the same values on the CLI. Example for the default `workflow` engine:\n\n```bash\nskillclaw-evolve-server --port 8787 --interval 300 \\\n  --storage-backend oss \\\n  --oss-endpoint \"$EVOLVE_STORAGE_ENDPOINT\" \\\n  --oss-bucket \"$EVOLVE_STORAGE_BUCKET\" \\\n  --group-id my-group\n```\n\nBy default, the `workflow` engine uploads accepted evolution outputs directly to the shared skill store at `{group_id}\u002Fskills\u002F\u003Cname>\u002FSKILL.md`.\n\nIf you want a staged publish flow where opted-in clients review candidates before they reach `skills\u002F`, switch the server to `validated` publish mode:\n\n```bash\nEVOLVE_PUBLISH_MODE=validated \\\nEVOLVE_VALIDATION_REQUIRED_RESULTS=1 \\\nEVOLVE_VALIDATION_REQUIRED_APPROVALS=1 \\\nEVOLVE_VALIDATION_MIN_MEAN_SCORE=0.75 \\\nEVOLVE_VALIDATION_MAX_REJECTIONS=1 \\\nskillclaw-evolve-server --port 8787 --interval 300 \\\n  --storage-backend oss \\\n  --oss-endpoint \"$EVOLVE_STORAGE_ENDPOINT\" \\\n  --oss-bucket \"$EVOLVE_STORAGE_BUCKET\" \\\n  --group-id my-group\n```\n\nIf you want the `agent` engine on the server side, install `openclaw` there and then run:\n\n```bash\nnpm install -g openclaw\n\nskillclaw-evolve-server --engine agent --port 8787 --interval 300 --no-fresh \\\n  --storage-backend oss \\\n  --oss-endpoint \"$EVOLVE_STORAGE_ENDPOINT\" \\\n  --oss-bucket \"$EVOLVE_STORAGE_BUCKET\" \\\n  --group-id my-group\n```\n\nOnly the server operator needs `openclaw` for `--engine agent`.\n\n## Quick Reference\n\n**Operational model** — Every user has their own `skillclaw` process and `~\u002F.skillclaw\u002Fconfig.yaml`. Each group maps to a `group_id`; one evolve server watches that namespace, turns sessions into skills, and writes them back. Joining a group only requires the correct sharing config on the client side.\n\n**First-run checks**\n\n| Command | Expected |\n|---------|----------|\n| `skillclaw status` | Reports `running` |\n| `curl http:\u002F\u002F127.0.0.1:\u003Cport>\u002Fhealthz` | Returns `{\"ok\": true}` |\n| `skillclaw config show` | Shows correct upstream URL, model, and sharing target |\n\n**Skill management**\n\n```bash\nskillclaw skills pull          # download shared skills\nskillclaw skills push          # upload local skills\nskillclaw skills sync          # bidirectional\nskillclaw skills list-remote   # browse shared skills\n```\n\n**Config reference**\n\n- Client config: `~\u002F.skillclaw\u002Fconfig.yaml` (created by `skillclaw setup`)\n- Server template: [`evolve_server\u002Fevolve_server.env.example`](.\u002Fevolve_server\u002Fevolve_server.env.example) (copy to `.env` to use)\n- Inspect \u002F update config: `skillclaw config show` | `skillclaw config \u003Ckey> \u003Cvalue>`\n- Repo entry points for contributors: `skillclaw\u002F` (client), `evolve_server\u002F` (backend), `scripts\u002F` (installers)\n\n\n## Acknowledgement\nThe repo is built upon these open-source repos.\n\n[MetaClaw](https:\u002F\u002Fgithub.com\u002Faiming-lab\u002FMetaClaw) - Just talk to your agent — it learns and evolves\n\n[WildClawBench](https:\u002F\u002Fgithub.com\u002FInternLM\u002FWildClawBench) - Can an AI agent do real work, end-to-end, without hand-holding\n\n[OpenClaw-RL](https:\u002F\u002Fgithub.com\u002FGen-Verse\u002FOpenClaw-RL) - Train a personalized agent simply by talking to it\n\n## Contributing\n\nSkillClaw is a community-driven project. We welcome contributions of all kinds — bug reports, feature requests, new skills, documentation improvements, and more. Feel free to open an issue or submit a pull request!\n\n## Citation\n\nIf you find SkillClaw useful in your research, please consider citing our paper:\n\n```bibtex\n@article{ma2026skillclaw,\n  title={SkillClaw: Let Skills Evolve Collectively with Agentic Evolver},\n  author={Ma, Ziyu and Yang, Shidong and Ji, Yuxiang and Wang, Xucong and Wang, Yong and Hu, Yiming and Huang, Tongwen and Chu, Xiangxiang},\n  journal={arXiv preprint arXiv:2604.08377},\n  year={2026}\n}\n```\n\n## License\n\nSee [LICENSE](.\u002FLICENSE) for details.\n","SkillClaw 是一个能够让AI代理技能通过真实互动集体进化的项目。它利用持续学习和集体智能技术，使AI代理在与用户交流的过程中自动进化其技能，无需额外的人工干预。支持Hermes、OpenClaw等平台，并兼容QwenPaw、IronClaw等多种AI代理。项目采用Python 3.10+开发，易于安装和配置，适用于需要不断优化用户体验、提升AI代理能力的场景，如客户服务、个人助理等领域。",2,"2026-06-11 02:38:55","CREATED_QUERY"]