skills-router

1|--- 2|name: skills-router 3|description: "Auto-load matching skills based on user intent keywords. Inspired by gstack's routing pattern." 4|version: 1.0.0 5|triggers: 6| - router 7| - "which skill" 8| - "auto skill" 9| - "enable auto skills" 10| - "disable auto skills" 11|allowed-tools: 12| - Bash 13| - Read 14| - skill_view 15| - skills_list 16|--- 17| 18|# Skills Router 19| 20|Auto-detects user intent from messages and loads matching skills. Based on gstack's routing pattern: "When in doubt, invoke the skill — a false positive is cheaper than a false negative." 21| 22|## How It Works 23| 24|1. Scan user's message for trigger keywords 25|2. Match against registered skill triggers 26|3. Auto-load top matching skill via skill_view() 27|4. Reply with loaded skill context 28| 29|## Routing Rules 30| 31|When you see these patterns, load the corresponding skill via skill_view(): 32| 33|### Agent & Infrastructure 34|- "agent rules", "agent behavior", "doctrine" → agent-doctrine 35|36|- "webhook", "subscribe", "event trigger" → webhook-subscriptions 37|- "cognee", "memory backend", "vector memory" → hermes-cognee 38|- "godmode", "jailbreak", "red team", "越狱" → godmode 39|- "python debug", "debugpy", "VSCode debug", "attach debugger" → python-debugpy 40|- "learn yourself", "自主学", "新工具不熟", "查官方文档", "自学技能" → self-learning 41| 42|### Gaming 43|- "minecraft", "MC server", "modpack" → minecraft-modpack-server 44|- "pokemon", "宝可梦", "pixelmon" → pokemon-player 45| 46|### Code & Development 47|- "lint", "format", "style check" → hooks-system (auto-format hooks) 48|- "hook", "pre-commit", "post-save" → hooks-system 49|- "review code", "code review", "PR review" → requesting-code-review 50|- "test", "TDD", "unit test" → test-driven-development 51|- "refactor", "clean up code" → simplify-code 52|- "debug", "bug", "why is this broken" → systematic-debugging 53|- "plan", "architecture", "design doc" → plan 54|- "repo map", "codebase map", "项目地图" → repomap 55|- "token save", "省 token", "简洁" → concise-output 56|- "agent team", "multi agent", "parallel agents", "orchestrate", "多 Agent", "kanban", "merge gate" → agent-orchestration (统一编排) 57|- "auto memory", "auto learn", "continuous learning", "session summary" → auto-memory 58|- "ECC", "everything claude code" → (research: ECC knowledge base node + GitHub API) 59|- "MCP gateway", "meta MCP", "MCP 网关", "MCP 编排", "MCP 401", "MCP auth", "add mcp tool", "REST to MCP" → mcp-gateway (统一 MCP 全栈) 60|- "build", "develop", "implement", "feature", "做功能", "开发", "全流程", "auto develop" → superpowers-workflow (全自动流水线) 🆕 61|- "codebase", "index", "codegraph", "代码索引", "project structure", "find in project" → codegraph-index (增量代码索引) 🆕 62| 63|### ECC Imports 64|- "instinct", "continuous learn", "confidence score", "本能学习" → continuous-learning-v2 65|- "iterative retrieval", "subagent context", "context problem" → iterative-retrieval 66|- "token budget", "response depth", "brief answer", "detailed" → token-budget-advisor 67|- "context budget", "context limit", "上下文预算" → context-budget 68|- "compact", "compress context", "压缩上下文" → strategic-compact 69|- "prompt optimize", "prompt engineering", "提示优化" → prompt-optimizer 70|71|- "verification", "eval loop", "pass@k", "验证循环" → verification-loop 72|- "agent eval", "agent benchmark", "评估" → agent-eval 73|- "security scan", "security audit", "安全扫描", "safety guard", "安全护栏", "guarail", "destructive" → security (统一安全) 74|75|- "knowledge ops", "知识运营", "knowledge sync", "ingest knowledge" → knowledge-base (吸收 knowledge-ops) 76|77| 78|### Architecture & Research 79|- "Claude Code", "agent framework", "agent 架构" → load hermes-agent then research via API 80|- "GitHub research", "挖 GitHub", "search repos" → skills-router (see references/github-api-research.md) 81| 82|### Project Management 83|- "GitHub", "github", "issue", "PR", "pull request", "merge", "clone", "fork", "code review", "gh", "git push", "branch", "release", "workflow", "actions", "CI", "token", "auth setup" → github-workflow (统一 GitHub 入口) 84|- "codebase stats", "LOC", "language breakdown", "pygount" → codebase-inspection 85|- "cron", "schedule", "定时" → (use cronjob tool, not a skill) 86| 87|### Infrastructure 88|- "deploy", "CI/CD", "pipeline" → github-workflow 89|- "config", "setup Hermes" → hermes-agent 90|- "memory", "remember this" → (use memory tool) 91|- "knowledge base", "笔记", "Obsidian", "知识库", "vault", "知识图谱", "加到图谱", "更新图谱", "全量更新" → knowledge-base (优先) 或 obsidian 92| 93|### Creative 94|- "ASCII", "figlet", "banner" → ascii-art 95|- "diagram", "架构图", "流程图" → architecture-diagram 96|- "UI", "design", "mockup" → popular-web-designs 97|- "heartmula", "Suno-like", "song generate", "AI 写歌" → heartmula 98| 99|### Social & Apple 100|- "xurl", "twitter share", "X post", "发推" → xurl 101|- "imessage", "iMessage", "Apple message", "苹果短信" → imessage 102|- "findmy", "Find My", "查找设备", "定位 iPhone" → findmy 103| 104|### Data & Research 105|- "paper", "arXiv", "论文" → arxiv 106|- "research paper", "paper writing", "学术写作", "论文写作" → research-paper-writing 107|- "YouTube", "transcript", "视频总结" → youtube-content 108|- "blog", "RSS", "监控" → blogwatcher 109|- "notebook", "Jupyter", "数据分析" → jupyter-live-kernel 110|- "audiocraft", "music gen", "audio generation", "音乐生成" → audiocraft 111|- "vllm", "LLM serving", "模型部署", "inference engine" → vllm 112|- "eval harness", "benchmark", "模型评测", "lm eval" → lm-evaluation-harness 113| 114|## Proactive Mode 115| 116|By default, the router is proactive: it suggests and loads relevant skills without being asked. The user can disable with "stop suggesting skills" or enable with "be proactive about skills". 117| 118|When proactive is ON: 119|1. Before answering any user request, scan for matching skill triggers 120|2. If a skill matches, load it via skill_view() FIRST 121|3. Then proceed with the task using the skill's context 122| 123|When proactive is OFF: 124|1. Only load skills when the user explicitly asks ("/skill name" or "use the X skill") 125|2. Answer directly without scanning 126| 127|## Usage 128| 129|Load this skill at session start: 130| 131|hermes -s skills-router 132| 133| 134|Or in-session: 135| 136|/skill skills-router 137| 138| 139|Then just talk normally — the router will auto-load relevant skills for each request. 140| 141|## Configuration 142| 143|Custom routing rules can be added to ~/.hermes/skills/router-rules.json: 144| 145|json 146|{ 147| "keyword": "skill-name", 148| "deploy k8s": "kubernetes-deploy" 149|} 150| 151| 152|## References 153| 154|- references/github-api-research.md — GitHub API + raw file fetch pattern for restricted networks (no git clone needed) 155|- references/github-access-china.md — GitHub access from China: status matrix, reliable patterns, anti-patterns 156|