mupeng-skill-router_skill

Overview

This skill is a context-based router that inspects user input to auto-select one or multiple skills, determine execution order, and run them as a coordinated pipeline. It acts as the first gateway for every request, composing single or federated skill chains and handling results, events, and hooks. The router enforces safety rules, approval gates for external actions, and cost/loop limits.

How this skill works

The router classifies the incoming natural-language intent as single or complex, then matches triggers in skill metadata to select candidates. It executes standalone skills directly or composes ordered/parallel chains, passing outputs via an events directory and applying post-execution auto-chain rules. Hooks (pre, post, on-error) and an event system let subsequent skills consume results; the router synthesizes final responses and requests approval for external side effects.

When to use it

• As the first entry point for every user request to ensure optimal skill selection
• When an input may require multiple coordinated skills (content pipelines, research→report workflows)
• To automate conditional follow-ups based on execution results (notifications, additional analysis)
• When you need consistent pre/post checks and centralized error handling
• For cost and loop control when composing federated skill chains

Best practices

• Maintain concise trigger phrases in each skill so the router can match quickly
• Use skill metadata to indicate run paths and event outputs for smooth chaining
• Model common chains as templates to reduce decision time and token usage
• Require explicit approval for any external deploy/send action to prevent accidental side effects
• Limit chain size and avoid circular dependencies to respect cost and safety limits

Example use cases

• Create a full content pipeline: seo-content-planner → copywriting → cardnews → insta-post (approval before posting)
• Analyze competitors and report: competitor-watch → daily-report → mail (mail sends after approval)
• Summarize a video and publish: yt-digest → content-recycler → cardnews → insta-post (preview then post)
• Release flow for code: code-review → git-auto → release-discipline with blockage on high-severity findings
• Weekly review routine: self-eval + tokenmeter + performance-tracker → daily-report