AgentSkillsCN

system-detective

自动识别调查情境，并引导用户前往合适的工具。触发条件：“调查 X”、“查询 Y”、“查看 Z”、“嗅探”、“检查”。决策树：Jira → jira-check.py，URL → 网络抓取/浏览器，系统 → 执行/进程，主题 → 图书管理员。大幅降低工具选择的复杂性。

SKILL.md
--- frontmatter
name: system-detective
description: "Auto-detect investigation context and route to right tool. Triggers: 'investiga X', 'consulta Y', 'vê Z', 'sniff', 'inspeciona'. Decision tree: Jira → jira-check.py, URLs → web-fetch/browser, systems → exec/process, topics → librarian. Reduces tool selection overhead."
type: public
version: 1.0.0
status: testing
dependencies: []
author: nonlinear
license: MIT

System Detective Skill

Triggers: investiga, consulta, vê, sniff, inspeciona

Purpose: Auto-detect the right tool to investigate/inspect systems, URLs, or topics.

Decision Tree

When user says "investiga X" or "vê X":

1. Detect context type:

A. Jira (Wiley tasks):

  • •URL contains jira.wiley.com or wiley-global.atlassian.net
  • •Keywords: task, ticket, UXPMS-XXX
  • •Action: Use jira-check.py or Jira REST API
  • •Credentials: ~/Documents/wiley/jira-check.py (API token embedded)

B. Figma (design files):

  • •URL contains figma.com/design or figma.com/board
  • •Keywords: figma, design, figjam
  • •Action: Use Figma MCP (mcporter call "Figma Desktop.get_metadata")
  • •Note: Figma Desktop must be running

C. Books/Topics (librarian):

  • •Default fallback when no specific system detected
  • •Any conceptual query, topic, or "what is X?" question
  • •Direct triggers: pesquisa, research
  • •Action: Use librarian skill (~/Documents/librarian/engine/scripts/research.py)
  • •Follow librarian rules: cite sources, check books first, NEVER generic advice

Detection logic:

  • •If detective trigger (investiga|consulta|vê|sniff) + no URL/system match → librarian
  • •Topics are NOT hardcoded (dynamic based on book index)
  • •Examples: "investiga servitors", "vê o que diz sobre familiars", "consulta chaos magick"

D. NAS apps (home server):

  • •URL contains 192.168.1.152 or keywords: kavita, paperless, homeassistant, changedetection, searxng
  • •Action:
    • •Browser relay (if badge ON)
    • •SSH + curl fallback: ssh nonlinear@192.168.1.152 "curl -s http://localhost:PORT"
  • •NAS services:
    • •Kavita: :5000 (ebooks)
    • •Paperless: :8010 (documents)
    • •Home Assistant: :8123 (devices, automation)
    • •changedetection.io: :5555 (price tracking)
    • •SearXNG: :8888 (meta-search)

E. RPM/Wiley systems:

  • •URL contains .wiley.host or .wiley.com
  • •Action: Manual (browser relay blocked by MDM)
    • •Ask for screenshot
    • •Guide navigation verbally
  • •Known issue: Chrome relay fails on Wiley domains (corporate restriction)

F. Web URLs (general):

  • •HTTP/HTTPS URL not matching above
  • •Action:
    1. •Try browser relay (if badge ON and not Wiley)
    2. •Fallback: web_fetch (markdown extraction)

G. Ambiguous/unclear:

  • •Ask user: "Quer que eu investigue via Jira, Figma, livros, ou web?"

Examples

User: "investiga UXPMS-168"
→ Jira API (get issue details)

User: "vê o Figma do RPM"
→ Figma MCP (needs URL or current selection)

User: "pesquisa tarot no livro"
→ Librarian skill (research.py)

User: "sniff kavita"
→ Browser relay (if ON) or SSH curl to :5000

User: "consulta RPM login page"
→ Manual (screenshot request, Wiley domain blocked)

User: "investiga https://example.com"
→ Browser relay or web_fetch

NAS Inventory (192.168.1.152)

Services running:

  • •Kavita (:5000) - Ebook library, EPUB/PDF reader
  • •Paperless-ngx (:8010) - Document management (OCR, tags, archive)
  • •Home Assistant (:8123) - Smart home automation, lights, climate
  • •changedetection.io (:5555) - Website monitoring, price tracking
  • •SearXNG (:8888) - Privacy-focused meta-search engine

Access methods:

  • •Local network: http://192.168.1.152:PORT
  • •Remote (Tailscale, iPad/iPhone only): http://media.adal-rigel.ts.net:PORT
  • •SSH: ssh nonlinear@192.168.1.152 (password: sambalele, sudo required for docker)

Browser relay: Works for Kavita, others untested (badge ON required)

Tools Reference

Jira:

  • •Script: python3 ~/Documents/wiley/jira-check.py
  • •API: https://wiley-global.atlassian.net/rest/api/3/
  • •Auth: Embedded in jira-check.py

Figma:

  • •MCP: mcporter call "Figma Desktop.get_metadata" --nodeId "X:Y"
  • •Requires: Figma Desktop app running

Librarian:

  • •Script: cd ~/Documents/librarian/engine/scripts && python3 research.py "query" --top-k N
  • •Always cite sources (book title, page, chapter)

Browser relay:

  • •Check tabs: browser action=tabs profile=chrome
  • •Snapshot: browser action=snapshot profile=chrome
  • •Badge must be ON (per-tab)

Web fetch:

  • •Tool: web_fetch url="https://..." extractMode="markdown"
  • •Fallback when relay unavailable

Notes

  • •Wiley MDM blocks Chrome DevTools Protocol on .wiley.host domains
  • •Figma MCP may be offline (check with mcporter list)
  • •NAS accessible only on home network (MacBook can't use Tailscale)
  • •Librarian priority: Check books BEFORE giving generic advice
  • •2-minute rule: If investigation >2min, create epic

Last updated: 2026-02-05