360 Secure SkillHub
360 安全技能中心

HK IPO Orchestrator

面向香港 IPO 研究的编排技能,按发现市场、抽取结构化事实、准备个性化配置、评分与决策输出、复盘优化的顺序推进,并把每一步路由到配套子技能;机器步骤间使用 JSON,最终面向用户输出文本或 Markdown,且在改动配置或写入复盘记录前给出提示。

hackstoic
hackstoic
热度 38.85
最近更新 2026-05-11 17:43:46
分类:数据依赖:代码执行 / 读写文件 / 第三方 API应用场景:市场分析 / 财务 / 研究 / 报告撰写 ...

安装技能

复制下面的话术发给你的 Agent,它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill: npx secure-skills-cli add hkipo-orchestrator
Metadata
name
hkipo-orchestrator
description
Coordinate the end-to-end Hong Kong IPO workflow across discovery, snapshot extraction, personalization, scoring, and review optimization. Use when the user wants one skill to choose the next step across the full HK IPO research loop.
version
0.1.0
metadata
{"openclaw":{"emoji":"🧭","requires":{"bins":["uv"],"config":["~/.hkipo-next/config/profile.json","~/.hkipo-next/config/watchlist.json","~/.hkipo-next/data/hkipo.db"]},"install":[{"id":"install-uv-brew","kind":"brew","formula":"uv","bins":["uv"],"label":"Install uv with Homebrew","os":["darwin","linux"]}]}}

HK IPO Orchestrator

Use this skill when the request spans multiple stages and you need to route the work to the right HK IPO skill.

Companion Skill Map

  • $hkipo-calendar-monitor: discover near-term IPO events
  • $hkipo-snapshot-extractor: extract one-symbol facts and quality signals
  • $hkipo-profile-watchlist-manager: manage personalized settings and watchlists
  • $hkipo-parameter-manager: tune scoring rules and active versions
  • $hkipo-decision-engine: produce scores, decision cards, and batches
  • $hkipo-review-optimizer: update actual outcomes and process suggestions

Default Flow

  1. Discover the market.
  2. Extract structured facts for shortlisted symbols.
  3. Ensure profile and watchlist state is ready when the task is personalized.
  4. Ensure an active parameter version exists before scoring.
  5. Produce score or decision outputs.
  6. Close the loop later with review and suggestions.

Operating Rules

  • Prefer JSON between machine steps and concise text or markdown for final user-facing output.
  • Surface degraded or partial data instead of hiding it.
  • Warn before mutating profile, watchlist, parameter, or review history when that side effect matters.
  • Do not run repeated score or batch jobs casually, because those commands create review records.