Die Oberfläche ist lokalisiert; Skill-Beschreibungen und -Dokumente werden per KI automatisch in Ihre Sprache übersetzt. Code, Befehle, Pfade und Fachbegriffe bleiben im Original, maßgeblich ist das Dokument des Autors.
Skills entdecken
Daten und Forschung · SKILL.md

uzi

uzi

A-share, Hong Kong, and US stock analysis skill for deep research, quick scans, investor panel review, hot-money/LHB analysis, trap detection, valuation, IC memos, and Bloomberg-style HTML reports.

Independent GitHub snapshot · Automated classification · Original documentation governs

Skill-Dokumentation

GitHub

name: uzi description: A-share, Hong Kong, and US stock analysis skill for deep research, quick scans, investor panel review, hot-money/LHB analysis, trap detection, valuation, IC memos, and Bloomberg-style HTML reports. version: 3.9.4 author: FloatFu-true license: MIT metadata: tags: [finance, stocks, a-share, hong-kong, us-stocks, dcf, valuation, investor-panel, youzi, lhb, trap-detection] related_skills: [deep-analysis, investor-panel, lhb-analyzer, trap-detector]

UZI Skill Root

This root file is the top-level entry for agents that expect a SKILL.md at the repository root.

Use the narrowest matching workflow:

  • Full stock research, valuation, IC memo, initiation, catalysts, earnings review, or HTML report: read skills/deep-analysis/SKILL.md.
  • Investor jury, "which investors would buy", panel-only voting, or persona review: read skills/investor-panel/SKILL.md.
  • Hot-money, LHB, seat recognition, or A-share short-term trader analysis: read skills/lhb-analyzer/SKILL.md.
  • Trap detection, pump-and-dump checks, "teacher/group/friend recommended this stock", or safety review: read skills/trap-detector/SKILL.md.
  • Command-specific requests: read the matching file under commands/.

Default Execution

From the repository root:

python3 run.py <ticker> --no-browser

For remote/mobile reports:

python3 run.py <ticker> --remote

For a single investor school, such as A-share hot-money:

python3 run.py <ticker> --school F --no-browser

Agent Rules

  1. Treat scripts as data and scoring tools, not as final analyst judgment.
  2. Do not invent numbers. Use script outputs, cached JSON, or current public evidence.
  3. For serious deep-analysis requests, complete the agent review loop described in skills/deep-analysis/SKILL.md before final report assembly.
  4. For hot-money analysis, apply LHB seat matching and is_in_range() before making a short-term judgment.
  5. For trap detection, scan all eight signals and include concrete evidence when risk is non-trivial.
  6. For report template or UI changes, update tests, version metadata, and release notes together.
Gespeicherte Skills