A prompt pack that teaches AI agents to hunt supply-chain bottlenecks

What it does

Serenity-skill is a prompt-and-workflow kit for AI coding agents (Claude Code, Codex, OpenAI Agent Skills). It guides the agent to decompose a hot sector—say, AI semiconductors or robotics—into downstream demand, components, materials, and infrastructure, then rank which layer is actually supply-constrained. The output is a prioritized research list with evidence checks, not stock picks.

The interesting bit

The method is borrowed from a Chinese financial blogger, Serenity (@aleabitoreddit), whose public posts trace how big tech rallies tend to concentrate in the hardest-to-expand links of a chain. The skill formalizes that intuition into a repeatable agent workflow: find low supplier counts, long qualification cycles, and hard-to-substitute inputs, then challenge whether a given stock really sits near that bottleneck or is just riding the theme.

Key highlights

• Ships as a SKILL.md file plus prompt templates, reference docs, and a Python scorecard script—no model training, no API calls.
• Includes explicit “evidence ladder” rules: social media is a clue, but final judgments must cite filings, exchange announcements, earnings calls, or regulatory docs.
• Covers both individual stocks and thematic funds/ETFs, checking whether holdings actually cluster near the bottleneck layer.
• Provides install paths for Claude Code, Codex, Hermes, and generic AgentSkills-compatible clients.
• MIT licensed; 636 stars, mostly Chinese-language docs with English README available.

Caveats

• The project is essentially a curated prompt pack and methodology document; the Python scripts are lightweight helpers (template generation, validation).
• All examples and documentation are China-market oriented (A-shares, Chinese exchange filings); adapting to other markets requires user effort.
• The README repeatedly notes this is “research support only” and disclaims trading decisions, which suggests the authors are wary of regulatory or liability exposure.

Verdict

Worth a look if you already use Claude Code or Codex and want a structured, repeatable way to interrogate thematic hype—especially in Chinese tech sectors. Skip it if you expect automated trading signals or deep quantitative modeling; this is a research discipline, not a strategy engine.