tw-trading-agentslisted

# Taiwan-stock TradingAgents (multi-agent) Applies the TradingAgents methodology (arXiv:2412.20138) — "simulate a real brokerage's division of labor + bull/bear debate + tiered risk control" — to the Taiwan market, with **FinMind** as the data source. Several analyst sub-agents analyze different dimensions in parallel; their reports flow through a bull/bear debate, a trader decision, and a risk-management review, after which the orchestrator (you) assembles an investment-research report containing a Buy/Sell/Hold call, a suggested position size, and a risk disclaimer. ## Cross-CLI compatibility (important) This skill uses a **fan-out / fan-in** pattern: each phase dispatches multiple sub-agents in parallel; each sub-agent analyzes its slice and **returns its report as its final output**; the orchestrator collects the reports and moves to the next phase. Sub-agents never communicate with each other directly. - The mechanism for dispatching sub-agents is **platform-specific** — see `references/platform-tools.md`. The one-line rule: - **Parallel when possible** (Claude Code: multiple `Task` calls in one message; Codex: `spawn_agent`); - **Sequential when not** (Gemini / Copilot: the main model plays each role in turn, producing the same structured intermediate output). - No team abstraction, no shared config file, no message bus, no manual shutdown — sub-agents "return as output." ## Architecture ``` Orchestrator (you) — doubles as Research Manager (debate