trading-decision-engine

name

trading-decision-engine

description

Use when users need productized trading decision support, conditional buy/sell-point style planning, Vibe-Trading backtest validation, Alpha Zoo factor screening, or full-market stock monitoring through this workspace. This skill routes through financial-services-skill-router, china-market-overlay for China securities, and trading_core CLI adapters to produce source-backed decision_card, conditional_trade_plan, execution feasibility checks, source logs, QA status, and Not investment advice without target prices, ratings, personal position sizing, return promises, or unconditional trade instructions.

Trading Decision Engine

Use this skill for daily trading decision-support workflows that need the project-native trading_core adapter, including HKUDS/Vibe-Trading bridge capabilities.

Required Order

1. financial-services-skill-router
2. .agents/SKILLS_INDEX.md
3. china-market-overlay when the security is A-share, 港股, China-linked, iFinD, 同花顺, or AKShare related
4. a-share-equity-research-workflow for A-share single-stock schema rules
5. a-share-research-product-workflow for decision-support packaging
6. This skill
7. financial-output-qa-gate before delivery or persistence

Workflow Recipes And References

• Use .agents/workflows/a_share_decision_support.json for conditional decision cards.
• Use .agents/workflows/vibe_backtest_validation.json for backtest validation.
• Use .agents/workflows/alpha_factor_bench.json for Alpha Zoo screening.
• Use .agents/references/output-contract.md for shared output fields.
• Use .agents/references/data-capability-registry.md for source priority and fallback disclosure.
• Use .agents/references/display-profiles.md for App/CLI/JSON presentation.
• Use .agents/references/local-research-artifacts.md for .research/ paths.

Core Commands

Use these commands from the project root when deterministic artifacts are needed:

python3 -m trading_core.cli decision --ticker 300033.SZ --market a_share --horizon 20d --mode conditional_strong
python3 -m trading_core.cli backtest --ticker 300033.SZ --strategy technical_breakout --start 2024-01-01 --end 2026-05-23
python3 -m trading_core.cli alpha-bench --universe csi300 --zoo gtja191 --period 2021-2026
python3 -m trading_core.cli review --path .research/runs/decision_support.jsonl --horizon 20

For local OHLCV files, add --ohlcv path/to/ohlcv.csv or --ohlcv path/to/ohlcv.json.

CLI output defaults to Markdown cards for human review. Add --format json for the full machine-readable schema, and add --output path.json when the full record should be saved while the terminal remains readable.

App Conversation Display

When presenting results directly in the Codex app conversation, read references/app-conversation-display.md and use the card-style Markdown layout: one-line orientation, decision card, conditional plan table, evidence matrix, execution constraints, next checks, and compact data/QA section. Keep the same source quality and guardrails as the JSON schema; do not show raw JSON unless the user asks for it.

Output Rules

• decision_card is the concise user-facing summary.
• conditional_trade_plan may include concrete trigger_level, invalidation_level, and risk_control_level, but each level must be conditional, source-backed, and marked as not a target price.
• A-share outputs should include execution_feasibility for T+1, price limits, 100-share lots, costs, and slippage assumptions.
• backtest_validation is evidence only. Do not convert factor or backtest results into standalone buy/sell recommendations.
• Every investment-related output must include Not investment advice, source logs, missing data, and QA status.
• Run the final record through financial-output-qa-gate or tools/check_research_integrity.py --profile a_share_decision before treating it as productized output.

Guardrails

• Do not invent market prices, financials, events, factors, or backtest results.
• Do not output target prices, buy/sell ratings, personal position sizing, or return promises.
• Do not issue unconditional trade instructions. Use conditional review language with triggers, invalidation, and risk controls.
• If current market data is unavailable, output source gaps and a framework rather than filling values.