Strategy design

Strategy Design

Design an execution plan for: $ARGUMENTS

Overview

• Implementation status: code-backed
• Local entry script: <bundle-root>/strategy-design/run.py
• Primary purpose: turn an accepted market view into a conditional execution style, actionable zones, and a monitoring checklist
• Workflow stages: stage 4 Feature Engineering & Signal Construction, stage 6 Risk Management & Position Sizing, and stage 7 Live Trading & Monitoring
• Local executor guarantee: produce a baseline strategy plan from the current snapshot and optional strategy preset
• Agent-required overlay: verify that the plan is appropriate for liquidity, price-limit behavior, cost assumptions, holding horizon, and monitoring discipline

Use When

• The user wants execution style rather than only a trade action.
• The user asks for entry, stop, target, and holding-horizon structure.
• The user wants a practical execution and monitoring checklist around the current market state.

Do Not Use When

• The user only wants a trading action and quantity. Use decision-support.
• The user wants a full report bundle. Use market-brief or analysis.
• The user wants to know whether the name belongs in the portfolio at all. Use decision-support or analysis first.
• The user expects a live-execution algorithm or broker integration. This skill does not provide that.

Inputs

• Normal case: one stock symbol.
• Optional --csv PATH: use a local CSV instead of the default market source.
• Optional --strategy STRATEGY_ID: force a migrated strategy preset.
• Optional --style auto|breakout|trend_pullback|range_trade|defensive.
• Optional --hold-days N, --capital N, --risk N.
• Optional --start, --end, --source.
• If symbol is omitted, the skill may reuse last_symbol from the same execution context.
• Assumption note:
  • style, hold-days, capital, and risk defaults materially affect the plan
  • the agent should say which were explicit and which were defaulted

Execution

Step 1: Confirm the planning boundary

Use strategy-design only after the investment judgment is accepted or at least conditionally accepted. If the user is still asking whether to own the name, route back to decision-support or analysis.

Step 2: Run the local executor

python3 <bundle-root>/strategy-design/run.py <symbol> [--strategy STRATEGY_ID] [--style STYLE] [--hold-days N]

Step 3: Validate execution realism around the baseline plan

The agent should explicitly review:

• regime and style fit
• price-limit risk for A-share execution
• gap and stop-slippage risk
• whether the expected position size appears too large relative to likely liquidity
• whether the holding period makes sense given signal decay and event risk
• whether the plan depends on a deterministic preset or an advisory-style framework

Step 4: Deliver a complete execution plan

When the user needs more than the raw local output, the skill should structure the answer into:

• strategy family and rationale
• entry conditions and entry zone
• stop logic and invalidation triggers
• target or profit-taking logic
• holding horizon and review cadence
• execution-realism notes and known non-modeled risks

Step 5: Define monitoring and handoff rules

The plan should identify what to watch next and which downstream skill should handle the next stage:

• paper-trading for simulated execution
• backtest-evaluator for historical validation of similar rules
• analysis-history or reports for prior comparable runs

Output Contract

• Minimum local executor output: human-readable text beginning with 策略设计.
• Core fields: strategy id or display name when applicable, holding days, entry zone, stop zone, take-profit zone, position percentage, checklist items, and risk notes.
• Side effects: updates session memory for the current execution context.
• Local executor guarantee:
  • baseline plan generation from the current snapshot and preset logic
  • conditional style selection when auto is used
• Agent-required delivery standard:
  • disclose style, preset, hold-period, capital, and risk assumptions
  • state whether the plan comes from deterministic preset logic or higher-level manual framing
  • treat zones as planning anchors rather than guaranteed fills
  • surface non-modeled risks when relevant, especially opening gaps, limit-up or limit-down behavior, liquidity, stop slippage, and event timing
  • keep execution mechanics separate from the question of whether the position should exist at all

Failure Handling

• Parse and argument errors: non-zero exit with a readable 命令错误 message.
• Data-loading or normalization errors: readable failure text beginning with 执行失败:.
• Missing symbol with no reusable session symbol: readable guidance instead of a traceback.
• If execution realism cannot be assessed from the available evidence, the plan should still return with explicit caveats rather than false precision.

Key Rules

• Treat the result as a conditional execution plan, not as an automatic order.
• Keep the style aligned with the observed market regime unless the user explicitly overrides it.
• Avoid false precision: the plan should acknowledge level slippage, price-limit behavior, and tape risk when they matter.
• If the user wants portfolio approval, route back to decision-support or analysis.

Composition

• Often follows market-analyze, decision-support, or analysis.
• Frequently appears as the planning section inside market-brief.
• Natural upstream companion to paper-trading.