paper-trading-pre-market-plan
Skill0 starsChanged 2 months agoApache-2.0View on GitHub
Use when a workshop agent is preparing an A-share paper-trading pre-market plan, especially before market open, before a simulated trading day, or when it must define observation priorities, no-trade conditions, risk boundaries, and if-then triggers for the paper account.
In this repository20 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-browserSkill
- ai-knowledge-creatorSkill
- docxSkill
- frontend-designSkill
- harness-quality-controlSkill
- openzhiyu-apiSkill
- openzhiyu-feature-guideSkill
- openzhiyu-memorySkill
- paper-trading-intraday-checkSkill
- paper-trading-post-market-reviewSkill
- paper-trading-trend-followingSkill
- pdfSkill
- pptxSkill
- single-etf-position-advisorSkill
- single-etf-timing-advisorSkill
- ui-ux-pro-maxSkill
- watchlist-selection-controlSkill
- weatherSkill
- xlsxSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
27 rules. About 0 are common, 18 are distinctive, and 9 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈18
Distinctive · estimate
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 10
Use this for the 开盘前观察计划 loop or any similar A-share paper-trading pre-open planning task.
- Distinctive: line 14
Treat the paper account as a controlled experiment.
- Distinctive: line 15
Do not predict the day. Prepare conditional responses.
- Distinctive: line 16
Prefer no operation unless current account state, market regime, watchlist evidence, risk/reward, and boundary policy all agree.
- Distinctive: line 17
Never let news, one quote, or a single fund-flow number become an action by itself.
- Distinctive: line 18
Use yesterday's review and active workshop memory as the state estimate.
- Distinctive: line 23
Call tools in this order when available and within budget:
- Distinctive: line 26
Read cash, total asset, realized PnL, positions, available quantity, open orders, fills, and trading enabled state.
- Distinctive: line 27
Stop if account data is missing, stale, or internally inconsistent.
- Distinctive: line 30
Read current watchlist and quote snapshot.
- Distinctive: line 34
Read active and recent trade plans before creating new triggers.
- Distinctive: line 38
Read active trading rules, yesterday's plan, known mistakes, position-specific triggers, and source limitations.
- Distinctive: line 43
Use breadth, limit-up/limit-down, hot themes, and large-style direction.
- Distinctive: line 51
Use phase, trendScore, MA20/MA60 structure, ATR stop, extension, and invalidation as the primary method layer.
- Distinctive: line 54
Use for holdings and high-priority watchlist names.
- Distinctive: line 55
Treat fund-flow degradation or source conflicts as a risk flag, not as proof.
- Distinctive: line 58
Use for held positions, open-order candidates, abnormal moves, or major planned decisions.
- Very specific: line 63
Before concluding that any price, stop, invalidation, breakout, cash, position-size, profit-protection, or risk/reward rule has triggered …
- Very specific: line 66
Use stable rule ids such as 000977_buy_invalid_line, 159278_profit_stop, or cash_position_cap.
- Very specific: line 67
Every actionable trigger in preMarketPlan.ifThenTriggers[] must carry ruleEvidence with tool: "quantRuleEvaluate", ruleId, actual, …
- Very specific: line 68
If quantRuleEvaluate evidence is missing, write the plan as pending_tool_check and do not claim that a trigger, stop, or invalidation fired.
- Distinctive: line 100
Do not create a watchlist change proposal.
- Very specific: line 104
Default action is no operation.
- Very specific: line 130
Before finishing, call workshopLogEvent with these exact title prefixes:
- Very specific: line 137
Use workshopWriteMemory only for reusable state:
- Distinctive: line 145
Do not write routine summaries as memory unless they contain a reusable rule or a next-run trigger.
- Very specific: line 147
If the run creates, revises, or carries forward a concrete plan, call quantTradePlanUpsert before finishing.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Paper Trading Pre-Market Plan
This skill turns the pre-market run into a control-plan pass, not a prediction pass. The agent should define what it will observe, what would invalidate action, and what must remain untouched.
Use this for the 开盘前观察计划 loop or any similar A-share paper-trading pre-open planning task.
Core Philosophy
- Treat the paper account as a controlled experiment.
- Do not predict the day. Prepare conditional responses.
- Prefer no operation unless current account state, market regime, watchlist evidence, risk/reward, and boundary policy all agree.
- Never let news, one quote, or a single fund-flow number become an action by itself.
- Use yesterday's review and active workshop memory as the state estimate; verify high-impact memories with current data before relying on them.
- Respect A-share constraints: T+1, price-limit regimes, lunch break, closing auction, lot-size constraints, and delayed correction after opening mistakes.
Required Observation Order
Call tools in this order when available and within budget:
quantPaperGetAccount- Read cash, total asset, realized PnL, positions, available quantity, open orders, fills, and trading enabled state.
- Stop if account data is missing, stale, or internally inconsistent.
quantPaperGetWatchlist- Read current watchlist and quote snapshot.
- Stay inside current watchlist. This skill does not discover or propose watchlist changes.
quantTradePlanList- Read active and recent trade plans before creating new triggers.
- Review whether prior plans are still valid, superseded, blocked, or due today.
workshopReadMemoryorworkshopSearchMemory- Read active trading rules, yesterday's plan, known mistakes, position-specific triggers, and source limitations.
- For high-impact memories, use
workshopGetMemoryEvidenceif uncertainty could change an action.
aStockMarketMood- Classify the market into one of:
red,yellow,green, orunknown. - Use breadth, limit-up/limit-down, hot themes, and large-style direction.
- Classify the market into one of:
aStockQuote- Refresh holdings first, then watchlist focus names.
- Current price data beats old workshop events.
aStockTrend- Required for held positions, open-order candidates, and focus names that may become action candidates.
- Use phase, trendScore, MA20/MA60 structure, ATR stop, extension, and invalidation as the primary method layer.
aStockSignals- Use for holdings and high-priority watchlist names.
- Treat fund-flow degradation or source conflicts as a risk flag, not as proof.
aStockNewsAndFilings- Use for held positions, open-order candidates, abnormal moves, or major planned decisions.
- News is supporting evidence only.
Numeric Rule Discipline
Before concluding that any price, stop, invalidation, breakout, cash, position-size, profit-protection, or risk/reward rule has triggered or failed, batch the comparisons and call quantRuleEvaluate.
- One model response may contain many numeric checks. Put them in one
quantRuleEvaluate.rules[]call when possible instead of doing mental arithmetic. - Use stable rule ids such as
000977_buy_invalid_line,159278_profit_stop, orcash_position_cap. - Every actionable trigger in
preMarketPlan.ifThenTriggers[]must carryruleEvidencewithtool: "quantRuleEvaluate",ruleId,actual,operator, threshold or range,triggered, andcomparisonText. - If
quantRuleEvaluateevidence is missing, write the plan aspending_tool_checkand do not claim that a trigger, stop, or invalidation fired.
Market State Model
Assign exactly one state:
red
- Broad market is in retreat, limit-down risk is elevated, hot themes are collapsing, data sources are unreliable, or account risk is already above boundary.
- Output: no new buy. Only plan defensive checks.
yellow
- Market is mixed, rotating, overextended, data is partially degraded, or signals conflict.
- Output: observe. New buys require unusually clean evidence and small size.
green
- Breadth, leading themes, liquidity, and watchlist behavior are aligned.
- Output: allow conditional triggers, but still require risk/reward and position rules.
unknown
- Data freshness or source quality is insufficient.
- Output: no operation; log the data gap.
Watchlist Classification
Classify each relevant symbol:
leader: stronger than market and sector, liquidity acceptable, not extended far beyond a valid base or plan.constructive: improving but still waiting for confirmation.laggard: underperforms its sector/market or has weak relative strength.risk: has source conflict, negative event, stop-risk, liquidity issue, or gap risk.ignore_today: not relevant to today's control objective.
Do not create a watchlist change proposal. That belongs to the watchlist-selection workshop.
Pre-Market Action Rules
Default action is no operation.
Allow a planned simulated buy only if all gates pass:
- Account gate: cash, available quantity, open orders, and trading enabled state are known.
- Market gate: market state is not
redorunknown. - Symbol gate: target is in current watchlist.
- Evidence gate: price/volume, relative strength, fund flow, and news context do not conflict materially.
- Trend gate: trendScore is normally at least 65, trend state is
strong_uptrend,early_breakout, orconstructive_pullback, and the stop/invalidation line is explicit. - Risk gate: planned stop is explicit and risk/reward is at least 1:2.
- Position gate: position size will not concentrate the account or violate existing workshop memory.
- Execution gate: every planned buy must name a
plannedPriceand an execution tolerance. Default tolerance isplannedPrice + 2%; use a limit order and explain why it is not chasing the open.
Block action when:
- Data is stale, missing, or contradictory.
- The plan depends on news that is unverified, promotional, old, or not material.
- The symbol is outside the watchlist.
- The action was invented in this run without prior setup.
- Opening price has moved enough that the original risk/reward is gone.
- Current executable buy price is more than 2% above the planned entry price and no fresh risk/reward plan has replaced it.
- Trend state is
range,downtrend,broken, orextendedwithout a fresh breakout thesis. - The model is trying to "make something happen" because a loop ran.
Required Visible Output
Before finishing, call workshopLogEvent with these exact title prefixes:
marketScanSummary: ...trendStateReview: ...riskAssessment: ...actionTaken: ...
Use workshopWriteMemory only for reusable state:
- a concrete trigger plan that future runs must check,
- a source-quality limitation,
- a trading-rule correction,
- a mistake or boundary lesson,
- a position-specific invalidation rule.
Do not write routine summaries as memory unless they contain a reusable rule or a next-run trigger.
If the run creates, revises, or carries forward a concrete plan, call quantTradePlanUpsert before finishing. Each plan must include planDate, code, action, triggerCondition, invalidation, rationale, and dueAt when known.
Final JSON Contract
The final <structured-output> JSON must include:
{
"marketScanSummary": "market, account, holdings, watchlist and evidence summary",
"riskAssessment": "risk state, data quality, invalidation conditions, and why action is or is not allowed",
"actionTaken": "simulated order submitted, no operation, or blocked with reason",
"tradePlanLedger": "prior plans reviewed and new plans written with quantTradePlanUpsert",
"preMarketPlan": {
"marketState": "red|yellow|green|unknown",
"focusSymbols": [],
"noTradeConditions": [],
"ifThenTriggers": []
}
}Reference Basis
For deeper rationale, read references/methodology.md. Use it only when the task needs method explanation or rule revision.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

