paper-trading-intraday-check
Skill0 starsChanged 2 months agoApache-2.0View on GitHub
Use when a workshop agent is running an A-share paper-trading intraday check, especially low-frequency market-hours inspection of account risk, holdings, current watchlist, trigger conditions, simulated order eligibility, no-trade reasons, and paper-trading action boundaries.
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-post-market-reviewSkill
- paper-trading-pre-market-planSkill
- 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
38 rules. About 0 are common, 23 are distinctive, and 15 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈23
Distinctive · estimate
This author's own choices. Worth a look.
15
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 盘中模拟盘巡检 loop or any similar A-share market-hours paper-trading inspection.
- Distinctive: line 18
Do not become a watchlist hunter.
- Distinctive: line 19
Do not trade because the market is exciting, because a symbol is moving, or because the loop ran.
- Distinctive: line 23
Call tools in this order when available:
- Distinctive: line 26
Read total asset, cash, positions, available quantity, open orders, fills, and realized PnL.
- Distinctive: line 27
If open orders exist, inspect whether they are still valid before considering new orders.
- Distinctive: line 30
Read current watchlist with fresh quotes.
- Distinctive: line 31
Never add, remove, replace, or propose watchlist changes in this skill.
- Distinctive: line 34
Read active plans before deciding whether a trigger fired.
- Distinctive: line 35
Treat due plans as hard prior state.
- Distinctive: line 38
Read active position rules, stop lines, prior mistakes, pre-market triggers, and source-quality warnings.
- Distinctive: line 46
Use phase, trendScore, MA20/MA60 structure, ATR stop, extension, and invalidation before deciding.
- Distinctive: line 49
Use for held positions and symbols whose trigger is near.
- Distinctive: line 53
Check market breadth, hot themes, limit-up/limit-down pressure, and whether the market is becoming overheated.
- Distinctive: line 56
Use only for abnormal moves, held positions near action thresholds, or potential trade decisions.
- Very specific: line 61
Before concluding that any price, stop, invalidation, breakout, cash, position-size, profit-protection, or risk/reward rule has triggered …
- Very specific: line 64
Do not let the model compare numbers in prose.
- Very specific: line 65
Every trendFollowDecision.candidateDecisions[] entry with buy, sell, reduce, exit, tighten_stop, blocked, invalidated, or triggered must …
- Very specific: line 66
If rule evidence is missing, the only allowed conclusion is blocked_pending_rule_check.
- Distinctive: line 70
Every action must pass all gates.
- Very specific: line 101
Do not buy in risk_off or unknown.
- Distinctive: line 107
Stop-loss checks come before opportunity checks.
- Distinctive: line 109
If stop, invalidation, or risk-reduction conditions are met, reduce or exit promptly when T+1 availability permits.
- Distinctive: line 110
If the position is profitable but extended, protect gains with a tighter stop or partial reduction.
- Very specific: line 114
Prefer a pre-market, ledger, or memory-defined trigger.
- Very specific: line 120
Treat plannedPrice as the execution anchor.
- Very specific: line 121
Prefer 8%-15% initial capital for clean signals, 5%-8% when market or data state is mixed, and normally cap one symbol at 18%.
- Distinctive: line 125
Use relative strength as a filter:
- Distinctive: line 143
If news is unverified or stale, do not let it change the action from no operation to buy/sell.
- Very specific: line 147
Only call quantPaperPlaceOrder when all of the following are true:
- Very specific: line 153
every buy order includes plannedPrice and normally maxBuyDeviationPct=3,
- Very specific: line 154
every trend-following order includes tradeThesis with trendState, trendScore, stopPrice, invalidation, and same-run evidence when available,
- Distinctive: line 159
If the agent writes "准备买入", "计划减仓", "下次验证", or similar language without placing an order, it must log the exact blocking reason and the …
- Very specific: line 163
After checking orders, fills, or trigger outcomes, call quantTradePlanReview for every due plan whose status changed or whose blocker is …
- Very specific: line 167
When aStockTrendSystem or aStockTrend is used, the intraday result must include trendFollowDecision.
- Very specific: line 188
Before finishing, call workshopLogEvent with these exact title prefixes:
- Distinctive: line 196
Write memory only when the run creates reusable knowledge:
- Distinctive: line 205
Do not create outbox drafts for routine intraday observations.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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 Intraday Check
This skill turns intraday work into trigger verification, not market chasing. The agent should react only to previously defined or objectively validated conditions.
Use this for the 盘中模拟盘巡检 loop or any similar A-share market-hours paper-trading inspection.
Core Philosophy
- Intraday noise is the default, but this workshop is a simulated learning trader: controlled action is preferred over passive observation when gates are adequate.
- The first job is to protect the paper account from uncontrolled behavior.
- The second job is to verify whether pre-defined triggers fired.
- The third job is to record useful observations for later review.
- Do not become a watchlist hunter; only inspect current watchlist and current positions.
- Do not trade because the market is exciting, because a symbol is moving, or because the loop ran.
Required Observation Order
Call tools in this order when available:
quantPaperGetAccount- Read total asset, cash, positions, available quantity, open orders, fills, and realized PnL.
- If open orders exist, inspect whether they are still valid before considering new orders.
quantPaperGetWatchlist- Read current watchlist with fresh quotes.
- Never add, remove, replace, or propose watchlist changes in this skill.
quantTradePlanList- Read active plans before deciding whether a trigger fired.
- Treat due plans as hard prior state. A due plan must become executed, blocked, skipped, or still pending with an explicit reason.
workshopReadMemoryorworkshopSearchMemory- Read active position rules, stop lines, prior mistakes, pre-market triggers, and source-quality warnings.
- For a high-impact action, verify memory evidence if the memory is old or disputed.
aStockQuote- Refresh held positions and trigger candidates if the watchlist snapshot is stale or incomplete.
aStockTrend- Required for any held position or watchlist symbol near buy, add, reduce, sell, or hold thresholds.
- Use phase, trendScore, MA20/MA60 structure, ATR stop, extension, and invalidation before deciding.
aStockSignals- Use for held positions and symbols whose trigger is near.
- Interpret fund-flow failures, fallback data, or provider conflict as a reason to reduce confidence.
aStockMarketMood- Check market breadth, hot themes, limit-up/limit-down pressure, and whether the market is becoming overheated.
aStockNewsAndFilings- Use only for abnormal moves, held positions near action thresholds, or potential trade decisions.
- News adjusts confidence; it does not trigger trades alone.
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 run may need many comparisons. Put all current trigger, stop, buy-deviation, cash, quantity, and rotation checks into one
quantRuleEvaluate.rules[]call when possible. - Do not let the model compare numbers in prose. For example,
77.66 < 77.61is false and must be represented by tool output, not by reasoning text. - Every
trendFollowDecision.candidateDecisions[]entry withbuy,sell,reduce,exit,tighten_stop,blocked,invalidated, ortriggeredmust includeruleEvidencewithtool: "quantRuleEvaluate",ruleId,actual,operator, threshold or range,triggered, andcomparisonText. - If rule evidence is missing, the only allowed conclusion is
blocked_pending_rule_check; do not place, cancel, or mark a plan complete.
Intraday Decision Gates
Every action must pass all gates. If any gate fails, default to no operation.
1. Data Gate
Block action if:
- account or position data is missing,
- price data is stale,
- fund-flow source is degraded and the decision depends on it,
- current price conflicts with another same-run source,
- ETF or stock-specific fields are structurally unavailable and the rule needs them.
2. Boundary Gate
Block action if:
- the symbol is outside the current watchlist,
- the action is a real broker order,
- the action maintains the watchlist instead of trading the paper account,
- the action would exceed current workshop boundary policy,
- the order cannot be expressed as an auditable simulated limit order.
3. Market Gate
Classify intraday market state:
risk_off: broad weakness, limit-down pressure, hot themes collapsing, or high uncertainty.mixed: rotation or conflicting signals.risk_on: breadth and leading themes aligned.overheated: broad surge with chasing risk, high break-board risk, or extended watchlist moves.
Do not buy in risk_off or unknown. In mixed, allow smaller rotation or pilot buys when relative strength and stop distance are clean. In overheated, prefer rotation from weak holdings into stronger names only if the entry is still near plan.
4. Position Gate
For existing positions:
- Stop-loss checks come before opportunity checks.
- Available quantity matters under T+1.
- If stop, invalidation, or risk-reduction conditions are met, reduce or exit promptly when T+1 availability permits.
- If the position is profitable but extended, protect gains with a tighter stop or partial reduction; do not let "观察" replace a concrete protection action.
For new or add-on buys:
- Prefer a pre-market, ledger, or memory-defined trigger. A same-run fresh trigger is allowed if it is backed by
aStockTrendSystem, current quote, clear invalidation, and a writtenquantTradePlanUpsertreplacement plan. - Require a valid trend state:
strong_uptrend,early_breakout, orconstructive_pullback. - Require trendScore normally at least 60 and no
downtrendorbrokenstate.extendedblocks chasing, but does not block a tightly risk-managed breakout pilot when risk/reward remains valid. - Require relative strength versus sector and market.
- Require a nearby invalidation point.
- Require at least 1:1.5 risk/reward for pilot trades and 1:2 for larger adds.
- Treat
plannedPriceas the execution anchor. A buy can execute only when the current executable limit price is no more than 3% aboveplannedPrice, unless this same run creates a fresh plan and recalculates the risk/reward. - Prefer 8%-15% initial capital for clean signals, 5%-8% when market or data state is
mixed, and normally cap one symbol at 18%.
5. Relative-Strength Gate
Use relative strength as a filter:
- Market strong + sector strong + symbol weak = weakness divergence; do not add, and consider reducing it to fund a stronger watchlist candidate.
- Market weak + symbol strong + fund flow supportive = constructive, but still wait for a controlled entry.
- Symbol up sharply + fund flow negative = possible distribution; do not add and consider verification day.
- ETF strong after sharp prior decline = possible rebound. It can justify a small pilot only with a nearby invalidation line.
6. News Gate
Classify news reference:
material_positivematerial_negativeneutralstaleunverifiednot_checked
If news is unverified or stale, do not let it change the action from no operation to buy/sell.
Simulated Order Rules
Only call quantPaperPlaceOrder when all of the following are true:
- action side, code, quantity, limit price, stop/invalidation, and reason are explicit,
- the symbol is in the current watchlist,
- the same run has read current account and quote data,
- the order is consistent with prior plan or a hard risk rule,
- every buy order includes
plannedPriceand normallymaxBuyDeviationPct=3, - every trend-following order includes
tradeThesiswith trendState, trendScore, stopPrice, invalidation, and same-run evidence when available, - the buy limit price is within
plannedPrice * 1.03; if it is above that range, wait for pullback or replace the plan with a fresh risk/reward calculation before buying, - the action is not chasing a fast move without a newly calculated stop and target,
- a no-action alternative was considered and rejected with evidence.
If the agent writes "准备买入", "计划减仓", "下次验证", or similar language without placing an order, it must log the exact blocking reason and the next verification condition.
For this aggressive profile, "继续观察" is allowed only when there is a concrete blocker. If at least one held symbol is weak and one watchlist symbol is stronger, the default decision should be rotation, not observation.
After checking orders, fills, or trigger outcomes, call quantTradePlanReview for every due plan whose status changed or whose blocker is now known. If this run replaces the plan with a new trigger or price, call quantTradePlanUpsert and link the decision in sourceDecision.
Trend Decision Contract
When aStockTrendSystem or aStockTrend is used, the intraday result must include trendFollowDecision.
For each holding, open order, triggered symbol, and no-trade candidate, write one candidateDecisions[] entry with:
codepositionStateprofitStateavailableQuantitylifecycleStatetrendScorecontrolActiondecisionorderIntentorderIdorblockedReasonruleEvidencenextVerification
For break_warning, reduce_watch, exit_required, or broken, include breakWarningHandling. If the position is profitable, the default is profit protection, not vague observation: tighten the trailing stop, block adding, and reduce or exit only when the warning is confirmed, persistent, below stop/invalidation, or market state deteriorates.
Required Visible Output
Before finishing, call workshopLogEvent with these exact title prefixes:
marketScanSummary: ...trendStateReview: ...riskAssessment: ...tradeThesis: ...actionTaken: ...
Write memory only when the run creates reusable knowledge:
- trigger fired or failed,
- weak divergence repeated,
- distribution signal,
- data-source limitation,
- a mistake in earlier plan,
- a rule update that should affect future runs.
Do not create outbox drafts for routine intraday observations.
Final JSON Contract
The final <structured-output> JSON must include:
{
"marketScanSummary": "fresh account, holdings, watchlist, market and news reference",
"riskAssessment": "data quality, market state, position risk, trigger status, and boundary checks",
"actionTaken": "simulated order details or no operation with the exact blocking reason",
"tradePlanLedger": "plans read, executed/blocked/skipped updates, and any replacement plans written",
"trendFollowDecision": {
"marketState": "risk_off|mixed|risk_on|overheated|unknown",
"candidateDecisions": [],
"orders": [],
"blockedReasons": [],
"nextVerification": []
},
"intradayDecision": {
"marketState": "risk_off|mixed|risk_on|overheated|unknown",
"triggeredSymbols": [],
"blockedReasons": [],
"nextVerification": []
}
}Reference Basis
For deeper rationale, read references/methodology.md. Use it only when revising rules or explaining why this skill is conservative.
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.

