pinescript
Pine Script v6: syntax, performance, error diagnosis, backtesting, visualization. Use when writing or debugging `.pine` files or TradingView Pine indicators/strategies.
iliaal · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add iliaal/ai-skills --skill pinescript技能说明
name: pinescript
class: language
description: >-
Pine Script v6: syntax, performance, error diagnosis, backtesting,
visualization. Use when writing or debugging .pine files or TradingView
Pine indicators/strategies.
paths: "**/*.pine"
Pine Script Development
Verify before implementing: For Pine Script version-specific syntax or new built-in functions, look up current docs via Context7 (query-docs) before writing code. TradingView updates Pine Script frequently and training data may be stale.
Critical Syntax Rules
- Keep simple ternaries readable; multiline expressions require valid continuation indentation. For complex ternaries, use intermediate variables:
isBull = close > open barColor = isBull ? color.green : color.red - Continuation lines outside parentheses MUST be indented by a non-multiple of 4 -- same indentation as the start errors, and 4/8/12 spaces parse as a local block and error too (2 spaces is the conventional choice). Inside parentheses (function calls, parenthesized expressions) any indentation works, including multiples of 4
- NEVER use plot() inside local scopes (if/for/functions) -- use conditional value instead:
plot(condition ? value : na) - Use
barstate.isconfirmedwhen signals require the chart bar's closing values. It does not establish that requested higher-timeframe values are confirmed; inspectrequest.security()offsets and lookahead separately.
Platform Limits
Check the current platform limits before sizing a script: 64 plot counts (one call can consume several); up to 500 line, box, or label IDs each and 100 polyline IDs; 40 unique request.*() calls, or 64 on Ultimate; 100,000 compiled tokens. History buffers, requested intrabars, and chart history have distinct limits; there is no general 500-bar request.security() history limit.
- Drawings positioned with
xloc.bar_indexreach at most 9,999 bars into the past and 500 into the future; for anything older, switch the drawing toxloc.bar_timeand pass a timestamp (a time value withoutxloc.bar_timeis treated as a future bar index and errors) - Set the relevant
max_*_countdeclaration parameter and cap growth with a rolling buffer: push each object, thenline.delete(arr.shift())after the intended line count is exceeded. The default display count is approximately 50 per drawing type.
Performance
- Tuple security calls -- one
request.security()returning[close, high, low]instead of 3 separate calls - Pre-allocate arrays with
array.new<type>(size)instead of push-and-resize - Short-circuit signals: build conditions incrementally, exit early when first condition fails
- Cache repeated calculations in variables -- Pine recalculates every bar
- Iterate collections with
for item in myArray(orfor [i, item] in myArray) instead offor i = 0 to array.size(...) - 1-- the indexed form re-evaluates the bound each pass and breaks when the loop mutates the array's size - Model related values as a user-defined type, not parallel arrays:
type Tradewithfloat entry,int startBar, plusmethodfunctions, stored in onearray<Trade>. Parallel arrays (entries,startBars, ...) desync on any missed push/remove and every operation must be repeated per array; one typed array keeps each object's fields together
Debugging
Use Pine Logs through log.info(), log.warning(), and log.error(), plus these visual checks:
- Label debugging:
label.new(bar_index, high, str.tostring(myVar))to inspect values; cap retained labels explicitly. - Table monitor:
table.new()withbarstate.islastfor real-time variable dashboard - Debug mode toggle: use
if input.bool(false, "Debug")for local debug code; keep plot calls global. - Repainting checks: record live signals with timestamps, then compare the same bars after reload.
value[1]refers to the preceding bar and does not detect revisions to an earlier calculation.
Strategy & Backtesting
- Use
strategy.*functions:strategy.wintrades,strategy.losstrades,strategy.grossprofit - Drawdown tracking:
maxEquity = math.max(strategy.equity, nz(maxEquity[1])), thendd = (maxEquity - strategy.equity) / maxEquity * 100 - Estimate annualized Sharpe from mean excess returns divided by their standard deviation, scaled by the square root of periods per year; state the sampling interval and annualization assumptions and handle zero variance.
- Walk-forward validation -- optimize on period 1, test on period 2, re-optimize on period 2, test on period 3. Compare degradation against sampling uncertainty, costs, and regime changes; no universal percentage establishes overfitting.
- Indicator accuracy testing -- at bar
t, scoreprediction[horizon]against the now-realized outcome, such asclose > close[horizon], excluding warmup bars. Positive offsets reference the past, never future bars; see history referencing. - Count evaluations per slice -- a slice scored N times during tuning is tuning data, whatever it is labelled, so a multi-parameter sweep run across every slice turns the "validation" numbers into selection bias. Reserve at least one slice with an explicit look budget, spend it after the parameters are locked, and treat "one more look" as the signal to stop
- Conflicting per-slice optima indicate instability -- compare a robust fixed parameter with a simpler strategy before adding a regime classifier. Fit any classifier using information available before entry and validate it on untouched data; conflicting optima alone do not prove that every fixed parameter fails.
- Re-run every parameter sweep with the regime gate active -- pre-gate sweeps do not transfer, because losing ungated sessions mask the parameter's real ef