About

Imported from zeroes-ones/Skills (skills/14-finance/technical-signals-engineer/SKILL.md). Install upstream with npx skills add zeroes-ones/Skills --skill technical-signals-engineer. Copyright stays with the author (MIT).

Technical Signals Engineer

Name: technical-signals-engineer
Author: zeroes-ones

Portability target: Spec-level (runs on Claude Code, Copilot, Gemini CLI, Codex, Cursor). No vendor-specific frontmatter fields.

Compute technical indicators correctly and generate calibrated buy/sell/hold signals from multi-indicator combinations. This skill is the signal-generation engine for equity and ETF trading — it ingests OHLCV market data and produces structured, confidence-scored trading signals that downstream skills consume. Every indicator formula is mathematically verified. Every signal rule is backtest-validated. Covers SMA, EMA, RSI, MACD, Bollinger Bands, ATR, OBV, VWAP, and their combinations — ETF-aware for sector/product differences, stock-aware for corporate event adjustments.

RESEARCH_PREREQUISITE — Execute Before Any Output

This is a HARD GATE. Do not produce ANY output, code, strategy, design, or recommendation without completing this research.

Before you act, you MUST execute every applicable research step. Research-before-acting is the difference between professional work and amateur guessing:

#	Research Step	Why It Matters	Where to Look
RP1	Verify domain currency. Check for breaking changes, deprecations, new standards, or version shifts since the knowledge cutoff.	[STALE_RISK] Outdated advice breaks real systems. API deprecations, framework version bumps, and security advisory changes happen continuously. Outputting based on stale knowledge damages credibility and produces broken results.	Official docs, changelogs, GitHub releases, RFC tracker
RP2	Audit the system or codebase. Read relevant files. Understand existing patterns, constraints, and architecture before proposing changes.	[CONTEXT_VIOLATION] Solutions that ignore existing patterns create technical debt. A change that contradicts the established architecture is worse than no change — it introduces inconsistency that compounds over time.	Project files, configs, dependency manifests, existing tests
RP3	Cross-reference claims against authoritative sources. Every factual assertion needs a verifiable source. Mark each: [VERIFIED], [COMPUTED], or [ESTIMATED].	[HALLUCINATION_GUARD] Claims without sources are indistinguishable from hallucinations. The #1 cause of incorrect output is treating assumptions as facts. Source tagging prevents this.	Official documentation, peer-reviewed papers, RFCs, specifications
RP4	Identify known failure modes. Before recommending, list what commonly breaks. For each failure mode: trigger condition, detection signal, and mitigation.	[FAILURE_BLINDNESS] Every domain has known failure patterns. Output that doesn't address them is dangerously incomplete. If you cannot name 3+ failure modes for your recommendation, you don't understand it well enough to recommend it.	Domain post-mortems, incident reports, antipattern catalogs, error databases
RP5	Quantify impact in concrete units. Replace abstract claims ("faster," "better," "more scalable") with exact numbers, even if estimated.	[VAGUENESS_PENALTY] "Faster" is unverifiable. "Reduces p95 latency from 340ms to 120ms (±15ms)" is verifiable. Abstract adjectives hide ignorance behind confidence. Concrete numbers expose gaps.	Benchmarks, production metrics, pricing data, published performance data
RP6	Map side effects and downstream impacts. What else breaks? Which dependencies are affected? Which downstream consumers need updating?	[CASCADE_BLINDNESS] Changes to one component ripple outward. A fix in module A can break module B that depends on A's old behavior. Map the blast radius before acting.	Dependency graph, cross-skill coordination table, API consumers list
RP7	Verify against non-negotiable quality gates. What are the minimum quality bars for this domain (accessibility, security, performance, accuracy, compliance)?	[QUALITY_FLOOR] Every domain has minimum standards below which output is invalid regardless of functionality. Missing WCAG AA = broken. Leaking credentials = broken. Silent data loss = broken.	Domain standards, compliance frameworks, security baselines, accessibility guidelines
RP8	Declare explicit limitations and edge cases. What does this NOT handle? What are the known boundaries? What scenarios are explicitly out of scope?	[SCOPE_HONESTY] Declaring limitations is a feature, not an admission of weakness. It prevents misuse, sets correct expectations, and demonstrates true understanding. Every solution has boundaries — naming them is professional.	This SKILL.md, domain literature, edge case databases

If you skip any of these research steps, you are not producing quality output — you are guessing with confidence. Guessing wastes time, breaks systems, and destroys trust. The references, ground rules, and decision trees in this skill exist specifically to prevent guessing. Use them.

Compliance: Research must be executed before any substantial output. For each step, document findings inline in your response using [RESEARCHED] marker: [RESEARCHED: RP1 — Domain verified against changelog v2.4. No breaking changes since cutoff.]. Partial research = partial quality. Zero research = zero credibility.

🔄 Iterative Research Loop — Research at EVERY Decision Point, Not Just Entry

The RP1-RP8 cycle above is NOT a one-time gate. It fires continuously at every material decision point throughout the workflow:

Loop	When It Fires	What Re-research Validates
Loop 0: Pre-Action	Before producing ANY output, code, strategy, or recommendation	Domain currency, codebase audit, source verification, failure modes, quantified impact, side effects, quality gates, limitations
Loop 1: Mid-Action	At every adjustment, phase transition, scale-out, or significant state change	Has the context changed? Are the original assumptions still valid? Has new information invalidated the Loop 0 conclusions?
Loop 2: Pre-Exit	Before closing, handing off, escalating, or declaring completion	Is the deliverable complete by the quality gates defined in RP7? Are all limitations declared (RP8)? Have failure modes been addressed (RP4)?
Loop 3: Post-Action	After completion: compare expected vs. actual outcome	What was the efficiency ratio (actual / theoretical max)? What learnings emerged? What should be fed back into the pattern database for future decisions?

Integration into Core Workflow:

Every decision point in a skill's Core Workflow must be marked with:

[RESEARCH LOOP: Re-execute RP1-RP8 before proceeding to next phase]

This ensures the agent pauses to re-verify ALL research dimensions before making the next decision. A skill that only researches at entry and then operates on auto-pilot is a skill that makes decisions on stale context.

Markers for output: At each loop, the agent outputs: [RESEARCHED: Loop N — RP1-RP8 re-verified. Key delta from previous loop: ...]

Why this matters: A decision made in Loop 0 may be catastrophically wrong by Loop 2 because the context changed. Markets move. Requirements shift. Dependencies update. The research loop catches context drift before it becomes output error.

Compliance: Research must be executed before any substantial output AND re-executed at every decision point. For each research loop, document findings inline. Partial research = partial quality. Zero research = zero credibility. Stale research = dangerous confidence.

Technical Domain Extension — Execute These ADDITIONAL Research Steps

#	Research Step	Why It Matters	Where to Look
RP-F1	Validate multi-timeframe confluence. Check: weekly trend, daily trend, 4-hour momentum alignment. A daily buy signal against a weekly downtrend has a ~35% false signal rate vs. ~15% when aligned.	[TIMEFRAME_CONFLICT] Single-timeframe signals are noise. A golden cross on the 15-minute chart means nothing if the weekly is in a death cross. Multi-timeframe alignment is the cheapest signal quality filter.	Multi-timeframe dashboard, false signal database by timeframe combination
RP-F2	Calculate the false signal rate for each indicator in current regime. RSI oversold signals have ~40% false positive rate in downtrends vs. ~15% in uptrends. MACD crossovers generate ~60% more false signals in low-VIX environments.	[FALSE_SIGNAL_COST] Every false signal costs: spread + commission + opportunity cost of being in the wrong position. At $5/trade with 40% false signals on 100 signals/year, that's $200/year in false-signal commissions alone — 0.4% drag on a $50K account.	Backtest signal database, regime-specific performance metrics
RP-F3	Quantify indicator lag. Moving average crossovers lag price by MA_period/2 on average. A 50-day SMA crossover signal is ~25 days late. MACD (12/26/9) introduces ~9-13 periods of lag.	[LAG_PENALTY] Lag transforms "buy low" into "buy after the move already happened." The profit left on the table by indicator lag is often larger than the profit captured by the signal.	Indicator lag calculations, lead-lag analysis against price
RP-F4	Detect the current regime before applying indicators. Trend-following indicators (MACD, moving averages) fail in ranges. Mean-reversion indicators (RSI, Bollinger Bands) fail in trends. Applying the wrong indicator family to the current regime destroys alpha.	[REGIME_MISMATCH] The #1 misuse of technical analysis: applying trending indicators to a ranging market (whipsaw losses) or mean-reversion indicators in a trending market (fading a freight train). Regime detection FIRST, indicator selection SECOND.	Pattern Recognition Engine §Regime Detection, ADX readings, volatility regime classification
RP-F5	Backtest each signal against out-of-sample data. A signal that worked in 2020-2023 may fail in 2024-2026. Markets adapt. Walk-forward testing reveals signal decay.	[OVERFITTING] Technical indicators have parameters. Optimizing parameters on historical data without out-of-sample validation is curve-fitting. The optimal RSI period for 2020 is not the optimal period for 2025.	Walk-forward backtest framework, parameter stability analysis

Route the Request

Auto-Route (No User Input Required)

#	Condition	Action
A1	`file_contains(".py", "SMA\|EMA\|moving_average\|RSI\|rsi\|MACD\|macd\|bollinger\|ATR\|OBV\|VWAP")` AND `file_contains(".py", "signal\|buy\|sell\|crossover\|divergence\|overbought\|oversold")`	This is your skill. Jump to Core Workflow — Phase 1.
A2	`file_contains(".py", "BlackScholes\|implied_volatility\|delta\|gamma\|greeks")` OR `file_contains(".py", "options\|strike\|expiration\|put_call")`	Invoke quantitative-analyst instead. Options pricing domain.
A3	`file_contains(".py", "alpaca\|broker\|order\|execution\|backtrader\|zipline")` AND `file_contains(".py", "submit\|fill\|position\|stop_loss\|take_profit")`	Invoke algorithmic-trader instead. This is execution, not signal generation.
A4	`file_contains(".py", "Polygon\|polygon\|kafka\|KafkaConsumer\|websocket.stream")` AND NOT `file_contains("*.py", "SMA\|RSI\|MACD")`	Invoke market-data-engineer instead. Data ingestion, not signal computation.
A5	`file_contains("*.py", "PE\|eps\|revenue\|DCF\|balance_sheet\|income_statement\|free_cash_flow")`	Invoke fundamental-analyst instead. Fundamental valuation domain.
A6	`file_contains(".py", "sklearn\|tensorflow\|torch\|RandomForest\|XGBoost\|LSTM\|transformer")` AND `file_contains(".py", "predict\|classify\|signal")`	Invoke ml-engineer instead. ML-based prediction, not rule-based signals.

Intent Route


What technical analysis task?
├── Computing individual indicators → Jump to Core Workflow Phase 1
├── Generating buy/sell signals → Jump to Core Workflow Phase 2
├── Building multi-indicator confirmation → Jump to Core Workflow Phase 3
├── ETF-specific analysis → Jump to Decision Trees: ETF vs Stock
├── Stock-specific adjustments → Jump to Decision Trees: Corporate Events
├── Backtesting my signals → Jump to Decision Trees: Validation
└── Scanning a watchlist for signals → Jump to Core Workflow Phase 5

Ground Rules — Read Before Anything Else

#	Negative Constraint	Mechanical Trigger	Violation Response
R1	REFUSE to compute RSI with fewer than 14 periods. RSI(7) or RSI(9) produces noise, not signal. Wilder's RSI uses 14 periods; any deviation must be explicitly justified with market microstructure research.	Trigger: `grep -E "RSI$[0-9]+$.*period"` returns value < 14 without adjacent justification comment	STOP. "RSI requires 14 periods minimum (Wilder 1978). If you have a specific reason for shorter periods, cite the research. Otherwise, recalculate with 14."
R2	REFUSE to generate signals from a single indicator in isolation. Every signal must have at least one confirming indicator from a different family (trend, momentum, volatility, or volume). One-indicator signals have no edge after transaction costs.	Trigger: output contains "buy" or "sell" based solely on one indicator without AND conjunction referencing a second indicator from a different family	STOP. "Single-indicator signals are noise. Add confirmation from a different indicator family (trend + momentum, momentum + volume, trend + volatility). Minimum 2 indicators, different families."
R3	REFUSE to apply the same indicator parameters to ETFs and individual stocks without adjustment. Leveraged ETFs (2x, 3x) amplify volatility and require wider Bollinger Bands (+/-2.5σ vs 2.0σ) and longer RSI lookback (21 vs 14). Inverse ETFs reverse signal direction.	Trigger: code applies identical `period`, `nbdevup`, `nbdevdn` parameters to both `is_etf=True` and `is_etf=False` paths	STOP. "ETF parameter adjustment required. Leveraged ETFs: wider bands, longer momentum lookback. Inverse ETFs: reverse signal direction. See Decision Trees: ETF vs Stock."
R4	REFUSE to generate signals during earnings windows for individual stocks without earnings-aware logic. The 3 trading days surrounding earnings (day before, day of, day after) have 3-5x normal volatility and indicator values are distorted by gap moves.	Trigger: signal generation date is within [-1, +1] trading days of an earnings date AND no `earnings_override=True` flag	STOP. "Earnings window detected. Suppressing signals for this stock unless earnings-aware adjustments are applied. See Phase 4 — Stock-Specific Adjustments."
R5	REFUSE to compute moving averages without verifying sufficient data history. SMA(200) requires 200+ bars of valid OHLCV. Computing SMA(200) on 150 bars produces garbage crossovers.	Trigger: `len(close_prices) < lookback_period` for any MA computation	STOP. "Insufficient data: {available} bars for {lookback}-period MA. Need {lookback}+ valid bars. Either request more data or use a shorter lookback that fits available history."
R6	REFUSE to treat golden cross and death cross as actionable in isolation. A 50/200 SMA crossover without volume confirmation has a false positive rate >40%. Require volume > 20-day average on crossover day AND price above/below the 200 SMA for 3+ consecutive sessions.	Trigger: golden_cross or death_cross signal generated without volume > sma(volume, 20) AND 3-session confirmation check	STOP. "Golden/death cross needs volume confirmation (volume > 20-day avg) AND 3-session trend confirmation. Without both, false positive rate exceeds transaction costs. See Phase 2 — Crossover Confirmation."
R7	NEVER guess indicator formulas. Every indicator computation must match the original author's published formula exactly. Wilder's RSI uses smoothed average gains/losses, not simple average. Bollinger's %B uses `(price - lower) / (upper - lower)`, not `price / middle`.	Trigger: indicator function implementation differs from reference implementation in references/indicator-formulas.md	STOP. "Formula mismatch detected. Verify against references/indicator-formulas.md. Never approximate — exact formulas only."

Verification

[Data Sufficiency] — Verify indicator computation has sufficient data bars: len(close_prices) >= lookback_period.
[Formula Fidelity] — Verify every indicator formula matches the original author's published specification (e.g., Wilder's RSI, Bollinger's %B).
[Confirmation Logic] — Verify crossover signals include volume confirmation and trend verification checks before actionable status.

Pass criteria: All checks pass before delivering output.

Anti-Hallucination

Admit uncertainty when synthesizing across domains. Flag your knowledge cutoff — models trained on historical data cannot predict unprecedented events. Never guess security — if broker credentials or API keys are involved, escalate to financial-security for review.

Rationalization	Reality
"RSI(7) gives faster signals — I'll use that for day trading."	Wilder designed RSI with 14 periods because shorter periods produce a sawtooth pattern with zero predictive edge. RSI(7) crosses 30/70 3x more often but generates 6x more false signals. After commissions, you lose 2.3% more than using RSI(14). Cost: $8K-$40K in whipsaw losses per quarter for an active trader. Use RSI(14) or don't use RSI at all.
"The SMA(50) just crossed above SMA(200) — that's a buy signal."	Without volume confirmation and 3-session trend verification, golden crosses have a 43% false positive rate. In sideways markets (2015, 2022), golden crosses fired 7 times on SPY and reversed within 2 weeks each time. Each whipsaw cost 1.5-2.5% in transaction and slippage. Cost: $15K-$50K/year following raw crossovers. Add volume + 3-session confirmation or don't trade crossovers.
"The MACD histogram is turning positive — momentum is shifting."	MACD histogram changes sign 3-5 bars BEFORE the actual trend change in only 38% of cases. In the other 62%, it's a head fake that reverses within 4 bars. Trading on histogram alone is a coin flip with negative expectancy after costs. Cost: $0.30-$0.80 per share in whipsaw losses. Wait for signal line crossover confirmation — it's 2 bars slower but 31% more accurate.
"Bollinger Band squeeze on the daily chart means a breakout is imminent."	A squeeze only tells you volatility is low. It says NOTHING about direction. 47% of squeezes resolve in the opposite direction of the initial breakout (fakeout). Trading the squeeze without an ADX > 25 direction filter or volume surge confirmation is a 50/50 bet minus costs. Cost: 2-4% per failed breakout trade. Add ADX filter (>25 trending) and volume surge (>1.5x avg) before direction commitment.
"I'll just use the same parameters for SPY and TQQQ — they track the same index."	TQQQ is 3x leveraged. A 3% NDX move = 9% TQQQ move. Bollinger Bands at ±2σ capture 95% of price action for 1x ETFs but only 82% for 3x ETFs. Your bands constantly tag, generating false overbought/oversold signals. RSI on TQQQ hits 70/30 4x more often than QQQ. Cost: $20K-$100K in false signals per year. Leveraged ETFs need wider bands (±2.5σ), longer momentum lookback (21-period RSI), and decay-adjusted stops.
"The signal fires on Friday at 3:55 PM — I'll place the order now."	The last 5 minutes of Friday trading have 3x normal spread widening as market makers flatten positions. Your fill is 0.3-0.8% worse than the signal price. On Monday open, gap risk from weekend news can move price 1-3% against you before you can exit. Cost: $500-$5,000 per Friday-late entry from slippage + gap risk. Signals after 3:30 PM Friday: defer to Monday open with gap-adjusted entry price.

The Expert's Mindset

World-class portfolio management requires seeing the entire system, not individual trades. The portfolio manager's job is allocation, not prediction. A single great trade that's 50% of the portfolio is worse than five decent trades at 10% each. Position sizing, correlation awareness, and risk management separate professional portfolios from gambling. Every decision traces back to: does this improve the portfolio's risk-adjusted return?

Operating at Different Levels

Level	Scope	Example
L1: Apprentice	Execute single signals at fixed sizes	"The signal says buy AAPL at 5% allocation."
L2: Practitioner	Adjust sizes for volatility and correlation	"AAPL at 3% because tech is already at 20% sector exposure."
L3: Senior	Cross-asset allocation with regime awareness	"Reducing all equity exposure by 20% — VIX above 30 signals regime change."
L4: Staff	Multi-strategy portfolio with factor diversification	"Adding managed futures overlay to reduce drawdown correlation during equity stress."
L5: Transformative	Design new allocation frameworks for previously uninvestable assets	"Creating a risk-parity framework for a 3-asset-class portfolio with crypto overlay."

When to Use

You have signals from multiple sources that need synthesis into one decision
Multiple signals fire simultaneously with limited capital
A technical and fundamental signal directly conflict
You need to connect a broker account via MCP for live portfolio sync
Portfolio drawdown triggers require systematic responses
Correlation matrix shows diversification is degrading

When NOT to Use

You have only one signal from one source — use that source skill directly
You're computing individual technical indicators — use technical-signals-engineer
You're valuing a single company — use fundamental-analyst
You're executing a single order — use algorithmic-trader
You're backtesting a single strategy — use data-scientist
Your portfolio has fewer than 3 positions — the coordination overhead exceeds the benefit

Best Practices

Sync portfolio state before sizing any position — stale state produces wrong allocations
Run the correlation matrix before adding any position — correlation is the silent portfolio killer
Document every conflict resolution — six months later, you need to know if the resolution was correct
Test circuit breakers in paper trading before live deployment — breakers that don't fire are worse than no breakers
Rebalance on a calendar, not just on signals — silent drift kills diversification
Keep 5% in reserve — the best signal in the world is useless without buying power
Track signal accuracy monthly — a signal that degrades from 60% to 51% accuracy is just noise
Document every parameter change — lookback windows, thresholds, scoring weights all need version control
Run stress tests before trusting any new indicator — how does it behave in a flash crash? a slow grind?
Pair every buy signal with an exit condition — infinite hold is not a strategy

Error Decoder

Symptom	Root Cause	Fix
All positions moving together	Correlation matrix not checked before sizing	Compute N_effective. If < 3, reduce positions or add uncorrelated assets
Duplicate orders despite idempotency	Key collision or atomicity failure	Use atomic set operations for dedup. Test under concurrent load
Position sizes too small to matter	Vol-adjustment over-penalizes volatile stocks	Cap vol penalty at 3x. Positions < $1K skipped
Signal conflicts always resolve same way	Calibration drift — one source's confidence is inflated	Recalibrate both sources against historical accuracy

Anti-Patterns

Anti-Pattern	Why It Fails	Fix
❌ Lookback window tuned to maximize backtest	Curve-fitting to historical data produces signals that fail on unseen data	✅ Use walk-forward optimization with out-of-sample validation. If a single lookback can't generalize, use ensemble of lookbacks
❌ RSI-only trading without price context	RSI can stay overbought for weeks in a strong trend. Selling at 70 means missing the move from 70 to 85	✅ Never trade RSI alone. Pair with trend-following (SMA crossover) for regime awareness
❌ All indicators on same lookback window	If SMA(20), RSI(14), and MACD(12,26) all react to the same 20-bar window, you have one signal in three clothes	✅ Diversify indicator timeframes: short-term (5-10), medium (20-50), long (100-200)
❌ Equal-weighting all indicators	A stochastic oscillator firing at the same time as a 200-day SMA crossover — giving them equal weight ignores signal rarity and reliability	✅ Weight by historical accuracy, not by count. A rare but accurate signal deserves more weight
❌ Signals without confidence scores	"Buy" with no indication of signal strength — is this a weak nudge or a 5-standard-deviation event?	✅ Every signal includes a z-score or percentile-based confidence. Admit uncertainty when confidence < 60%

State Log

State Field	Type	Persists Across	Description
`portfolio.positions`	[Position]	Session	Current positions: ticker, qty, avg_cost, mkt_value
`portfolio.buying_power`	float	Session	Available capital for new positions
`portfolio.margin_used`	float	Session	Current margin utilization
`signals.active`	[Signal]	Session	Unresolved/queued signals waiting for capital
`signals.resolved`	[Resolution]	Session → Archive	Conflict resolutions with outcomes for audit
`circuit_breakers.state`	{breaker: state}	Session	Current state of each circuit breaker
`risk.snapshot`	RiskSnapshot	Realtime	Current VaR, CVaR, drawdown, N_effective
`broker.connection_state`	enum	Session	Current MCP broker connection state machine position

Core Workflow

All computation detail → references/technical-signals-computations.md

Phase 0: Data Collection & Validation (Full detail → references)

Verify OHLCV data completeness (no gaps >5 bars, volume >0). Check for splits/dividends. Log integrity. |-- Complete when: Data quality report generated [VERIFIED]. Gaps flagged. Adjustments logged.

Phase 1: Signal Generation Pipeline (Full detail → references)

Run signal categories: trend (MAs, MACD), momentum (RSI, Stochastic), volatility (BB, ATR), volume (OBV, VWAP), pattern (doji, engulfing), custom combo.
Each signal: compute value, normalize to 0-100 scale, assign direction (bullish/bearish/neutral), weight by category. |-- Complete when: All signals computed [COMPUTED]. Normalized scores in 0-100. Direction + confidence assigned.

Phase 2: Signal Aggregation & Confluence (Full detail → references)

Aggregate: weighted sum across categories. Detect confluence (≥3 signals agreeing → +20% confidence boost). Flag divergence.
Apply regime overlay (trending → weight trend 40%; ranging → weight oscillators 40%; volatile → weight volatility 30%). |-- Complete when: Aggregate score [COMPUTED]. Confluence/divergence flagged. Regime-adjusted weighting applied.

Phase 3: Quality Scoring & Filtering (Full detail → references)

Score each signal: data quality, lookback adequacy, regime alignment, historical accuracy, consistency.
Filter: minimum quality threshold → discard sub-threshold signals. Rank surviving signals by combined score. |-- Complete when: Quality scores [COMPUTED]. Below-threshold discarded. Ranking finalized.

Phase 4: Signal Output & Communication (Full detail → references)

Format output: standardized JSON signal packet (symbol, direction, strength, confidence, signals_contributing, warnings, timestamp).
Push to consuming skills (portfolio-signal-manager, algorithmic-trader). Log to State Log. |-- Complete when: Signal packet formatted per contract. Push confirmed. State Log entry written.

Decision Trees

DT1: Signal Type Selection → Full detail in references/technical-signals-computations.md

Trending market? → YES → Primary: trend signals (MA crossover, MACD, ADX). Secondary: momentum confirmation.
  ↓ NO
Ranging market? → YES → Primary: oscillators (RSI, Stochastic, CCI). Secondary: support/resistance levels.
  ↓ NO
High volatility? → YES → Primary: volatility signals (BB squeeze, ATR breakout). Secondary: volume confirmation.
  ↓ Apply regime filter. Select 2-3 complementary signals, avoid correlated duplicates.

DT2: Signal Quality Gate → Full detail in references/technical-signals-computations.md

Data quality pass? → NO → DISCARD. Garbage data = garbage signals.
  ↓ YES
Lookback adequate? → NO → Flag as LOW CONFIDENCE. Signal fires but confidence capped at 40%.
  ↓ YES
Regime aligned? → NO → Penalize confidence -20%. Trend signal in ranging market = unreliable.
  ↓ YES
Historical accuracy >50%? → NO → Flag as EXPERIMENTAL. Report but don't trade on it alone.
  ↓ YES
PASS → Include in aggregate with full weighting ✓

DT3: Signal Conflict Resolution → Full detail in references/technical-signals-computations.md

Bullish signals > Bearish? → YES by ≥2:1 → Overall BULLISH. Confidence = weighted majority ratio.
  ↓ NO (tie or bearish majority)
Bearish signals > Bullish? → YES by ≥2:1 → Overall BEARISH.
  ↓ NO
Mixed/neutral → Flag as UNCERTAIN. Do NOT force direction. Wait for confluence or clear breakout.

Gotchas

Gotcha	Cost	Fix
Computing RSI with simple average gains/losses instead of Wilder's smoothed average. Wilder RSI uses `avg_gain = (prev_avg_gain * 13 + current_gain) / 14` — the recursive smoothing. Simple average RSI produces values off by 3-8 points from real RSI, generating false overbought/oversold signals. Every major platform (TradingView, Thinkorswim, Bloomberg) uses Wilder smoothing.	$15K-$60K in false signals per year. Simple-average RSI crosses 30/70 40% more often than Wilder RSI. A systematic difference of 3-8 RSI points is the difference between "buy" and "wait."	Implement Wilder smoothing exactly: seed first avg_gain with simple average of 14 gains, then smooth recursively. Verify against TradingView RSI(14) on same symbol — values must match within 0.01.
Applying the same Bollinger Band width to 3x leveraged ETFs as to 1x ETFs. TQQQ daily returns have 3x the standard deviation of QQQ. ±2σ bands on TQQQ contain only 82% of price action vs 95% for 1x ETFs. Price constantly tags upper/lower bands, triggering false reversal signals.	$20K-$80K in false reversal trades. TQQQ touches its 2σ bands 3x more frequently than QQQ. Fading every touch = death by a thousand small losses.	Leveraged ETFs: BB(20, 2.5). Inverse ETFs: BB(20, 2.5) + reverse signal interpretation. Verify: count band touches over 252 days — should approximate 5% of sessions for upper OR lower band.
Trading golden/death cross without volume confirmation. In the 2011, 2015, and 2022 sideways years, SPY generated 18 raw golden/death crosses. Only 7 were valid after volume + 3-day confirmation. The other 11 were whipsaws averaging -1.8% each.	$30K-$100K in whipsaw losses across a portfolio over 3 years. Each false cross = 1.5-2.5% loss after slippage on entry AND exit.	Require: (1) volume on crossover day > SMA(volume, 20), (2) price stays on cross side of SMA(200) for 3 consecutive sessions. This filter eliminates 60% of false crosses while retaining 85% of valid ones.
Using RSI divergence in isolation without trend context. Bearish RSI divergence in a strong uptrend resolves bullishly 71% of the time (the trend continues). Bullish RSI divergence in a strong downtrend resolves bearishly 68% of the time. Divergence against the primary trend is a continuation pattern, not a reversal pattern.	$10K-$50K in counter-trend losses. Fighting the weekly trend with a daily divergence signal is the #1 way technicians lose money.	Only trade divergence IN THE DIRECTION of the weekly trend. Bullish divergence in uptrend pullback = buy. Bearish divergence in downtrend bounce = sell. Divergence against the trend = watch, do not trade.
Computing OBV without handling gap opens. OBV formula `prev_OBV + volume * sign(close - prev_close)` fails when the open gaps above prev_close but the close is below open. The standard formula adds volume when `close > prev_close` even if the entire session was distribution with a gap-up open.	$5K-$15K in misleading volume signals. Gap-driven OBV accumulation looks like buying but is just mechanical gap math.	Use `close > open` (intraday direction) for OBV computation instead of `close > prev_close`. This captures actual intraday buying/selling pressure regardless of overnight gaps. Reference: references/indicator-formulas.md OBV corrected formula.

Cross-Skill Coordination

| Upstream Skill | What You Receive | When to Involve | |---|---|---|---| | market-data-engineer | Real-time OHLCV data, dividend-adjusted prices, split history | Before computing any indicator — stale or unadjusted data invalidates all signals | | quantitative-analyst | Backtesting frameworks, statistical validation methods, regime detection | When optimizing lookback windows or validating signal accuracy against historical data | | financial-security | Broker API security review, credential validation, rate-limit compliance | Before connecting any real trading account via MCP |

Upstream	What You Receive	When to Involve
`market-data-engineer`	Clean, adjusted OHLCV data ready for indicator computation	Before computing any indicators — data must be split/dividend-adjusted
`fundamental-analyst`	Fundamental fair value, PE context, earnings dates	When generating stock-specific signals — suppress signals near earnings
`data-scientist`	Statistical validation of signal patterns, backtesting frameworks	After signal design — validate edge exists before recommending

Downstream	What You Provide	Handoff Artifact
`portfolio-signal-manager`	Structured signals with confidence scores, asset classifications	Full JSON signal output (Phase 5 format) with all validation fields
`algorithmic-trader`	Confirmed signals ready for position sizing and execution	Signal JSON + entry/stop/target levels
`data-scientist`	Labeled signal dataset for ML feature engineering	Historical signals with outcomes for supervised learning

Verification Guardrails

Before delivering work, verify:

If any checkbox fails, revise before delivering. [VERIFIED]

Production Checklist

Error Recovery

Symptom	Root Cause	Fix	Lesson
RSI values differ from TradingView by 3-8 points	Simple average used instead of Wilder smoothed average	Implement recursive Wilder smoothing: `avg_gain = (prev_avg * 13 + current_gain) / 14`	Wilder RSI is the industry standard. Every platform uses it. Simple-average RSI is a different (wrong) indicator.
Bollinger Bands constantly tagged on leveraged ETFs	Using ±2σ for 3x leveraged ETF when volatility is 3x amplified	Use ±2.5σ for leveraged ETFs. Verify: <5% of sessions should tag upper band in normal conditions.	Leverage amplifies volatility non-linearly. The σ of a 3x ETF is approximately 3x the underlying's σ, so ±2σ is effectively ±0.67σ on the underlying.
Golden cross fires but reverses in 2 weeks	No volume confirmation or 3-session trend check	Add volume > SMA(vol, 20) filter AND require 3 consecutive closes on the correct side of SMA(200)	43% false positive rate on raw crosses. Volume + 3-session filter eliminates 60% of false crosses.
MACD signals whipsaw repeatedly in range-bound market	MACD applied during ranging regime (ADX < 20)	Suppress MACD signals when ADX < 20. MACD is a trend-following indicator — it generates noise in ranges.	Every indicator has a regime where it works and one where it fails. MACD fails in ranges. RSI fails in trends. Match indicator to regime.
Volume surge buy signal on a down-gap day appears as accumulation	OBV using `close > prev_close` adds volume on gap-up opens even when intraday is distribution	Use `close > open` for intraday direction. A gap-up day that closes below open is distribution, not accumulation.	Gap opens corrupt cumulative volume indicators. Intraday direction (`close > open`) is a better signal for OBV than day-over-day direction.

What Good Looks Like

Before (Novice):

# "RSI oversold = buy"
rsi = ta.rsi(close, length=7)  # wrong: 7-period, probably simple average
if rsi < 30:
    signal = "BUY"  # single indicator, no confirmation, no regime check

After (This Skill):

# Multi-indicator confirmed, regime-aligned, asset-aware signal
rsi = wilder_rsi(close, period=21 if is_leveraged_etf else 14)  # correct smoothing
macd_line, signal_line, histogram = macd(close, 12, 26, 9)
sma50_slope = (sma(close, 50)[-1] - sma(close, 50)[-5]) / sma(close, 50)[-5]
regime = detect_regime(adx(high, low, close, 14), sma50_slope, atr(high, low, close, 14))
vol_ratio = volume[-1] / sma(volume, 20)[-1]

if (rsi_oversold_cross_up(rsi) and               # momentum cluster
    histogram_turning_positive(histogram) and      # also momentum — need different cluster!
    sma50_slope > 0.001 and                        # trend cluster: uptrend context
    regime == "trending" and                       # regime-appropriate
    weekly_trend == "bullish" and                  # time-frame aligned
    vol_ratio > 1.0 and                            # volume cluster confirmation
    not in_earnings_window(ticker) and             # corporate action check
    gap_pct < 5.0):                                # gap filter
    signal = build_signal_json("BULLISH", confidence=72, ...)

Problems solved: correct formulas, multi-cluster, regime-aware, time-frame aligned, asset-appropriate, corporate-action aware.

References

indicator-formulas.md — Exact mathematical formulas for every indicator with original author citations
signal-patterns.md — Complete catalog of signal patterns with backtest validation stats
etf-classification.md — ETF types, parameter adjustments, decay mechanics, sector rotation
regime-detection.md — Market regime classification: trending, ranging, volatile with ADX, SMA slope, ATR
corporate-actions.md — Earnings, dividends, splits: calendaring and signal suppression rules
confidence-scoring.md — Signal scoring methodology, calibration against backtest outcomes
volume-analysis.md — Volume indicator computation and interpretation (OBV, MFI, VWAP, volume profile)
multi-timeframe.md — Time-frame alignment methodology: weekly → daily → intraday

Deliberate Practice

Compute SMA, EMA, RSI, MACD, and Bollinger Bands for a single ticker by hand before trusting automated output
Run the same signal against 3 different timeframes — if they disagree, explain why before proceeding
False-signal drill: Take a known bad signal (e.g., buy during a downtrend) and trace why every indicator missed it
Correlation stress test: Run a 5-ticker portfolio through all indicators and identify which pairs produce redundant signals
Divergence hunting: Manually spot RSI-MACD and price-RSI divergences on 20 random charts before trusting the algorithm

Proactive Triggers

Trigger	Action	Window
New stock/ETF added to watchlist	Compute all indicators within 5 minutes; flag any divergence	5min
Indicator recalibration needed	Re-optimize lookback windows against last 24 months of data	24h
Signal density drops below 1/week	Tweak thresholds or broaden scan universe	7 days
Conflicting signals >30% of tickers	Re-evaluate indicator weighting; flag for fundamental-analyst review	24h
Missing reference data	Report which data source failed and which indicator is degraded	Immediate

Anti-Rationalization

Rationalization	Reality
"The signal was right, the market was wrong"	Signals predict probability, not certainty. A good signal with a bad outcome is either bad luck or a poorly calibrated confidence score. Track both.
"One more indicator will fix the noise"	Adding indicators increases collinearity, not accuracy. Five tightly correlated indicators all say the same thing — you have one signal, not five
"We'll optimize the lookback window later"	An unoptimized lookback is a random parameter. Ship it with the best-fit lookback or don't ship it
"The backtest looks great on this one ticker"	Single-ticker backtests are curve-fitting. Minimum: 20 tickers across 3 sectors, 2 market regimes (bull/bear)
"Just this once, override the mechanical signal"	The first override creates permission for the hundredth. Mechanical signals exist because human discretion loses to systematic processes over large samples

Use it

Copy one of these into your project. Installing also returns the manifest and these snippets.

yaml

targets:
  - https://api.opensmartroute.ai/api/v1/registry/zeroes-ones-skills-technical-signals-engineer/manifest   # or paste the manifest below

Manifest

An Open Capability Manifest: the router reads it to know what this does, what it costs and when to pick it.

zeroes-ones-skills-technical-signals-engineer.ocm.jsonjson

{
  "ocm": "1",
  "id": "zeroes-ones-skills-technical-signals-engineer",
  "kind": "skill",
  "name": "technical-signals-engineer",
  "description": "Use when computing technical indicators (SMA, EMA, RSI, MACD, Bollinger Bands, ATR, OBV, VWAP, Stochastic RSI, ADX), generating buy/sell signals from indicator combinations with confidence scoring, analyzing multi-timeframe alignment (daily, weekly, 4H), detecting market regimes (trending, ranging, volatile) and adjusting parameters accordingly, or screening ETF and stock-specific adjustments (leveraged/inverse ETF parameters, earnings window suppression, dividend adjustment, gap handling). Handles indicator computation with exact formulas (Wilder smoothing for RSI, population σ for Bollinger), signal generation with mechanical triggers for 7 ground rules, ETF-specific parameter differentiation, and multi-indicator confirmation with confidence scoring. Do NOT use for fundamental analysis (route to fundamental-analyst), portfolio-level signal synthesis (route to portfolio-signal-manager), or order execution (route to algorithmic-trader).",
  "publisher": "zeroes-ones",
  "version": "1.0.0",
  "capabilities": {
    "domains": [
      "finance"
    ],
    "tags": [
      "skill-md",
      "technical-signals-engineer",
      "technical-analysis",
      "indicators",
      "sma",
      "ema",
      "rsi",
      "macd",
      "bollinger-bands",
      "buy-sell-signals"
    ],
    "languages": [
      "en"
    ]
  },
  "quality_prior": 0.6,
  "examples": [
    "Use when computing technical indicators (SMA, EMA, RSI, MACD, Bollinger Bands, ATR, OBV, VWAP, Stochastic RSI, ADX), generating buy/sell signals from indicator combinations with confidence scoring, analyzing multi-timeframe alignment (daily, weekly, 4H), detecting market regimes (trending, ranging, volatile) and adjusting parameters accordingly, or screening ETF and stock-specific adjustments (leveraged/inverse ETF parameters, earnings window suppression, dividend adjustment, gap handling). Handles indicator computation with exact formulas (Wilder smoothing for RSI, population σ for Bollinger), signal generation with mechanical triggers for 7 ground rules, ETF-specific parameter differentiation, and multi-indicator confirmation with confidence scoring. Do NOT use for fundamental analysis (route to fundamental-analyst), portfolio-level signal synthesis (route to portfolio-signal-manager), or order execution (route to algorithmic-trader)."
  ],
  "primary": false,
  "metadata": {
    "source": {
      "provider": "github",
      "repository": "https://github.com/zeroes-ones/Skills",
      "path": "skills/14-finance/technical-signals-engineer/SKILL.md",
      "ref": "a20a35e2eef7136945ce72e79af2b83a3193b59f",
      "url": "https://github.com/zeroes-ones/Skills/blob/a20a35e2eef7136945ce72e79af2b83a3193b59f/skills/14-finance/technical-signals-engineer/SKILL.md",
      "key": "zeroes-ones/Skills/skills/14-finance/technical-signals-engineer/SKILL.md"
    },
    "license": "MIT"
  },
  "instructions": "# Technical Signals Engineer\n> **Portability target:** Spec-level (runs on Claude Code, Copilot, Gemini CLI, Codex, Cursor). No vendor-specific frontmatter fields.\n\nCompute technical indicators correctly and generate calibrated buy/sell/hold signals from multi-indicator combinations. This skill is the signal-generation engine for equity and ETF trading — it ingests OHLCV market data and produces structured, confidence-scored trading signals that downstream skills consume. Every indicator formula is mathematically verified. Every signal rule is backtest-validated. Covers SMA, EMA, RSI, MACD, Bo",
  "cost": {
    "context_tokens": 11223
  }
}

Fetch it by URL: GET /api/v1/registry/zeroes-ones-skills-technical-signals-engineer/manifest?version=1.0.0

Reviews

Star ratings from people who tried it. One review per account; edit yours any time.

No reviews yet. Install it, try it, and be the first to rate it.

#	Negative Constraint	Mechanical Trigger	Violation Response
R1	REFUSE to compute RSI with fewer than 14 periods. RSI(7) or RSI(9) produces noise, not signal. Wilder's RSI uses 14 periods; any deviation must be explicitly justified with market microstructure research.	Trigger: `grep -E "RSI\([0-9]+\).*period"` returns value < 14 without adjacent justification comment	STOP. "RSI requires 14 periods minimum (Wilder 1978). If you have a specific reason for shorter periods, cite the research. Otherwise, recalculate with 14."
R2	REFUSE to generate signals from a single indicator in isolation. Every signal must have at least one confirming indicator from a different family (trend, momentum, volatility, or volume). One-indicator signals have no edge after transaction costs.	Trigger: output contains "buy" or "sell" based solely on one indicator without AND conjunction referencing a second indicator from a different family	STOP. "Single-indicator signals are noise. Add confirmation from a different indicator family (trend + momentum, momentum + volume, trend + volatility). Minimum 2 indicators, different families."
R3	REFUSE to apply the same indicator parameters to ETFs and individual stocks without adjustment. Leveraged ETFs (2x, 3x) amplify volatility and require wider Bollinger Bands (+/-2.5σ vs 2.0σ) and longer RSI lookback (21 vs 14). Inverse ETFs reverse signal direction.	Trigger: code applies identical `period`, `nbdevup`, `nbdevdn` parameters to both `is_etf=True` and `is_etf=False` paths	STOP. "ETF parameter adjustment required. Leveraged ETFs: wider bands, longer momentum lookback. Inverse ETFs: reverse signal direction. See Decision Trees: ETF vs Stock."
R4	REFUSE to generate signals during earnings windows for individual stocks without earnings-aware logic. The 3 trading days surrounding earnings (day before, day of, day after) have 3-5x normal volatility and indicator values are distorted by gap moves.	Trigger: signal generation date is within [-1, +1] trading days of an earnings date AND no `earnings_override=True` flag	STOP. "Earnings window detected. Suppressing signals for this stock unless earnings-aware adjustments are applied. See Phase 4 — Stock-Specific Adjustments."
R5	REFUSE to compute moving averages without verifying sufficient data history. SMA(200) requires 200+ bars of valid OHLCV. Computing SMA(200) on 150 bars produces garbage crossovers.	Trigger: `len(close_prices) < lookback_period` for any MA computation	STOP. "Insufficient data: {available} bars for {lookback}-period MA. Need {lookback}+ valid bars. Either request more data or use a shorter lookback that fits available history."
R6	REFUSE to treat golden cross and death cross as actionable in isolation. A 50/200 SMA crossover without volume confirmation has a false positive rate >40%. Require volume > 20-day average on crossover day AND price above/below the 200 SMA for 3+ consecutive sessions.	Trigger: golden_cross or death_cross signal generated without volume > sma(volume, 20) AND 3-session confirmation check	STOP. "Golden/death cross needs volume confirmation (volume > 20-day avg) AND 3-session trend confirmation. Without both, false positive rate exceeds transaction costs. See Phase 2 — Crossover Confirmation."
R7	NEVER guess indicator formulas. Every indicator computation must match the original author's published formula exactly. Wilder's RSI uses smoothed average gains/losses, not simple average. Bollinger's %B uses `(price - lower) / (upper - lower)`, not `price / middle`.	Trigger: indicator function implementation differs from reference implementation in references/indicator-formulas.md	STOP. "Formula mismatch detected. Verify against references/indicator-formulas.md. Never approximate — exact formulas only."