technical-analysis-workflow v1.0.0

name: technical-analysis-workflow description: Run disciplined chart analysis with a repeatable workflow. Use when reading a price chart, validating a trading setup, confirming a trend with market breadth and volume, designing a backtest, or explaining what technical analysis can and cannot claim.

Technical Analysis Workflow

Purpose

Most chart-pattern writeups are one-liner stubs: "double bottom = bullish." This skill replaces that with a repeatable discipline: state the thesis up front, require breadth and volume confirmation before trusting a price signal, backtest any rule with real bias controls, and write the limitations section every time. It is for analysts, builders, and agents who want chart analysis they can defend, not chart reading as decoration.

Workflow

1. Write the thesis first. One sentence: what do you believe the chart says, what trade or conclusion follows, and what price level invalidates it. If you cannot name the invalidation level, you do not have a thesis.
2. Map the trend on two timeframes. Higher highs and higher lows (or the reverse) on both the trading timeframe and one level up. Moving averages (20/50/200-day) are trend filters, not buy signals on their own. Record which trend regime you are in: uptrend, downtrend, range.
3. Require breadth confirmation. A price signal counts as strong only if market internals agree. Check at least two of: advance-decline line direction, McClellan Oscillator position and slope, share of index components above their 50-day moving average, up-volume share. If the index makes a new high while breadth indicators make lower highs, that is a divergence, and it weakens the bullish case. See references/ta-workflow.md for indicator details.
4. Require volume confirmation. Trend moves should carry volume. Check On-Balance Volume direction against price, Chaikin Money Flow sign, and whether breakouts print volume clearly above the 20-day average. A breakout on thin volume gets a weaker grade.
5. Scan for divergences. Price versus RSI, MACD, and OBV. A new price high that an indicator fails to confirm is a warning flag, not an entry.
6. Backtest any trading rule you derive. In-sample/out-of-sample chronological split (hold out at least 30%), walk-forward validation, transaction costs included (commissions, spread, slippage), and the five bias checks: look-ahead, survivorship, overfitting, multiple-testing (too many tuned variants inflates the best result), and cost drag. Details in references/ta-workflow.md.
7. Write the limitations section. Mandatory, not optional. Cover the academic evidence status, the efficient-market objection, regime change, and the fact that a passed backtest predicts nothing with certainty. Frame the output as a disciplined hypothesis, never as certainty.
8. Produce the analysis report per the Output Contract.

Output Contract

Every analysis produced under this skill ships a report with these sections, in this order:

1. Thesis: the one-sentence claim plus invalidation level.
2. Trend map: regime per timeframe, key levels, trend-filter readings.
3. Confirmation scorecard: breadth checks, volume checks, divergence scan, each graded: confirms / mixed / contradicts. The helper bin/scorecard.py turns a set of indicator readings into a composite grade.
4. Backtest appendix (if a rule was proposed): data window, in/out-of-sample split, costs modeled, biases checked, risk metrics: Sharpe, max drawdown, win rate, profit factor. Never report win rate alone.
5. Limitations: honest notes on evidence, assumptions, and what would change the conclusion.
6. Verdict: one of: supported, weakly supported, unsupported. No hedge words.

Operating Rules

1. Every factual claim gets verified against a primary source (docs, standards bodies, published curricula, or direct measurement). Unverifiable claims get cut or framed explicitly as opinion.
2. Never invent numbers, dates, quotes, or events. No confident unverified claims, ever.
3. An indicator reading without its setting (period, universe, session) is not a reading. RSI(14) on daily closes is a different claim than "RSI is high."
4. Breadth indicators are index-universe dependent. The McClellan Oscillator built on NYSE advances/declines says nothing about a single stock or a crypto pair; say which universe you measured.
5. A backtest that omits transaction costs, uses current index members on old data, or peaks at future prices is not evidence of anything. Run the bias checks or label the result illustrative.
6. Keep the output plain and direct: no hype, no motivational framing, no "game-changer" language. If a sentence sounds like a pitch, rewrite it.
7. Bulky detail goes in references/; the report itself stays tight.