obai-stock-synthesis

name

obai-stock-synthesis

description

Use when finalizing any non-terminal analysis built from one or more evidence-supplier specialists (`market_data_analysis`, `fundamentals_analysis`, `events_news_analysis`, `options_analysis`, `screener_lookup`, `portfolio_analysis`, `research_analysis`). Covers stock, ETF, company, sector, thematic, portfolio, and research-only queries. Do not use for terminal strategy backtest or terminal prediction-market output.

OBaI Stock Synthesis

Purpose

Use this skill to turn evidence from non-terminal specialist agents into a concise user-facing answer.

This skill is for synthesis. It is not a routing override and it is not a substitute for specialist tools.

When to use

Use this skill when the final response depends on one or more evidence-supplier specialists:

• market_data_analysis
• fundamentals_analysis
• events_news_analysis
• options_analysis
• screener_lookup
• portfolio_analysis
• research_analysis

Use it only after the needed specialist outputs are available or after deciding that a requested dimension is unavailable.

When not to use

Do not use this skill when a terminal specialist output controls the final response.

Do not use it to reshape completed or pending output from:

• strategy_analysis
• prediction_market_analysis

Do not use the regular stock-analysis structure if it would remove, compress, or rename sections from a terminal specialist artifact.

Synthesis rule

The final answer should preserve the decision-relevant facts from each specialist used.

A fact is decision-relevant when it changes one of these:

• valuation view
• risk view
• catalyst view
• liquidity view
• timing view
• portfolio fit
• tradeoff assessment
• confidence level

If a specialist result does not materially affect the answer, mention that briefly or omit it only when the omission does not hide a relevant risk, conflict, or missing data condition.

Coverage gate

For each specialist used, include at least one concrete takeaway from its main dimension unless unavailable.

Lead with the most diagnostic facts — the ones that drive the conclusion. Do not dump every number from the specialist output; pick the metrics that change the answer.

Use these dimensions:

• market_data_analysis: price context, trend, returns, range, volume, volatility, or relative movement
• fundamentals_analysis: valuation, profitability, growth, leverage, cash flow, estimates, capital returns, or balance-sheet quality
• events_news_analysis: catalyst, event timing, sentiment direction, regulatory item, earnings item, macro item, or material uncertainty
• options_analysis: implied volatility, liquidity, spread, open interest, skew, Greeks, positioning, or expiration structure
• screener_lookup: filters used, matching rationale, ranking rationale, ticker resolution, or candidate exclusion reason
• portfolio_analysis: exposure, concentration, correlation, drawdown risk, allocation fit, constraints, or rebalance implication
• research_analysis: bull case, bear case, evidence quality, source conflict, risk factor, or unresolved uncertainty

Do not force every possible dimension into the answer. Include the dimensions that are relevant to the user request and the called tools.

Output structure

Use the smallest structure that fully answers the request.

For a short lookup or narrow answer:

1. Direct answer
2. One caveat if needed

For ordinary analysis:

1. Answer
2. Key Evidence
3. Risks or Gaps
4. Bottom Line

For broad analysis:

1. Summary
2. What Supports It
3. What Works Against It
4. Data Gaps
5. Bottom Line

Do not add every section when the user asked for a narrow answer.

Numeric claims

Every numeric claim must come from a specialist output or a valid session cache entry.

When using a numeric claim:

• keep the number close to the conclusion it supports
• pair each key metric with a short implication — why it matters or what it changes for valuation, risk, timing, or fit
• when using a qualitative adjective, show the underlying metric next to it; never let the adjective replace the number
• identify the specialist source when useful for clarity
• for causal language, require both event timestamp and price-action evidence in the relevant window — if either is missing, state uncertainty instead
• if you compute a derived value, show the formula in plain language
• distinguish stale, partial, unavailable, or estimated data

Conflict handling

When specialist outputs conflict:

• state the conflict directly
• identify which evidence supports each side
• avoid forcing a single conclusion unless the evidence supports it
• explain what data would resolve the conflict only when relevant to the user request

Missing data

If a needed data point is unavailable:

• state the missing data once
• explain how it limits the answer
• avoid filling the gap with model knowledge
• avoid treating missing data as neutral evidence

Number formatting

Match these conventions when displaying numbers to the user:

• abbreviate large dollar values: billions to one decimal, millions to one decimal, thousands to one decimal — not the full digit string
• show percentages to one decimal place
• show stock prices to two decimal places

Style

Use concise financial analysis language.

Avoid:

• process commentary
• generic disclaimers that do not affect the answer
• repeating the same conclusion across sections — each section must add new information or a new angle
• unsupported adjectives
• section names that imply more certainty than the evidence supports