market-analysis

name	Market Analysis
description	Analyze macroeconomics, policy trends, and industry momentum, then combine the result with market index data for a broader judgment
allowed-tools	["web_search","data_fetch"]

Market Analysis Skill

Use the tools according to the user's question and combine current-turn market quotes, macro evidence, and dated news to provide a deeper market view. This skill covers broad or regional markets and sector or industry themes; it must not fall back to generic chat merely because the request has no single ticker.

This skill must always anchor the analysis to the current session time before making any macro judgment. Treat the session's current time and date as the source of truth for words such as "today", "latest", "this week", "tonight", or "just announced".

Tool Guide

Tool call	Purpose
web_search(query="...")	Fetch macroeconomic data, policy interpretation, and industry developments
data_fetch(data_type="search", query="ticker or name")	Resolve every representative index proxy, ETF, or listed company before using it
data_fetch(data_type="quote", ticker="comma-separated exact symbols")	Fetch same-symbol latest-available price, change, and provider timestamps for representative benchmarks
data_fetch(data_type="sector_performance")	Fetch current sector breadth when the requested scope supports it
data_fetch(data_type="gainers_losers")	Inspect current market leaders/laggards as breadth context, not as entity proof

Analysis Framework

1. Time anchor first and Interactive answer ownership: the main Agent completes one full final answer inside the current-turn tool loop and publishes that completed body once. The Agent itself starts every Interactive market or sector answer with 数据时间：北京时间 YYYY-MM-DD HH:MM；行情口径：..., using the current Beijing time from the Session context and the current-turn quote provider timestamp, market session, and latest-available/non-tick-by-tick basis; do not emit any preamble before it
2. Subject and entity discovery first: identify every requested market scope; broad-market turns resolve representative benchmarks, while sector turns discover listed representatives from current theme evidence and exact-resolve at least three same-theme securities
3. Current quote first: fetch same-symbol quotes and provider timestamps for every representative before analyzing direction; never reuse prior assistant prices or let one market overwrite another in a mixed-scope request
4. Target session before cause: for a decline/rally explanation, preserve the user's explicit date, weekday, and session; verify whether the named market/security actually made that move in that exact interval before explaining it. A latest quote cannot prove another historical session, and a broad index, sector, and single stock are different scopes
5. Query rewrite first: convert relative-time wording into absolute market-local civil dates before calling web_search; keep Beijing time as the user-visible anchor. Civil-calendar hints do not prove that an exchange was open or that the quote is a close
6. Macro and policy: interest rates, inflation, employment, central-bank, fiscal, and regulatory evidence only when relevant
7. Industry and breadth: sector trends, representative-company dispersion, and capital flows
8. Fact/inference split: dated source facts and causal inference must be visibly separate

Broad / Regional Market Output Contract

Use exactly five substantive numbered sections after the Agent-authored time and quote-basis line:

1. Conclusion
2. Verified market facts: one independent line per representative, with exact symbol, current-turn price, change, and quote timestamp basis
3. Market-move reasons: dated verified events with a current-turn source domain, followed separately by causal inference; if evidence is insufficient, say 原因本轮未完全核验
4. Bull / Bear / Base Case and primary risks
5. Action framing, triggers, and falsification conditions

Do not ask “which stock?” as a substitute for a broad-market answer. In mixed-market requests, retain separate local dates, benchmark entities, and evidence for every scope.

Sector / Industry Output Contract

Use exactly nine substantive numbered sections after the Agent-authored time and quote-basis line:

1. What the technology or theme is
2. Its core change versus alternatives
3. Why it matters now and the adoption timeline
4. The next 2–3 years of market space and mainstream views; label unsupported numbers as not verified in the current turn
5. Value-chain layers and bargaining power
6. Listed-company comparison, with an independent current-turn same-symbol quote for every verified representative
7. High-certainty, high-beta, and concept-only mappings
8. Risks and falsification conditions, including Bull / Bear / Base scenarios and verifiable catalysts
9. Final investment framing and trigger conditions

Never use SPY/QQQ or a previous-turn ticker merely to fill a sector list. Every representative must be supported by current theme evidence plus exact-symbol search and quote results.

Mandatory Query-Rewrite Rules

1. If the user asks about macro data, policy headlines, geopolitical events, or uses relative time such as "today", "latest", "this week", or "just announced", first read the current session date.
2. Rewrite the search query into an absolute-time form before search. Do not search with ambiguous wording.
3. The rewritten query must include the exact year, month, and day. Add "latest", "today's release", or the event name only after the absolute date is present.
4. Example rewrite: User asks: "How was today's nonfarm payroll?" Search query should become: "2026-04-04 latest US nonfarm payroll release"
5. For Interactive market and sector answers, the Agent-authored Beijing data-time and quote-basis line is always the first visible line. Do not emit a preamble or a second time line.
6. For "why did the market/stock fall" questions, the absolute search date must be the verified target-session date, not merely the current market-local calendar date. If the quote snapshot belongs to a different date, do not silently move the question to that date.
7. If broad benchmarks do not support the user's "crash/selloff" premise, state the scope mismatch and check the named sector/security or ask one minimal clarification. Never substitute a larger move from another day. If same-date causal evidence remains weak, report the verified move and say 原因本轮未完全核验.

Notes

• Always focus on the dimensions that match the user's question instead of speaking in broad generalities
• Use current-turn DataFetch quotes for market facts and dated web_search evidence for causes; do not expose raw tool names or payloads in the user-facing answer
• Separate hard facts from market expectations or opinions in the final answer
• For macro search, never issue a web_search query that omits the absolute date when the user intent is time-sensitive
• If quotes succeeded, never claim that Hone lacks real-time/current market data or did not request it. Describe the provider result as latest available and non-tick-by-tick.
• A failed news result does not erase valid quotes: report the verified prices and say the cause is not fully verified. A failed quote for one scope does not permit copying another scope's quote.
• If the user asks for a trend, curve, distribution, or side-by-side visual and you already have the numbers, use the Chart Visualization skill instead of describing the chart only in prose. Native Codex loads that skill through its own skill discovery; legacy Hone runners may use skill_tool(skill_name="chart_visualization", execute_script=true, ...).