market-sentinel

Market Sentinel

Scope of responsibility: This Skill only performs the aggregation and interpretation of "four-source sentiment evidence → composite score, phase, counter-evidence, confidence band"; it does not provide abnormal-movement scan lists, does not provide specific buy/sell/add/reduce position/leverage suggestions, and does not replace market-scanner 's time-segment monitoring and event-window output.

With market-scanner: To find "who is surging today, sector abnormal movements, pre-market/intraday scans," use market-scanner; for "overall sentiment phase, framework judgment on whether heavy positions are possible (not instructions)," use this Skill.

Examples of incorrect usage (counterexamples)

• Incorrect: Using this Skill's output to directly place orders or set stop-loss prices. Correct: Use it only for environmental judgment; trading decisions must be borne by the user's strategy and compliance process.
• Incorrect: Treating "a large number of limit-ups" alone as the basis for heavy positions while ignoring capital flow and northbound counter-evidence. Correct: All four tools must be run in parallel to align evidence, and contradictions must be stated in risk_counterevidence .
• Incorrect: After market-scanner has already provided an abnormal-movement list, using this Skill again to request "today's hot sector rankings." Correct: Sector-ranking scans are left to market-scanner; this Skill only needs to read the sector structure sub-score.

Objective

Within a single day or a short lookback window, provide aninterpretable comprehensive assessmentof A-share market sentiment: composite score, phase label, dominant factors, and risk counter-evidence; does not replace scanning-type Skills.

Input

• User question (may include date、lookback_days、risk_mode）
• structured evidence returned by the four sentiment tools

Output (fixed structure)

1. Composite score and phase (0–100 score + freezing point/repair/climax/ebb/oscillation/chaos)
2. Sub-item scores and attribution of dominant factors
3. Data completeness ratio and (if any) degradation notes
4. Action-tendency framework and confidence band

Evidence table (required)

• List the four dependent tool calls and their respective quality_status; the composite score must be traceable to each sub-score source.

Counter-evidence and limitations (required)

• Contradictions among the four sources, and the explanation of dilution of the composite score when a certain source degraded .

Mandatory rules

• Call dependent tools only through manifest / tool_runner ; direct connections are prohibited plugins.data_collection。
• Must call in parallelthe four dependent tools (asynchronous/concurrent); serial blocking for convenience is prohibited; aggregated results may be reused via config/market-sentinel_config.yaml 's short TTL cache.
• When core evidence is insufficient to support a conclusion, output insufficient_evidence and explain the missing items.
• Outputting specific buy/sell points, position ratios, or leverage suggestions is prohibited.
• Weights,risk_mode, dynamic reweighting rules, and phase thresholds are read only from config/market-sentinel_config.yaml .

Dependent tools

• tool_resolve_symbol(L2; when the code is in mixed format)
• tool_fetch_limit_up_stocks
• tool_fetch_a_share_fund_flow
• tool_fetch_northbound_flow
• tool_fetch_sector_data

Common output fields

• overall_score
• sentiment_stage
• sub_scores
• factor_attribution
• data_completeness_ratio
• action_bias
• risk_counterevidence
• confidence_band
• degraded