Market Research Copilot

L2 composite skill — orchestrates 12 read-only MCP Tool Calls across Gate-Info MCP (8 tools) and Gate-News MCP (4 tools) to aggregate multi-source market intelligence into structured, citation-backed reports. No trading, swaps, staking, or any fund-moving operations.

Trigger Scenarios: User intent is "research / understand / analyze" — market overview, single-coin analysis, multi-coin comparison, technical trends, event attribution, risk check, or daily briefing.

Target Users: Casual / beginner traders, content-oriented users, and pre-decision users who need to quickly understand the market, a coin, or risk before acting.

Read and follow exchange-runtime-rules.md first.

Domain Knowledge

Common Misconceptions

Design Rationale: 5-Dimension Signal Stacking

User queries often span multiple analysis dimensions (e.g., "How is BTC doing?" implies fundamentals + technicals + news). Forcing a single-intent classification is fragile. This Skill uses dimension signal stacking instead of N-choose-1 classification: independently detect 5 signal dimensions from the query, each activated dimension contributes a set of Tool Calls, and the final tool set is their union. Running 5 independent binary classifiers is far more stable than 1 nine-way classifier.

Tool Inventory (12 unique tools, all read-only, no auth required)

L1 Skill to Tool Mapping

Key Parameter Notes

• info_marketsnapshot_get_market_snapshot: Always pass source="spot" to avoid non-standard pair data contamination.
• info_markettrend_get_kline: Always pass source="spot" — without this constraint, price data may include DEX non-standard pairs, producing incorrect prices. This is a mandatory parameter.
• info_compliance_check_token_security: Requires chain parameter; skip for major coins (BTC/ETH) where contract audit is not applicable.
• info_onchain_get_token_onchain: Some tokens return empty results — this is expected; skip the on-chain section and annotate "On-chain data unavailable".

Workflow

MCP tool calls: Only Step 3 invokes Gate-Info / Gate-News MCP tools (info_*, news_*). Steps 1, 2, 4, and 5 are orchestration, signal planning, aggregation, and formatting — no MCP tool call in those steps.

Step 1: Intent Gate — Research vs. Execution

Determine if the user wants to research or execute.

• Research intent (enter this Skill): understand, analyze, compare, briefing, risk check, trend, why pumping/dumping, sentiment, fundamentals, worth buying, overview, report.
• Execution intent (route away): buy, sell, swap, stake, transfer, withdraw, place order, close position.

If the user mixes both (e.g., "analyze BTC then buy"), split into two phases: complete the research phase within this Skill, then route the execution phase to the appropriate trading skill (e.g., gate-exchange-spot).

Key data to extract:

• intent_type: "research" or "execution"

Step 2: Parameter Extraction & 5-Dimension Signal Detection

Extract parameters from the user query:

• symbols[]: Coin tickers (0 to N). Map project names to tickers (Solana -> SOL, Chainlink -> LINK).
• sector: Optional sector (DeFi, Layer2, Meme, etc.)
• time_range: Time scope (default 24h)

Then independently evaluate each of the 5 signal dimensions (non-exclusive, multiple can fire):

Fallback rules (when no signal is explicitly activated):

• symbols == 0 -> activate S1 (market overview)
• symbols == 1 -> activate S2 + S3 (single-coin comprehensive = fundamentals + technicals)
• symbols >= 2 -> activate S2 x N (multi-coin comparison)
• symbols >= 2 and the user semantically asks for a comprehensive / holistic multi-coin view (same spirit as single-coin S2+S3) -> also activate S3 x N (i.e. S2 x N + S3 x N)

Screening mode (when symbols == 0 and query contains "filter", "find", "recommend", "which", "ranking", "top"):

• Phase 1: Activate S1 to fetch top-coin snapshots
• LLM intermediate aggregation: Extract symbols from Phase 1 results
• Phase 2: Activate S2/S3/S4/S5 per remaining signal keywords

Key data to extract:

• symbols[]: List of coin tickers
• sector: Sector filter if any
• time_range: Time scope
• activated_signals[]: List of activated signal dimensions (S1-S5)
• screening_mode: Boolean

Signal Routing Examples:

Step 3: Assemble Tool Set & Execute

Combine all activated signal tool sets into a union, deduplicate by tool name + parameters, then execute:

Parallel execution (tools within the same Phase have no dependencies — call simultaneously):

• All tools in a Phase are dispatched in parallel.

Serial dependencies (next Phase depends on previous Phase LLM aggregation):

• Screening mode: Phase 1 snapshot results -> LLM extracts symbols -> Phase 2 uses those symbols.
• Attribution branch (S4): news_events_get_latest_events result determines whether to call news_events_get_event_detail (event found) or skip (no event).

Deduplication rule: Same tool + same parameters within one request is called only once; the result is reused across multiple aggregation steps.

Call each activated tool with the parameter conventions below:

Call info_marketsnapshot_get_market_snapshot with: symbol={symbol}, timeframe="1d", source="spot" Call info_coin_get_coin_info with: query={symbol}, scope="full" Call info_markettrend_get_kline with: symbol={symbol}, timeframe="1d", limit=90, source="spot" Call info_markettrend_get_indicator_history with: symbol={symbol}, indicators=["rsi","macd","bollinger","ma"] Call info_markettrend_get_technical_analysis with: symbol={symbol} Call info_platformmetrics_get_defi_overview with: category="all" Call info_compliance_check_token_security with: token={symbol}, chain={chain} Call info_onchain_get_token_onchain with: token={symbol} Call news_events_get_latest_events with: time_range={time_range}, limit=10 Call news_events_get_event_detail with: event_id={event_id} Call news_feed_search_news with: coin={symbol}, limit=5, sort_by="importance" Call news_feed_get_social_sentiment with: coin={symbol}

Key data to extract:

• tool_results{}: Map of tool name to response data
• failed_tools[]: Tools that timed out or returned errors

Step 4: LLM Multi-Source Aggregation

Pass all tool responses to the LLM for aggregation:

1. Merge data from all dimensions into a coherent narrative.
2. Deduplicate overlapping information (e.g., price from multiple snapshot calls).
3. Conflict resolution: If sources contradict each other, present both viewpoints with attribution — never merge contradictory conclusions into one sentence.
4. Data timestamp: Label each data point with its source tool and approximate data time.
5. Missing data: If a tool failed or returned empty, annotate "Data unavailable for this section" — never fabricate data.

Key data to extract:

• aggregated_report: Structured report content
• data_sources[]: List of tools that contributed data
• data_timestamps: Approximate data freshness

Step 5: Output Structured Report

Select the appropriate report template based on the activated signals, then output the final report. Always include the disclaimer at the end.

Key data to extract:

• final_report: Markdown string — user-facing report body (template-filled; may refine aggregated_report from Step 4); must include the standard disclaimer.

Report Template

Template A: Market Overview Brief (S1 dominant)

## Market Overview Brief

**Data Time**: {timestamp}

### Market Snapshot

| Metric | Value | 24h Change |
|--------|-------|------------|
| BTC | ${btc_price} | {btc_change_24h}% |
| ETH | ${eth_price} | {eth_change_24h}% |
| Total Crypto Market Cap | ${total_mcap} | {mcap_change}% |
| Fear & Greed Index | {fear_greed} | {sentiment_label} |

### DeFi Overview

| Metric | Value |
|--------|-------|
| Total TVL | ${tvl} |
| DEX 24h Volume | ${dex_volume} |
| Stablecoin Total Supply | ${stablecoin_supply} |
| Long/Short Ratio | {long_short_ratio} |

### Key Events (Past 24h)

1. **{event_title}** — {impact_summary} (Source: `news_events_get_latest_events`)
2. ...

### Market Sentiment

- Social Sentiment Score: {score} ({bias})
- Key Discussions: {summary of trending topics}

### Summary & Outlook

{LLM 3-5 sentence synthesis: current market phase, primary drivers, key risks}

> Data sourced from Gate-Info MCP and Gate-News MCP. Analysis is data-driven and does not constitute investment advice.

Template B: Single-Coin Deep Dive (S2+S3 dominant)

## {symbol} Research Report

**Data Time**: {timestamp}

### 1. Fundamentals

| Metric | Value |
|--------|-------|
| Project | {project_name} |
| Sector | {category} |
| Market Cap | ${market_cap} |
| FDV | ${fdv} |
| Circulating / Total Supply | {circulating} / {total} |
| Key Investors | {investors} |

{Tokenomics summary; flag upcoming large unlocks}

### 2. Market & Technical Analysis

| Metric | Value | Status |
|--------|-------|--------|
| Price | ${price} | — |
| 24h Change | {change_24h}% | {Up/Down/Flat} |
| 7d Change | {change_7d}% | {Up/Down/Flat} |
| RSI(14) | {rsi} | {Overbought/Oversold/Neutral} |

**Technical Signals** (from `info_markettrend_get_technical_analysis`):
- Short-term (1h/4h): {signal}
- Medium-term (1d): {signal}
- Support: ${support} | Resistance: ${resistance}

### 3. News & Sentiment

**Recent News**:
1. {title} — {summary} ({source}, {time})
2. ...

**Social Sentiment**: {sentiment_score} ({bias}) — {summary}

### 4. Risk Alerts

{Data-driven alerts: RSI extremes, volume anomalies, funding rate, upcoming unlocks, low liquidity}

### 5. Overall Assessment

{LLM 3-5 sentence synthesis: market phase, drivers, risks}

> Data sourced from Gate-Info MCP and Gate-News MCP. Analysis is data-driven and does not constitute investment advice.

Template C: Multi-Coin Comparison (S2 x N)

## Multi-Coin Comparison: {coin1} vs {coin2} vs ...

**Data Time**: {timestamp}

### Comparison Table

| Metric | {coin1} | {coin2} | {coin3} |
|--------|---------|---------|---------|
| Price | ${price1} | ${price2} | ${price3} |
| Market Cap | ${mcap1} | ${mcap2} | ${mcap3} |
| 24h Change | {chg1}% | {chg2}% | {chg3}% |
| RSI(14) | {rsi1} | {rsi2} | {rsi3} |
| Sector | {cat1} | {cat2} | {cat3} |
| Risk Level | {risk1} | {risk2} | {risk3} |

### Per-Coin Highlights

**{coin1}**: {1-2 sentence highlight}
**{coin2}**: {1-2 sentence highlight}

### Comparative Assessment

{LLM analysis: relative strengths/weaknesses, risk comparison, suitability for different strategies}

> Data sourced from Gate-Info MCP and Gate-News MCP. Analysis is data-driven and does not constitute investment advice.

Template D: Event Attribution (S4 dominant)

## Market Movement Attribution

**Data Time**: {timestamp}

### What Happened

{Brief description of the price movement or market event}

### Root Cause Analysis

1. **{event/cause 1}**: {explanation with data} (Source: `news_events_get_latest_events`)
2. **{event/cause 2}**: {explanation}

### On-Chain Evidence

{On-chain flow data if available, otherwise "On-chain data unavailable"}

### Current Sentiment

- Social Score: {score} ({bias})
- Recovery Signal: {recovering/still negative/neutral}

### Outlook

{LLM forward-looking assessment based on event + sentiment + technical data}

> Data sourced from Gate-Info MCP and Gate-News MCP. Analysis is data-driven and does not constitute investment advice.

Template E: Risk Check (S5 dominant)

## {symbol} Security & Risk Report

**Data Time**: {timestamp}

### Contract Security Audit

| Check | Result | Detail |
|-------|--------|--------|
| Honeypot | {Yes/No} | {detail} |
| Hidden Owner | {Yes/No} | {detail} |
| Buy/Sell Tax | {buy_tax}% / {sell_tax}% | {normal/abnormal} |
| Holder Concentration | {top10_pct}% | {concentrated/distributed} |
| Overall Risk Level | {Low/Medium/High} | {flags_count} flags |

### Fundamentals Cross-Check

| Metric | Value |
|--------|-------|
| Project | {project_name} |
| Market Cap | ${market_cap} |
| Active Addresses | {address_count} |

### Risk Summary

{LLM synthesis: overall risk assessment, specific concerns, red flags}

> Data sourced from Gate-Info MCP and Gate-News MCP. Analysis is data-driven and does not constitute investment advice.

Judgment Logic Summary

Error Handling

L1/L2 Routing Strategy

When a user query can be handled by a single L1 Skill's tool set (e.g., "How is BTC?" only needs coinanalysis's 5 tools), this L2 still serves as the unified entry point for all "research" intents:

• Short-circuit optimization: If the activated signals map to only one L1's tools, call only that subset and produce the report — no need to invoke all 12 tools.
• Principle: Unified entry via this L2 ensures consistent "research" experience; internally, 1 to 12 tools are called as needed.

DEX / On-Chain Boundary Handling

This Skill uses only Gate-Info MCP and Gate-News MCP. It does not include gate-dex-* or on-chain address tracking capabilities.

No Confirmation Required

This L2 does not involve any trading confirmation mechanism because:

• All 12 Tool Calls are read-only queries (info_* / news_*); zero cex_* write operations
• No asset changes, order placement, transfers, or mode switches
• No API Key / Secret / OAuth2 required

If the user appends execution intent (e.g., "analyze then buy"), this Skill completes the research phase only. The execution phase is routed to the appropriate trading skill (e.g., gate-exchange-spot), which applies its own Action Draft -> User Confirm -> Execute flow.

Tool Degradation Matrix

Cross-Skill Routing

Safety Rules

1. No investment advice: All output is data-driven analysis. Every report must include the disclaimer: "does not constitute investment advice."
2. No price predictions: Do not output specific target prices or directional predictions (e.g., "will reach $100K").
3. No fund operations: This Skill calls only read-only info_* and news_* tools. Never call cex_* tools.
4. Data transparency: Label data source (tool name) and approximate data time for each section.
5. No data fabrication: When a tool returns empty or fails, clearly state "Data unavailable" — never fabricate numbers.
6. Conflict disclosure: When sources contradict, present both sides with attribution rather than silently picking one.
7. No auth required: All 12 tools are public read-only; no API Key or OAuth2 is needed.