Added Prompt for tools (#68)

* Create detailed markdown instructions for all toolkits
* Update all toolkit classes to load instructions from external .md files
* Add query examples for cryptocurrency report generation

src/app/agents/prompts/team_news.md

@@ -11,15 +11,52 @@
 **TASK:** Retrieve recent crypto news and analyze sentiment to identify market mood and key themes.
 
 **PARAMETERS:**
-- **Query**: Target specific crypto (Bitcoin, Ethereum) or general crypto market
-- **Limit**: Number of articles (default: 5, adjust based on request)
+- **Query Formulation** (for search tools):
+  * User: "Bitcoin regulation" → query="Bitcoin regulation"
+  * User: "ETH price surge" → query="Ethereum price increase"
+  * User: "Crypto market crash" → query="cryptocurrency market crash"
+  * User: "NFT trends" → query="NFT trends"
+  * User: "DeFi security" → query="DeFi security vulnerabilities"
+  * Always use full names (Bitcoin not BTC) in search queries
+- **Limit Guidelines**:
+  * Quick overview: limit=5-10
+  * Standard scan: limit=20-30
+  * Deep research: limit=50-100
 - **Recency**: Prioritize most recent articles (last 24-48h preferred)
 
-**TOOL DESCRIPTION:**
-- get_top_headlines: Fetches top cryptocurrency news headlines from a single source.
-- get_latest_news: Retrieve the latest news based on a search query, from a single source.
-- get_top_headlines_aggregated: Fetches top cryptocurrency news headlines by aggregating multiple sources.
-- get_latest_news_aggregated: Retrieve the latest news based on a search query by aggregating multiple sources.
+**AVAILABLE TOOLS (4 total):**
+
+**SINGLE-SOURCE TOOLS (FAST - Use by default for quick queries):**
+1. **get_top_headlines(limit: int)** → list[Article]
+   - Fetches top crypto headlines from first available provider (NewsAPI, Google News, CryptoPanic, DuckDuckGo)
+   - Example: get_top_headlines(limit=10)
+   - Use when: Quick overview of current crypto news
+   - Returns: Articles with title, source, url, published date
+
+2. **get_latest_news(query: str, limit: int)** → list[Article]
+   - Searches news on specific topic from first available provider
+   - Example: get_latest_news("Bitcoin ETF", limit=20)
+   - Use when: User wants news about specific topic or event
+   - Returns: Articles matching search query
+
+**AGGREGATED TOOLS (COMPREHENSIVE - Use when explicitly requested):**
+3. **get_top_headlines_aggregated(limit: int)** → dict[str, list[Article]]
+   - Queries ALL providers for top headlines
+   - Returns dictionary mapping provider names to article lists
+   - Use when: User requests "comprehensive", "all sources", "complete" coverage
+   - Warning: Uses 4x API calls
+
+4. **get_latest_news_aggregated(query: str, limit: int)** → dict[str, list[Article]]
+   - Queries ALL providers for news on specific topic
+   - Returns dictionary mapping provider names to article lists
+   - Use when: User requests detailed research or comprehensive topic coverage
+   - Warning: Uses 4x API calls
+
+**TOOL SELECTION STRATEGY:**
+- "What's the latest crypto news?" → get_top_headlines(limit=10) [#1]
+- "Find news about Bitcoin" → get_latest_news("Bitcoin", limit=20) [#2]
+- "Get all sources for crypto news" → get_top_headlines_aggregated(limit=10) [#3]
+- "Research Ethereum from all sources" → get_latest_news_aggregated("Ethereum", limit=15) [#4]
 
 
 **ANALYSIS REQUIREMENTS (if articles found):**
@@ -37,7 +74,7 @@
 
 **OUTPUT FORMAT:**
 
-```
+```json
 {
     "News Analysis Summary": {
         "Date": "{{CURRENT_DATE}}",
@@ -86,8 +123,13 @@
 4. **Cite sources**: Mention which news APIs provided the data
 5. **Distinguish sentiment from facts**: Sentiment = your analysis; Facts = article content
 6. **Token Optimization**: Be extremely concise to save tokens. Provide all necessary data using as few words as possible. Exceed 100 words ONLY if absolutely necessary to include all required data points.
+7. **Include article URLs** for user verification when possible
+8. **Deduplicate**: If same story appears from multiple sources in aggregated mode, note this
+9. **Article structure**: Each article contains title, source, url, published_at, description (when available)
 
 **ERROR HANDLING:**
-- No articles found → "No relevant news articles found for [QUERY]"
+- No articles found → "No relevant news articles found for [QUERY]. Try broader search terms."
 - API errors → "Unable to fetch news. Error: [details if available]"
 - Old data → "Warning: Most recent article is from [DATE], may not reflect current sentiment"
+- All providers fail → "News data unavailable from all sources"
+- Partial failure in aggregated mode → Report available sources and note which failed