Better Prompt (#45)
* Enhance report generation and team leader prompts with conditional output rules for empty sections. * Update .gitignore and configuration for model adjustments; refine query classification prompt * Enforce max response length in prompts * Documentazione tool essenziale per il loro utilizzo da parte degli LLM. * istruzioni chiare sui tool disponibili nei promt degli agenti * corretto incongruenze nel report generation prompt * convertito i promt in file markdown
This commit was merged in pull request #45.
This commit is contained in:
80
docs/Flow_Sequence_Diagrams.md
Normal file
80
docs/Flow_Sequence_Diagrams.md
Normal file
@@ -0,0 +1,80 @@
|
||||
# Diagrammi di Flusso e Sequenza (Sintesi)
|
||||
|
||||
Documentazione breve con blocchi testuali e mermaid per flussi principali.
|
||||
|
||||
## Flusso Gradio Chat
|
||||
|
||||
```mermaid
|
||||
flowchart LR
|
||||
U[User] --> CH(ChatInterface)
|
||||
CH --> RESP[gradio_respond]
|
||||
RESP --> PL(Pipeline.interact)
|
||||
PL --> WF(Workflow run)
|
||||
WF --> OUT(Report)
|
||||
CH --> HIST[history update]
|
||||
```
|
||||
|
||||
## Flusso Telegram Bot
|
||||
|
||||
```
|
||||
/start
|
||||
│
|
||||
├─> CONFIGS state
|
||||
│ ├─ Model Team ↔ choose_team(index)
|
||||
│ ├─ Model Output ↔ choose_team_leader(index)
|
||||
│ └─ Strategy ↔ choose_strategy(index)
|
||||
│
|
||||
└─> Text message → __start_team
|
||||
└─ run team → Pipeline.interact_async
|
||||
├─ build_workflow
|
||||
├─ stream events (Query Check → Gate → Info Recovery → Report)
|
||||
└─ send PDF (markdown_pdf)
|
||||
```
|
||||
|
||||
## Pipeline Steps (Workflow)
|
||||
|
||||
```mermaid
|
||||
flowchart TD
|
||||
A[QueryInputs] --> B[Query Check Agent]
|
||||
B -->|is_crypto true| C[Team Info Recovery]
|
||||
B -->|is_crypto false| STOP((Stop))
|
||||
C --> D[Report Generation Agent]
|
||||
D --> OUT[Markdown Report]
|
||||
```
|
||||
|
||||
## Team Leader Loop (PlanMemoryTool)
|
||||
|
||||
```
|
||||
Initialize Plan with tasks
|
||||
Loop until no pending tasks:
|
||||
- Get next pending task
|
||||
- Dispatch to specific Agent (Market/News/Social)
|
||||
- Update task status (completed/failed)
|
||||
- If failed & scope comprehensive → add retry task
|
||||
After loop:
|
||||
- List all tasks & results
|
||||
- Synthesize final report
|
||||
```
|
||||
|
||||
## Tools Aggregazione
|
||||
|
||||
```mermaid
|
||||
flowchart LR
|
||||
TL[Team Leader] --> MT[MarketAPIsTool]
|
||||
TL --> NT[NewsAPIsTool]
|
||||
TL --> ST[SocialAPIsTool]
|
||||
MT --> WH(WrapperHandler)
|
||||
NT --> WH
|
||||
ST --> WH
|
||||
WH --> W1[Binance]
|
||||
WH --> W2[Coinbase]
|
||||
WH --> W3[CryptoCompare]
|
||||
WH --> W4[YFinance]
|
||||
WH --> N1[NewsAPI]
|
||||
WH --> N2[GoogleNews]
|
||||
WH --> N3[CryptoPanic]
|
||||
WH --> N4[DuckDuckGo]
|
||||
WH --> S1[Reddit]
|
||||
WH --> S2[X]
|
||||
WH --> S3[4chan]
|
||||
```
|
||||
Reference in New Issue
Block a user