Data Flow
Quick Reference
- Pattern: Request-Response + State Machine (Consensus)
- Protocol: JSON-RPC (internal agent communication)
- Serialization: JSON (standard) / Markdown-wrapped JSON (robust)
End-to-End Data Flow
The flow of data from raw user intent to persisted validated state.
mermaid
graph TB
Input["📥 User Request (String)"] --> Parse["⚙️ Analysis (DAG JSON)"]
Parse --> Reasoning["⚖️ Debate Logic (Argument Map)"]
Reasoning --> Transform["🛠️ Execution (Task Results)"]
Transform --> Persistence["🗄️ Memory (SQLite)"]
Persistence --> Output["📤 Final Artifacts"]Main Workflows
Task Result Injection
When executing a DAG, results from upstream tasks are injected into downstream task parameters to enable data-driven swarming.
mermaid
sequenceDiagram
participant E as Execution Engine
participant A1 as Task 1 (Coding)
participant A2 as Task 2 (CLI)
E->>A1: Execute(context)
A1-->>E: { "status": "success", "file_path": "auth.py" }
Note over E: Inject results into Task 2
E->>A2: Execute(context + Task 1 results)
Note right of A2: Uses 'auth.py' for testing
A2-->>E: { "status": "success", "test_report": "..." }- Execution Engine identifies Task 1 as ready.
- CodingAgent completes Task 1 and returns a structured dictionary.
- Execution Engine maps Task 1 output to Task 2 input based on the DAG dependencies
(execution/engine.py:75). - CLIAgent receives the context and performs actions on the newly created artifacts.
External Integrations
| Service | Protocol | Direction | Data | Rate Limit |
|---|---|---|---|---|
| Google Vertex AI | HTTPS (SDK) | Outbound | Prompts/Completion | ~20 RPM (Free Tier) |
| Mistral AI | HTTPS (SDK) | Outbound | Prompts/Completion | ~10 RPM |
| Kaggle API | HTTPS (CLI) | Outbound | Notebooks/Submissions | Per Kaggle limits |
Reasoning Traces
The reasoning_trace column in the decisions table provides a chronological dump of how the system arrived at a conclusion.
| Step | Action | Origin | Sample Payload |
|---|---|---|---|
| 1 | start_analysis | AnalysisAgent | {"request": "..."} |
| 2 | start_challenge | CriticAgent | {"logical_flaws": [...]} |
| 3 | start_mitigation | SolutionAgent | {"revised_plan": "..."} |