How juleslscx/ds-star works
DS-STAR accepts data files and a query, analyzes the data, then iterates through planning, code generation and execution, debugging failures, and verifying/refining results before producing a formatted answer. Its CLI and pipeline controller also support resuming runs, interactive pauses, manual code edits, and artifact persistence. The README documents these capabilities and the provider options; the excerpts show the central pipeline, storage/controller responsibilities, and provider dispatch, but not all agent internals.
Main entry points: dsstar.py Pipeline Controller · provider.py Provider Dispatch · config.yaml Run Configuration · prompt.yaml Agent Prompts
Components 18
Entry and Control
- CLI
dsstar.py - Run Configuration
config.yaml - Pipeline Controller
dsstar.py
Analysis Workflow
- Analyzer
dsstar.py - Planner
dsstar.py - Coder and Executor
dsstar.py - Debugger
dsstar.py - Verifier and Router
dsstar.py - Finalizer
dsstar.py - Agent Prompts
prompt.yaml
Run State
- Run Artifacts
dsstar.py - Code Editing
dsstar.py
Model Providers
- Provider Dispatch
provider.py
Other components
- User
- Data Files
- Gemini
- OpenAI
- Ollama
Connections 18
- User → CLI: submits query
README.md - Data Files → CLI: provides inputs
README.md - Run Configuration → CLI: configures run
README.md - CLI → Pipeline Controller: starts or resumes
dsstar.py - Pipeline Controller → Analyzer: runs analysis
dsstar.py - Pipeline Controller → Planner: requests plans
dsstar.py - Pipeline Controller → Coder and Executor: generates and executes
README.md - Pipeline Controller → Debugger: repairs failures
README.md - Pipeline Controller → Verifier and Router: checks results
dsstar.py - Pipeline Controller → Finalizer: formats answer
dsstar.py - Pipeline Controller → Run Artifacts: saves and resumes
dsstar.py - Pipeline Controller → Code Editing: supports code edits
README.md - Pipeline Controller → Agent Prompts: uses prompts
- Pipeline Controller → Provider Dispatch: dispatches model calls
dsstar.py - Provider Dispatch → Gemini: calls model
README.md - Provider Dispatch → OpenAI: optionally calls
README.md - Provider Dispatch → Ollama: optionally calls
README.md - Verifier and Router → Planner: refines plan
README.md