Extending and Contributing to DeerFlow
1. Start with the workflow stage you want to improve​
DeerFlow contributions are easiest when you pick one clear layer:
| Goal | First place to look |
|---|---|
| Change orchestration logic | src/ runtime and graph logic |
| Improve the web experience | web/ |
| Add examples or patterns | examples/ |
| Clarify operations | docs and harness guidance |
2. Why this matters​
Because DeerFlow is step-oriented, it is especially important not to mix planning, evidence gathering, synthesis, and UI changes into one large patch unless they truly belong together.
3. Best first contributions​
The easiest high-value first contributions are usually:
- docs clarifications,
- example workflows,
- better prompts or stage boundaries,
- small integration fixes,
- output-quality improvements.
4. Think in reusable workflows​
A strong DeerFlow contribution usually makes one of these things better:
- research reliability,
- workflow clarity,
- human review ergonomics,
- report quality.
5. Before opening a PR​
Validate the full research flow your change touches. A patch that looks correct in one function but breaks the end-to-end research process is still a real regression.