Who it is forA designer or engineer who needs several options for a title, idea, or design and wants to make the final call.
The problemOne generated answer rarely gives enough options, and comparing them by hand takes time.
The outcomeYou get four complete candidates shown side by side with their costs, and you pick the one you prefer.
What you need first
- DeepSeek Harness (dsh) installed
- The lowtide plugin installed into dsh
- A model connected to your Harness
- A workspace directory for the project the tasks will touch
Steps
- In the new-task modal, choose the sampling strategy and set the candidate count to 4.
- Describe the design problem clearly in the task prompt.
- File the task so it lands as pending review, then release it.
- Open the execution report once the run finishes.
- Compare the candidates side by side and choose one, since the machine makes no aesthetic judgment.
Prompt for your agent
Using the lowtide plugin in DeepSeek Harness, run the sampling strategy with 4 candidates on the design problem described in <design brief file>. Expected output: four complete candidates shown side by side with their costs in the execution report. Verify that exactly four candidates appear and that each has a cost listed.
Caveat. The README says the sampling strategy costs about N times a single run, so four candidates cost roughly four times as much.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for KelaoHu/dsh-lowtide and "Generate four design candidates and pick the best"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer