5 Prompts That Improved My AI Agent Workflow: Tested Hacks for Claude Code
Five battle-tested prompts for Claude Code: ask clarifying questions, don't stop till done, show-don't-tell with playground, multi-agent hiring, memory from mistakes.
Prompts that improved my AI agent experience
My personal hacks after intensive work with Claude Code.
1. "Honey, let's discuss this"
Explicitly ask the agent to ask clarifying questions about your idea. Helps articulate what you actually want — getting closer to the expected result. Then plan mode, then execute.
2. "Don't stop until it's done"
Sometimes the agent stops halfway: "I did the task, but create the PR yourself." Nope — pedal to the metal.
3. "Show, don't tell"
Install the official playground skill → get interactive HTML. Improving architecture? Clickable graph. Design? Mockup with sliders. Can even generate videos with Remotion (+ they added prompt examples).
4. "Hire whoever you need"
Claude Code recently added a gastown-like feature (needs manual enabling) where you can set roles for parallel agents. I was too lazy to configure it — and Claude did great on its own.
5. "Review the chat, draw conclusions, remember"
At the end of a session, ask Claude to analyze where you corrected it or it made mistakes. I'm too lazy to explain the same thing repeatedly. Save memories to .md files in the repo — they'll help future sessions.
Original post: @danokhlopkov
Dan Okhlopkov — AI agent practitioner. Building tools for TON Blockchain analysis and Telegram automation.