Who it is forA product minded developer or PM who keeps getting the wrong feature because the prompt was vague.
The problemVague requests get silent assumptions, and you only find the gaps after the code is written.
The outcomeA clear set of requirements gathered one question at a time, ready to feed into /spec.
What you need first
- Node.js for `npx skills`, or Claude Code, Codex, Gemini CLI or Antigravity for their native installs
- Git
- SSH keys for the Claude Code marketplace, or the HTTPS URL workaround in the README
Steps
- Run
npx skills add addyosmani/agent-skills --skill interview-me - Describe the feature in a sentence and ask the agent to use interview-me
- Answer its questions one at a time until it has enough to write a spec
- Run
/specwith the answers so the spec reflects what you actually need - Add
/constraintsto record quality rules once so they are enforced everywhere
Prompt for your agent
Use the interview-me skill from addyosmani/agent-skills to interrogate me about <feature idea>, one question at a time. When you have enough, write the spec with /spec and list every assumption you still made.
Caveat.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for addyosmani/agent-skills and "Have your agent interview you before it writes any code"
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