PROMPTS
Make Claude Catch Its Mistakes ⚡
Asking AI to check its own work in the same chat is useless. The fix is a subagent. Here are the exact prompts I use, ready to copy.
When you ask the same chat that wrote something to go back and check it, it's reading its own work with the same blind spots that made the mistake in the first place. It's like proofreading your own email at midnight. You'll miss the typo every time, and it'll swear it looks perfect.
The fix is a subagent: a second instance of Claude that opens with a totally blank memory and reviews your work like a coworker seeing it for the first time. Fresh eyes catch what the first one talked itself into. Here's exactly how to use it, plus the prompts.
What A Subagent Actually Is
A subagent is the same Claude model, but spawned in its own separate context window with none of the original conversation. Because it can't see the reasoning that produced the work, it can't inherit the same blind spots. And don't let the name fool you, a subagent isn't less capable, it's just another agent doing one focused job.
This lives in Claude Code (Anthropic's coding tool), where you can spawn subagents on command or save your own under the /agents menu. No Claude Code? The same principle works anywhere: open a brand-new chat, paste only the work, and ask it to review cold. Either way, the magic is the clean context.
The Cold Review
The one you'll reach for most. Works on a doc, an email, a plan, a caption, code, anything. Paste your work, then run this.
The Devil's Advocate
Run this before any decision, pitch, or plan. It stress-tests your thinking now, so reality doesn't do it for you later.
The Fact-Checker
AI sounds most confident right when it's wrong. Send a subagent in to verify, not to agree.
Your Reader's Reaction
The fix for anything that makes sense to you but lands flat on everyone else. Just fill in who it's for.
The Expert Panel
Three independent opinions instead of one, and Claude picks the right experts for whatever you're working on. No echo chamber, because none of them can see each other's notes.
The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.