Anthropic just released ALL the Claude Code secrets Their Prompting best practices just went live in their do...

Anthropic just released ALL the Claude Code secrets

Their Prompting best practices just went live in their docs and I spent hours reading it and testing out all the tips

Here are the 10 that make Claude Code so much better: 🧵 

🔶 Use Github early and often

Once you connect CC to git, it gets access to your entire commit history

Meaning it will know about EVERY change you've ever committed, making it SO much more powerful

This will reduce hallucinations a ton. Connect CC to git right away 

🔶Load up your context

CC's context management has gotten MASSIVE improvements the last few weeks

It now compacts intelligently, allowing you to keep context for a LONG time

Don't be afraid to load context by having CC review your architecture at the beginning of sessions 

🔶Adjust eagerness

Claude Opus 4.5 is an incredibly eager model. It asks for forgiveness, not permission

The eagerness to make changes is why I love CC. But if you don't like this eagerness, Anthropic wrote a new CLAUDE rule you can put in your rules file: 

🔶Be careful with the word 'think'

'Think' is the only word you can use with Claude Code that triggers extra token usage

If you are tight on budget or tokens, avoid using the word 'think' 'think more' or 'ultrathink'

Instead use 'evaluate' or 'consider'

🔶Use vision

Claude Opus 4.5 is the greatest vision model of all time

Meaning, it can analyze and understand images better than any other model

Make sure to use as many images as you can. I like to use them for debugging and visual inspiration

Just paste directly into CC!

🔶Parallel tool calls

Claude Opus 4.5 is the best model ever at tool usage. It's incredibly agentic

With this new CLAUDE rule Anthropic created, you can have CC use tools in parallel, increasing efficiency

Just place this in CLAUDE .MD: (give image to claude to extract text) 

🔶 Reduce hallucinations

Claude's eagerness has some downsides

It will often make edits to files without truly understanding its purpose first

This new CLAUDE .md rule straight from Anthropic will fix that: 

Implement these tips, and Claude Code will perform so much better for you!

There's much more to their prompting guide, I'll link it in the next post.

Make sure to hit like on the first post if you learned anything!

https://x.com/AlexFinn/status/2007585393584353688?s=20

The full Anthropic prompting guide for Claude Code:

https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-4-best-practices#research-and-information-gathering

I started a Claude Code community if you want to join. Over 100 people in it. Tons of Claude Code videos plus weekly calls with everyone. Great place to learn vibe coding and meet builders:

http://vibecodingacademy.dev