We put together a prompting guide for Claude Opus 4.5 based on extensive internal testing by our research and applied AI teams.
Here's what we've learned so far about getting the best results:
1. The new effort parameter is powerful because it controls approximately how many Claude will use for an output.
You can trade off intelligence for cost/latency with a single dial. Works on all tokens including thinking, responses, and tool calls.
2. Tool triggering rates may change. Opus 4.5 is more responsive to system prompts, so if your old prompts used aggressive language to reduce undertriggering, you may now see overtriggering.
Dial back "CRITICAL: You MUST use this tool" to just "Use this tool when..."
Instead of loading all tool definitions upfront, Claude discovers tools on-demand. Mark tools with defer_loading: true and only pays tokens for tools Claude actually needs.
Up to an 85% token reduction and big boost in accuracy on our MCP evals (79.5% to 88.1%)
Programmatic Tool Calling
Claude orchestrates tools through code instead of individual round-trips. It writes Python, processes outputs in a sandbox, and controls what enters context.
Today we're introducing Skills in claude dot ai, Claude Code, and the API.
Skills let you package specialized knowledge into reusable capabilities that Claude loads on demand as agents tackle more complex tasks.
Here's how they work and why they matter for the future of agents:
At a high level, the best analogy I've heard for Skills is something like Neo learning Kung Fu in seconds in the Matrix.
We're "loading in" specialized knowledge to our general agents at runtime.
At their core, Skills are simple. They're just a folder with a file.
The file starts with a name and description, then contains instructions, code, and resources. This simplicity means anyone can now specialize Claude without building custom agents. SKILL.md
We’re running a “Built with Claude Sonnet 4.5” challenge.
We want to see the coolest things you can build with 4.5 in the next week.
Four winners will receive one year of Claude Max 20x and $1k in Claude API credits.
We will select four winners:
“Keep Coding” Award - most technically impressive implementation
“Keep Researching” Award - most compelling exploration of a topic
“Keep Learning” Award – best educational application
“Keep Creating” Award – most artistic use-
To enter, quote post the first tweet of this thread through October 7 with what you built with Claude 4.5:
- How you built it (prompts, agents, MCP servers, workflows)
- Screenshots or demos
- Must be your own work, built with Claude Sonnet 4.5 (Claude.ai, Claude app, Claude Code, Claude Code SDK)
- We will select winners based on ingenuity, creativity, and community response.
We’ve rolled out another update to Claude Code to help customize your workflows: Hooks.
Hooks are user-defined shell commands that execute at various points in Claude Code’s agent loop.
They give you deterministic control over Claude Code’s behavior to ensure certain actions always happen at certain times.
You can create hooks for:
- Notifications (e.g. via Slack) on prompt completions
- Logging and observability
- Custom permissions and approvals
- Running lints after every write
We've simplified local MCP usage by creating something new we call Desktop Extensions (.dxt files).
These package your local server, handle dependencies, and provide secure configuration so you can one-click share and install local servers on Claude Desktop and other apps.
dxt's are zip archives containing the local MCP server as well as a manifest.json, which describes everything Claude Desktop and other apps supporting desktop extensions need to know.