Claude Skills: what they are, which to enable, and when to use them
Skills give Claude superpowers — web search, code execution, file creation. Here is which ones matter, how to set them up, and when to turn them off.
Skills are the capabilities you enable inside Claude.ai. Each one gives Claude access to a specific tool — searching the web, running code, creating files, generating images. They are not on by default (some are, depending on your plan). Understanding which to use when is one of the quickest ways to get better results.
The skills that matter most for operators
Web Search. Claude searches the internet and incorporates current information into its response. Turn this on when you need up-to-date information: current pricing, recent news, live documentation. Turn it off when you are working with your own documents and don't want Claude pulling in external context that might conflict.
Code Execution. Claude writes and runs Python code in a sandbox. Essential for data analysis: paste a CSV, ask Claude to analyse it, and it writes and executes the code to give you actual numbers — not estimates. If your work involves any data, this should be on.
File Creation. Claude produces downloadable files — spreadsheets, documents, presentations. The office document skills (PPTX, XLSX, DOCX) are Anthropic-managed skills that produce professional-quality files. If you need Claude to create a deliverable, not just text in a chat window, enable this.
Deep Research. Claude spends extended time crawling multiple web sources and produces a cited research report. Use this for thorough investigation, not quick questions. It takes longer and uses more tokens, but the output is qualitatively different — comprehensive, multi-source, cited.
How to set them up
In Claude.ai, skills are managed per conversation or per Project:
Per conversation: Click the skills icon in the message bar. Toggle on what you need. These settings apply to this conversation only.
Per Project: In your Project settings, enable the skills that everyone using this Project should have access to. This is better for team consistency — the marketing Project always has web search on, the data analysis Project always has code execution on.
When to turn skills OFF
This is the part most people miss. Skills add tokens and can introduce noise.
Turn off web search when working with internal documents. If you uploaded your product documentation to a Project and want Claude to answer from that, web search can pull in conflicting or outdated information from the internet. Disable it to keep Claude focused on your content.
Turn off connectors you are not using. Each active connector is a potential source of context that consumes tokens. If you are not actively using your Jira integration in this conversation, disable it.
Turn off code execution for pure writing tasks. It won't interfere much, but keeping your skill set focused keeps Claude's behaviour predictable. If you are writing blog posts, you don't need code execution trying to run things.
Custom skills
Beyond Anthropic's built-in skills, you can create custom SKILL.md files — instruction sets that teach Claude specialised workflows. A custom skill might define how your team writes product briefs, how to format customer reports, or a specific analysis methodology.
Custom skills live in your Project. They are essentially structured system prompts — but scoped to a specific capability rather than general behaviour.
The practical setup for a team admin
If you are rolling Claude out to a team:
- Create one Project per team function
- Enable the relevant skills per Project (CS gets web search + file creation; data team gets code execution; marketing gets web search + deep research)
- Disable skills that don't apply — it reduces confusion and token waste
- Document which skills are enabled and why, so team members don't randomly toggle things
The goal: each team member opens their Project and has exactly the capabilities they need, nothing more.