Key Takeaways
- Claude skills provide structured instructions using SKILL.md files to make recurring workflows more repeatable and consistent across different tasks.
- Setup methods vary by environment, allowing browser users to upload skills through settings, Claude Code users to place files in specific directories, and Cowork users to apply specialist methods to document work.
- Skills differ from temporary prompts because they stay available for repeated use, enforcing stable standards and methodologies over many interactions.
Claude can already help with writing, research, documents and analysis, but skills make those workflows more repeatable. A skill gives Claude structured instructions for a specific type of task instead of making you explain the process again in every conversation. For users who want to explore ready-made options, Agensi is a marketplace for Claude skills, including skills designed for Claude Code and Cowork. The useful part is not collecting as many skills as possible. It is finding one that matches something you already do regularly and adding it to the version of Claude you use.
In This Article
In short: how do you add a skill to Claude?
The setup depends on how you use Claude. Browser users can upload skills through Claude’s settings, Claude Code users place SKILL.md files in a skills directory and Cowork users can use skills to add specialist methods to document and knowledge-work tasks. Once installed, a matching skill provides Claude with the instructions it should follow for that type of work.
A practical setup looks like this:
- Choose one recurring task you want Claude to perform more consistently.
- Find or create a skill for that task.
- Install or upload the skill in the appropriate Claude environment.
- Enable it where necessary.
- Give Claude a task that matches the skill.
- Check whether the result actually follows the intended method.
What is a Claude skill?
A Claude skill is based around structured instructions, commonly stored in a SKILL.md file. The skill describes how Claude should approach a particular job. Instead of merely telling Claude to “write a report”, for example, a reporting skill could define the required sections, checks and output structure.
That makes skills different from ordinary prompts. A prompt is usually temporary and applies to the current interaction. A skill stays available for repeated use, which is useful when the same standard or method should apply across many tasks.
Method 1: add a skill in the Claude browser
Claude Pro users can use skills without working in a terminal. According to Agensi’s setup material, the browser workflow starts by opening Settings > Capabilities and making sure code execution and file creation are enabled. From there, users can open Customize > Skills, where built-in skills can be activated and custom skills can be uploaded.
To upload your own skill:
- Open Customize > Skills.
- Select Upload.
- Choose the skill ZIP or SKILL.md file.
- Wait for it to appear in the skills list.
- Toggle the skill on.
- Start a conversation that matches the skill’s purpose.
When the request matches an active skill, Claude can use the instructions from that skill while completing the task.
Method 2: Install skills in Claude Code
Claude Code uses folders rather than a browser upload flow. Personal skills can be placed in ~/.claude/skills/, while project-specific skills belong in .claude/skills/ inside the repository. Personal skills are available across projects, while project skills can be shared with everyone working on the same codebase.
A typical structure looks like this:
~/.claude/skills/
└── code-reviewer/
└── SKILL.md
After adding a skill, start a new Claude Code session. Claude scans the relevant directories at session startup and makes discovered skills available. Developers should also check that the SKILL.md file is not accidentally nested one folder too deep, which is a common installation problem.
Method 3: use skills with Claude Cowork
Claude Cowork is aimed at broader knowledge work rather than software development. Agensi describes it as a visual environment for tasks involving documents, spreadsheets, presentations, file organisation, research and reporting. Both Cowork and Claude Code support the same broader SKILL.md approach, although they are designed for different types of work.
For Cowork, a skill can act like a reusable professional method. A proposal skill might define how a client proposal should be structured. A research skill could specify how sources should be compared and how findings should be organised. A reporting skill could define the expected sections and checks. The point is not that Cowork suddenly receives access to everything. Skills define methodology, while connectors and permissions determine what Cowork can actually reach.
Which everyday tasks work well with skills?
Skills are not only for programmers. Browser and Cowork users can apply them to tasks where a clear method improves consistency.
| Everyday task | What a skill can add | What you still need to check |
| Writing and editing | Tone, structure and style rules | Accuracy and final wording |
| Research | Research method and reporting format | Source quality and conclusions |
| Presentations | Preferred structure and content checks | Storytelling and visual judgement |
| Document creation | Reusable templates and requirements | Whether the document fits the situation |
| Content repurposing | Rules for adapting one source to several formats | Platform relevance |
| Job applications | CV context, target roles and formatting preferences | Personal accuracy |
Agensi’s material specifically identifies writing, editing, research, document creation, content repurposing and job applications as categories that can work well without a coding environment.
How do you know whether a task deserves a skill?
Not every request needs reusable instructions. A one-off question is usually better handled with a normal prompt. Skills become more useful when you notice yourself repeatedly explaining the same standards.
Three signals make a task a good candidate:
- You repeat it frequently: the same type of work returns every week or month.
- You keep giving Claude the same corrections: the desired method is stable enough to document.
- The result has a recognisable structure: you can describe what good output should contain.
If you repeatedly tell Claude to use the same writing style, follow the same reporting structure or apply the same research checklist, that process may be better suited to a skill.
Should you install many Claude skills at once?
Usually not. Starting with one skill makes it easier to judge whether it actually improves the workflow. Agensi’s guidance for browser users recommends beginning with the task you perform most often rather than enabling a large collection immediately.
Too many overlapping skills can also make it less obvious which instructions Claude should follow. Focused skills with clear purposes are easier to test. Once one workflow produces consistently useful results, add another skill for a different recurring task.
What should you check before installing a third-party skill?
A skill may contain instructions rather than a conventional application, but those instructions still influence agent behaviour. Source and provenance therefore matter. Agensi’s marketplace material describes security scanning and review as part of its curation process, while skills obtained from open repositories may require users to inspect the files themselves.
Before installing a third-party skill, check:
- Purpose: Does the skill clearly explain what it is designed to do?
- Instructions: Do the contents match that stated purpose?
- Permissions: Does the task require the access being requested?
- Supporting files: Are there scripts or references you should inspect?
- Source: Is the creator or marketplace identifiable?
This is especially important for skills used alongside agents that can access files or external tools.
How to test whether your new skill works
Do not judge a skill based on whether Claude mentions that it is installed. Give it a realistic task and inspect the result.
If you installed a research skill, ask for a small research project and check whether the output follows the expected method. If you added a writing skill, compare the result with the structure or style you normally request manually. For Claude Code, test the skill on a low-risk project task before relying on it for production work.
The most useful measure is simple: does the skill save repeated explanation while making the result more consistent?
Start with the job, not the skill collection
Claude skills become useful when they solve a recurring problem. Installing dozens of capabilities because they look interesting can create more clutter than value. Start by identifying one everyday task you already perform with Claude and define what a good result should look like.
From there, a skill can turn that preferred method into something reusable. Whether you use Claude in the browser, Claude Code or Cowork, the principle stays the same: choose a specific job, give Claude a clear procedure and keep human review where judgement still matters.
