Claude Skills vs Custom Instructions
People put the wrong things in the wrong place constantly — then wonder why their agent is inconsistent. Here's the actual line between them.
The test that cuts through the confusion
One question:
Is this a preference or a process?
Preferences → Custom Instructions
- • "Always use British English"
- • "Keep responses concise"
- • "Use TypeScript not JavaScript"
- • "Never add emojis"
Processes → Skills
- • "When reviewing code, check X then Y then Z"
- • "For contract review, flag these clause types…"
- • "When debugging, reproduce first, isolate second…"
- • "For keyword research, structure output like this…"
Side-by-side comparison
| Agent Skills | Custom Instructions | |
|---|---|---|
| Scope | Task-specific — invoked for a particular workflow | Global — applies to every conversation |
| Purpose | Defines a repeatable process | Sets standing preferences |
| When active | When the relevant task is happening | Always, in every session |
| Complexity | Can be detailed — step-by-step workflows | Better kept short and focused |
| Multiple at once | Yes — install many, use relevant ones | One set applies everywhere |
| Shareable | Yes — installable from GitHub | Personal only |
| Portable across agents | Yes — Claude Code, Codex CLI, OpenClaw, Gemini CLI | Platform-specific (Claude.ai only) |
| Best for | Repeated complex workflows | Persistent personal preferences |
Custom instructions: your baseline
Custom instructions load in every conversation, every time. Use them for things that should be true regardless of the task — preferred response length and tone, coding language and style conventions, things you never want Claude to do, background context about who you are.
Think of them as standing orders. They don't change between tasks — they set the frame for everything that happens inside a session.
Skills: for specific jobs
A skill defines: what task it handles, what inputs it needs before starting, the step-by-step process to follow, and what the output should look like. That structure is why skills outperform prompts for complex repeated work. A prompt tells the model what you want. A skill tells it how to get there.
Skills are also modular. A developer can have one skill for PR review, another for debugging, another for API design — each with its own checklist and output format. None of that complexity needs to live in custom instructions.
The mistakes that cost the most time
Putting everything in custom instructions
A 2,000-word custom instruction block that tries to cover every workflow degrades response quality. The model holds all of it in context at all times — even when 90% is irrelevant to the current task. Shorter, focused custom instructions consistently outperform comprehensive ones.
Skills that are too broad
"Be a great marketer" is not a process. A skill needs to specify what inputs it requires, what steps it follows, and what the output looks like. Narrow skills that handle one workflow extremely well beat vague skills every time.
Duplicating instructions in both places
If "British English" is in custom instructions and also embedded in your writing skill, you're adding noise. Pick one home for each instruction.
Using a skill for one-off tasks
Skills are for repeated work. If you're doing something once, a well-written prompt is faster. Invest in a skill when you know you'll run the same workflow again and again.
How they work together
Custom instructions set the style. Skills add process on top of it. Each skill inherits the tone from custom instructions but adds its own structure on top. That's the right layering — style is global, process is per-task.
Role
Developers
Custom Instructions
Language preferences, code style, communication style
Skills
PR review, systematic debugging, API design critique
Role
Marketers
Custom Instructions
Brand voice, target audience, channel preferences
Skills
Keyword research, content briefs, campaign analysis
Role
Founders
Custom Instructions
Communication style, company stage, strategic context
Skills
Board prep, user feedback synthesis, competitive teardowns
Role
Lawyers
Custom Instructions
Jurisdiction, formality, preferred output format
Skills
Contract review, NDA triage, legal risk assessment
Frequently asked questions
Where do I install Agent Skills?
For Claude Code: ~/.claude/skills/{slug}/SKILL.md. For OpenAI Codex CLI: ~/.codex/skills/. Install with one command: npx skills add {owner}/{repo} --skill {path}. See the full install guide for all methods.
Can I use skills with Claude.ai (not just Claude Code)?
Yes. For Claude.ai, paste the SKILL.md content into a Project's instructions. The skill then applies to all conversations in that Project. The install guide covers this method.
How many skills can I install at once?
Claude Code supports multiple skills simultaneously. Skills are modular and designed to work together. The practical limit is context window size, but most professional setups run 5–10 skills without issue.
Are the skills on findskills.co free?
Yes. All 405 skills are open source (MIT or Apache 2.0) and free to install. The directory is free to browse forever.
Find skills for your workflow
405 source-verified skills organized by profession. Browse by your role or explore curated Skill Stacks.