# Am I the only one deliberately avoiding Skills and preferring saved prompts instead? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/a043b53b416a3c6a

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-10-30 23:12 ingested 2026-10-31 02:35

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1xjuepl/am_i_the_only_one_deliberately_avoiding_skills/)

## retained text



```
I'm curious if anyone else is taking this approach, because I seem to be going in the opposite direction of the whole Skills trend.

Don't get me wrong, I do use Skills. But I only install the ones I genuinely need as a skill. I think I have about 5 or 6 installed across my AI harnesses.

For everything else, I prefer **saved prompts, custom instructions, and reusable templates**.

Here's my reasoning.

**1. I prefer explicitly telling the LLM what to do rather than hoping it discovers the right Skill.**

I have a collection of saved prompts for specific tasks, workflows, and missions. Whenever I need something done, I just hit my shortcut key, open my prompt menu, select the exact prompt I want, and send it.

That's it. No Skill discovery, no ambiguity. I know exactly what instructions I'm giving the model.

**2. Skills introduce another decision point for the LLM.**

Even with progressive disclosure and indexed Skills, the model still needs to evaluate what's available and decide whether a particular Skill is relevant.

Will it discover the right one? Will it ignore one that would have been useful? Will it invoke a Skill when I didn't actually intend it to?

Sure, modern models are pretty good at this, but why introduce another layer of decision-making when I can just specify exactly what I want?

**3. More Skills = more clutter and potential noise.**

I know Skills aren't necessarily loaded entirely into context, but their descriptions and discovery mechanisms still add another layer of information for the model to deal with.

And as your collection grows, so does the potential for overlapping instructions, ambiguous triggers, and unintended behavior.

I prefer keeping my harness as lean as possible.

**4. Saved prompts are portable and incredibly easy to maintain.**

This is probably my favorite part.

My prompts aren't tied to a particular harness or agent framework. I can use the exact same prompt in OpenCode, Codex, an IDE, a browser-based AI chat, or practically any other LLM interface.

If I want to improve a workflow, I simply edit the saved prompt. No installation, no worrying about Skill compatibility, no managing multiple copies across different environments.

**5. I like being in control of when a workflow gets triggered.**

Sometimes I want the model to approach a task naturally. Other times, I want it to follow a very specific set of instructions.

With saved prompts, that decision is entirely mine.

I don't have to wonder whether the model might decide to use some installed Skill that changes how it approaches the task.

I'm not saying Skills are useless. They're genuinely valuable for certain tasks, especially when they package specialized knowledge, scripts, tools, or complex repeatable workflows.

I just feel like a lot of what people are turning into Skills could just as easily be saved prompts.

And personally, **I'd rather have 50 saved prompts I deliberately invoke than 50 installed Skills the model has to evaluate.**

Maybe I'm missing something. Maybe Skill discovery has become so efficient that my concerns are largely irrelevant.

But so far, this approach has been working really well for me.

**Anyone else doing something similar?**
```



document fetched 2026-10-31 00:18 · canonical https://reddit.com/r/AI_Agents/comments/1xjuepl/am_i_the_only_one_deliberately_avoiding_skills
