Best AI Agent Skills for Prompt Engineering in 2026
A. Frans
Published July 11, 2026
Table of Contents
- 01The short list
- 02Prompt Master — rough ask in, precise prompt out
- 03LangGPT — prompts you can reuse and version
- 04Get Shit Done — meta-prompting and spec-driven work
- 05Prompt Engineering Guide — the reference you settle arguments with
- 06Skill Creator — package your prompting into something reusable
- 07Context Engineering Kit — fix the context, not just the prompt
- 08How to choose
- 09Start with structure, add technique later
- 10Trust notes
- 11FAQ
Most bad agent output isn't a model problem — it's a prompt problem. The instruction was vague, the context was dumped in raw, the format was left to chance, and the model did something reasonable that wasn't what you meant. Prompt engineering is the discipline of closing that gap, and in 2026 it's less about clever one-liners and more about structure: giving the model a clear role, the right context in the right order, and an output shape it can't misread. That's work an agent can help with, because a good prompt-engineering skill turns the model into a partner that drafts, critiques, and tightens its own instructions. The skills below do exactly that — they teach structured prompt formats, meta-prompting, and context engineering, and a couple are reference libraries deep enough to settle an argument about technique.
If you write prompts for a living — or you're building agents and tired of coaxing them — here are the skills worth installing. Everything runs in Claude Code, and I've flagged trust tier and license for each so you know what you're bringing into your workflow.
The short list
| Skill | Best for | Trust tier | License | Install target |
|---|---|---|---|---|
| Prompt Master | Turning a rough ask into a precise prompt | Verified | MIT | Claude Code |
| LangGPT | Structured, reusable prompt templates | Verified | Apache-2.0 | Claude Code |
| Get Shit Done | Meta-prompting and spec-driven context | Verified | MIT | Claude Code |
| Prompt Engineering Guide | The reference for technique and research | Verified | MIT | Claude Code |
| Skill Creator | Building and optimizing your own skills | Official | MIT | Claude Code |
| Context Engineering Kit | Improving output by fixing the context | Community | GPL-3.0 | Claude Code |
Prompt Master — rough ask in, precise prompt out
Start here if the daily pain is writing the prompt itself. Prompt Master (nidhinjs/prompt-master, verified, MIT, ~10k stars) is a Claude skill that takes a rough description of what you want and writes an accurate, well-structured prompt for whatever tool you're aiming at.
claude skill add nidhinjs/prompt-master
It's the first install because it attacks the most common failure directly: you know what you want but phrase it loosely, and the model fills the gaps its own way. Prompt Master makes the instruction-writing itself the job — you describe the goal, it produces a prompt with the role, constraints, and output format spelled out. That's especially useful when you're prompting a different tool than the one you're in, like drafting an image or video prompt from inside Claude Code. MIT-licensed and widely starred, it's a low-friction way to stop hand-tuning wording and let the model do the part it's good at.
LangGPT — prompts you can reuse and version
When you want prompts that behave like components instead of one-off text, LangGPT (langgptai/LangGPT, verified, Apache-2.0, ~12k stars) is the pick. It's a structured prompt framework — think of it as a schema for prompts, with roles, rules, and sections you fill in the same way every time.
claude skill add langgptai/LangGPT
The value shows up the second time you need a similar prompt. Ad-hoc prompting means rewriting from scratch and getting inconsistent results; LangGPT gives you a repeatable structure, so a prompt you tuned once becomes a template you reuse across tasks and teammates. That structure also makes prompts easier to review — when every prompt has the same shape, spotting the missing constraint or the vague instruction is faster. For anyone maintaining a library of prompts rather than firing them off individually, this is the organizing layer.
Get Shit Done — meta-prompting and spec-driven work
Some tasks are too big for one prompt, and that's where Get Shit Done (gsd-build/get-shit-done, verified, MIT, ~65k stars) comes in. It's a lightweight meta-prompting and context-engineering framework built around spec-driven work — you describe the outcome as a spec, and it structures the context the agent needs to hit it.
claude skill add gsd-build/get-shit-done
This is the pick when the problem is scope, not wording. A single clever prompt won't carry a multi-step build; what carries it is a spec the agent can decompose and the right context loaded at each step. Get Shit Done leans into that — meta-prompting to shape the plan, context engineering to feed each stage only what it needs. At ~65k stars it's one of the most adopted frameworks in this space, and it's the right mental model for anyone moving from single prompts to agent workflows that have to hold a thread across many turns.
Prompt Engineering Guide — the reference you settle arguments with
Not a tool so much as the canon. Prompt Engineering Guide (dair-ai/Prompt-Engineering-Guide, verified, MIT, ~76k stars) from DAIR.AI is a massive collection of techniques, papers, and worked examples — chain-of-thought, few-shot patterns, the research behind why they work.
claude skill add dair-ai/Prompt-Engineering-Guide
Pointing an agent at this raises the ceiling on its advice. When it's reasoning about which technique to apply — should this be few-shot, does this task need explicit reasoning steps, is there a known pattern for structured extraction — it's drawing from a documented body of work instead of improvising. At ~76k stars it's one of the most referenced prompt-engineering resources anywhere. It's the depth install: less about generating today's prompt, more about making sure the technique behind it is sound.
Skill Creator — package your prompting into something reusable
Once you've found prompt patterns that work, the next step is turning them into skills. Skill Creator (anthropics/skills, official, MIT) is Anthropic's own skill for creating, testing, and optimizing agent skills — the tool you use to formalize a good prompt into a repeatable capability.
claude skill add anthropics/skills/skill-creator
This is where prompt engineering graduates into skill engineering. A prompt you keep re-typing is a skill waiting to be written; Skill Creator gives you the official path to build, test, and refine one, so the pattern lives in your setup instead of your clipboard. Being first-party from Anthropic means it tracks the skill format as it evolves, which matters for anything you want to keep working across Claude Code updates. If you're serious enough about prompting to be repeating yourself, this is the tool that ends the repetition.
Context Engineering Kit — fix the context, not just the prompt
Often the prompt is fine and the context is the problem. Context Engineering Kit (NeoLabHQ/context-engineering-kit, community, GPL-3.0, ~1.2k stars) focuses on the other half of the equation — improving output quality by shaping what the agent has in front of it, not just how you ask.
claude skill add NeoLabHQ/context-engineering-kit
It's the pick for the failure mode where a well-written prompt still produces mediocre results because the model is working from too much noise or too little signal. Context engineering is about curation — the right files, the right prior turns, the right examples, nothing extraneous — and this kit gives the agent patterns for doing that deliberately. It's community-tier and GPL-3.0, so mind the license if you're embedding it commercially, but as a way of thinking about the input side of prompting, it fills a real gap the wording-focused tools leave open.
How to choose
Match the install to your actual bottleneck:
- Writing precise prompts faster: Prompt Master.
- Maintaining a reusable prompt library: LangGPT for structure.
- Running multi-step agent workflows: Get Shit Done for spec-driven context.
- Getting the technique right: Prompt Engineering Guide as the reference.
- Turning repeated prompts into skills: Skill Creator.
- Fixing quality problems that aren't about wording: Context Engineering Kit.
If you're building with these rather than just writing prompts, our agent skills for AI engineers roundup covers the model-and-eval side, agent skills for developers is the broader coding toolkit, and agent skills for content writers is the one to read if your prompting is aimed at writing output.
Start with structure, add technique later
There's one habit that beats collecting tools: write structured prompts before you reach for clever techniques. Most output problems dissolve once the prompt has a clear role, the context in a sensible order, and an explicit output format — the plain mechanics that LangGPT and Prompt Master enforce. Fancier methods like chain-of-thought or few-shot examples are real levers, but they're multipliers on top of a well-structured prompt, not substitutes for one. The trap is reaching for a research technique to fix a prompt that's just vague. Get the structure right first, confirm the model actually understood the ask, and only then layer on the technique the Prompt Engineering Guide would point you to. The skills here are most useful in that order — structure, then refinement, then packaging.
Trust notes
This is a lower-stakes category than most — a prompt skill shapes instructions, it doesn't touch your cloud or your files the way an infra or database tool does. Still, the lineup is chosen for signal: the anchor picks are verified-tier with real adoption (Prompt Master ~10k, LangGPT ~12k, Get Shit Done ~65k, the DAIR.AI guide ~76k stars), and Skill Creator is first-party from Anthropic. Licenses are mostly permissive MIT and Apache-2.0; the one to watch is Context Engineering Kit at GPL-3.0, which is fine to use but worth noting before you embed it in a commercial product. The one real caution: a prompt skill can inject instructions into your agent's context, so prefer verified and official sources over unknown ones, and skim what a skill actually tells the model before you trust it in a serious workflow.
FAQ
What's the difference between prompt engineering and context engineering? Prompt engineering is about how you phrase the instruction — role, constraints, output format. Context engineering is about what the model has in front of it — which files, prior turns, and examples are loaded. Prompt Master and LangGPT handle the first; Context Engineering Kit and Get Shit Done lean into the second. Good output usually needs both.
Do I still need prompt engineering with newer, smarter models? Yes — arguably more, because you're handing bigger tasks to agents that run many steps. Smarter models forgive sloppy wording on simple asks, but multi-step agent work still fails on vague specs and messy context. The discipline shifts from clever phrasing toward structure and context, which is exactly where these skills focus.
Can a skill write prompts for tools other than Claude? Yes. Prompt Master specifically writes prompts for any target tool, so you can draft an image, video, or other-model prompt from inside Claude Code. The structured formats from LangGPT are model-agnostic too — the schema transfers across tools.
When should I turn a prompt into a skill? When you catch yourself retyping the same prompt pattern. A one-off prompt is fine as text; a prompt you use weekly is a skill waiting to be written. Skill Creator gives you the official path to build, test, and optimize one so it lives in your setup instead of a notes file.
Is the Prompt Engineering Guide a tool or a reference? A reference — a large collection of techniques, papers, and examples rather than something that generates a prompt for you. Its value in an agent setup is grounding: when the agent picks a technique, it's drawing from documented research instead of guessing. Pair it with a generator like Prompt Master for the day-to-day writing.
Share this article
⚙Related Tools
📄Related Articles
Get More AI Tool Guides
New comparisons and guides every week. Join thousands of professionals staying ahead of the AI curve.