Build for AI Agents
Don’t build another chatbot. Teach an AI agent something useful.
AI is moving from chatbots to agents that do real work. But agents waste tokens, lose the thread in complex workflows, and need the same instructions again and again. Build a skill that fixes that, publish it for everyone, and put it in front of real users.
Certificate for every valid entry Goodies for the top 20 Showcased on LLM SkillHub- 6–26 Oct 2026
- Results on 29 Oct 2026
- College students across India
- No AI/ML expertise needed
- No app to build, no model to train
--- name: gst-invoice-check description: Check an Indian GST invoice for every field the rules require. Use when someone shares an invoice and asks if it is complete or ready to file. --- # Checking a GST invoice 1. Confirm each GSTIN is 15 characters; the first two digits are the state code. 2. Same state as the place of supply: CGST + SGST. Different state: IGST. 3. List every missing field. Never invent a value.
What you win
Certificates, goodies and a skill people use
Certificate of achievement
Build for AI Agents
LLM SkillHub Student Challenge 2026
awarded to
Your Name
for building and publishing an AI-agent skill
Goodies for the top 20
For the top skill publishers, as ranked by the judges.
- Top 10T-shirt, Sipper and Keychain
- Ranks 11–20T-shirt
A certificate for every valid submission
Submit a valid entry and you get a certificate: proof of a real AI-agent project, ready for your CV and LinkedIn.
Published and showcased
Your skill listed on LLM SkillHub, where agents and developers can find and install it.
A portfolio project
A practical AI-agent project you can link to, with real users behind it.
Mentorship from industry experts
For the best skills that target a real problem and solve it.
Preference for future internships
If internship opportunities come up, winners will be preferred. A preference, not a guaranteed selection.
Open source that gets used
A skill other developers and AI-agent users can install, learn from and build on.
The problem
Agents are capable. They are not yet practised.
Tokens, spent again
Without guidance, an agent works the same task out from scratch every time, and someone pays for every token of it.
Workflows that lose the thread
Multi-step jobs go wrong in the middle: a skipped check, a guessed value, a step done out of order.
The same instructions, again
People paste the same guidance into chat after chat. A skill is that guidance, written once and loaded when the task calls for it.
What is a skill?
A folder that teaches an agent one job. The agent reads its description to decide when to use it, then follows the instructions inside. Only SKILL.md is required.
Works with Claude CodeCursorCodexGemini CLIopencode
Read the developer guide →my-skill/ ├── SKILL.md # when to use it, and how ├── scripts/ # optional: code it can run ├── references/ # optional: docs it reads on demand └── assets/ # optional: templates, data
Your mission
Six steps from a problem to a published skill
- 1
Identify a real-world problem
Pick something people actually struggle with, ideally people you can reach. Specific beats broad: “check a GST invoice” over “help with taxes”.
- 2
Build an AI-agent skill
Write SKILL.md: when an agent should use it, the steps, and the mistakes to avoid. Add scripts or reference files only if they earn their place.
$ llmsh init - 3
Publish it publicly on LLM SkillHub
As a public skill, from your own LLM SkillHub account: its email is the one you add to your Unstop registration form. Only public skills can be judged and showcased, and every public version is read by a person before it is listed, so publish early.
$ llmsh publish ./my-skill - 4
Create evaluation casesOptional
Good to have, not required. Define what “good” looks like: realistic situations, what a good answer has to contain, and the fraction that has to pass. Publish them as an eval.
- 5
Test with real users
Put it in front of people who have the problem. Note what worked, what confused them, and what you changed because of it.
- 6
Submit for judging
Your skill, your eval and your results, submitted on Unstop before the deadline.
Format and timeline
How the challenge runs
Registration
Register on Unstop, and add your LLM SkillHub-linked email in the registration form.
6–26 Oct 2026Build and publish
Write your skill and publish it on LLM SkillHub.
Test and refine
Test with real users, and run your eval if you wrote one.
Submission
Skill, write-up and feedback, on Unstop.
deadline 26 Oct 2026, 11:59 PM ISTResults
Winners announced and skills showcased.
announced 29 Oct 2026
Judging
What the judges look for
Real-world usefulness
Does it solve a problem real people have, and would they reach for it again?
Quality and reliability
Does an agent follow it the same way every time, including on the awkward cases?
Evaluation performanceIf included
Counts when you include an eval: how the skill does on it, and how honest and demanding the cases are.
Problem-solving and originality
A fresh angle or an under-served problem, not a renamed copy of an existing skill.
Documentation and usability
A description that tells an agent exactly when to use it, and a skill anyone can install and use in minutes.
Feedback from real users
Evidence that people tried it, what they said, and what changed because of it.
Submission format
What to hand in
llmsh publish is public unless you add --private, and the Publish page asks. Drafted it in your private space? Open the skill’s page and choose Make public. Public versions are read by a person before they are listed, so leave time for that.Submit on Unstop
- Your skill: the link to its public page,
llmskillhub.com/you/your-skill - A short write-up (up to 500 words): the problem, who has it, how the skill solves it, and what you changed after testing
- User feedback: how many people tried it and what they said, as quotes or screenshots
Optional, but good to have
- An eval: the link to it on LLM SkillHub, with its cases and pass bar
- Eval results: the agent and model you ran it with, and how many cases passed against your bar
- A demo video of up to 2 minutes, showing an agent using your skill
Your skill must
- Be public on LLM SkillHub. A skill in a private space can’t be judged.
- Be published from the LLM SkillHub account whose email you gave in your Unstop registration form
- Pass
llmsh validatewith no errors - Have a description that says what it does and when to use it
- Carry a licence, so others can use it
- Contain no secrets, keys or personal data
- Be your own work, with anything borrowed credited
Do’s and don’ts
Build it the way it will be judged
Do
- Solve one problem well. A narrow skill that works beats a broad one that sort of works.
- Write the description for the agent: say exactly when the skill should be used.
- Publish your skill as public. Skills in a private space can’t be judged.
- Add an eval if you can. It is optional, but it is the clearest proof your skill works; include hard and failing cases, not only the happy path.
- Test with real people and change the skill based on what they tell you.
- Publish early, so review is finished well before the deadline.
- Credit your sources and respect the licence of anything you include.
Don’t
- Build another chatbot, or a wrapper around one.
- Paste one giant prompt with no structure and call it a skill.
- Put API keys, passwords or anyone's personal data in your skill.
- Include instructions that could harm users, bypass safeguards or leak data. These are rejected in review.
- Submit a skill that is still private. It must be public on LLM SkillHub.
- Copy someone else's skill or eval, or pick eval cases just to flatter your results.
- Submit something you have not run with a real agent.
Ideas
Your skill could help with anything
Starting points, not a list to choose from. The best entries come from a problem you have seen up close.
Review a database migration for changes that could lose data
Summarise a paper with every claim tied to a page number
Walk through an NCERT maths problem step by step without giving the answer away
Turn meeting notes into owners, deadlines and follow-ups
Clean a messy government open-data CSV and explain every change
Tailor a resume to a campus placement job description
Write a scheduled job safely, with a way to roll it back
Check a GST invoice for every field it must carry
Who can participate
College students across India
Interested in AI, software development, automation, research, productivity, or simply building useful technology? You are who this is for. You don’t have to be an AI/ML expert.
Start in five commands
$ npm install -g llmskillhub $ llmsh login $ llmsh init $ llmsh validate ./my-skill $ llmsh publish ./my-skill # public by default
Sign in with Google or GitHub and note the email on your account page: it goes in your Unstop registration form. Then create a token there for llmsh login.
FAQ
Questions
Do I need to be an AI/ML expert?
No. What matters most is finding a useful problem and building a skill that solves it well. If you can explain a task clearly, you can write a skill.
Do I need to build an app or train a model?
No. A skill is a folder with a SKILL.md file (instructions, and when to use them), plus optional scripts and reference files. You are teaching an existing agent, not building a new one.
Which agents can use my skill?
Skills use the open Agent Skills format. llmsh installs them for Claude Code, Cursor, Codex, Gemini CLI and opencode, and any agent that speaks MCP can search the catalogue.
Which email do I give in the Unstop registration form?
The email linked to your LLM SkillHub account: the Google or GitHub account you sign in with, shown on your account page. It does not have to be the email you registered on Unstop with. It is how your submission is matched to your skill, so publish your skill from that account.
Who gets a certificate, and who gets goodies?
Everyone with a valid submission gets a certificate. Goodies go to the top 20 skill publishers as ranked by the judges: the top 10 get a T-shirt, a sipper and a keychain, and ranks 11–20 get a T-shirt.
Do I have to submit an eval?
No. An eval is optional but good to have: it is the clearest evidence that your skill does what it claims, and it counts towards judging when you include one.
How do I run my eval?
If you add one: run each case through an agent with your skill installed, judge each answer against what the case expects, and report how many passed against your pass bar. LLM SkillHub stores and shares evals; it does not run them for you, so say which agent and model you used.
Can my skill be private?
No. Your submission must be a public skill: judges, the showcase and other users all need to see and install it. You can draft in your private space, then open the skill’s page and choose Make public before you submit.
How long does review take?
Every public version is read by a person before it is listed. Publish well before the deadline; while it waits, you can see it on your own account.
Does publishing cost anything?
No. Publishing and installing skills on LLM SkillHub is free.
Will this lead to an internship?
If internship opportunities come up in future, winners will be given preference. It is a preference, not a guarantee of selection.
Where do I register and submit?
On Unstop. This page is the brief; Unstop handles registration, submission and announcements.
Who do I contact with questions?
Email hello@llmskillhub.com.
Still have a question? Write to hello@llmskillhub.com.
Terms and conditions
The fine print, in plain words
Top 20 only. Top 10: T-shirt, sipper and keychain. Ranks 11–20: T-shirt. No one else receives goodies.
From industry experts, for the best skills only. Those that target a real problem and solve it, as selected by the judges; not everyone will receive it.
- Goodies go only to the top 20 skill publishers, as ranked by the judges: the top 10 each receive a T-shirt, a sipper and a keychain, and the next 10 (ranks 11–20) each receive a T-shirt. No other participant receives goodies.
- Mentorship from industry experts is offered only for the best skills that target a real problem and solve it, as selected by the judges. Not every participant will receive it.
- If internship opportunities arise in future, winners will be given preference. This is not a guarantee of selection, and no internship is promised.
- Every participant with a valid submission receives a certificate. A submission is valid when it meets the requirements on this page.
- Your LLM SkillHub linked account Email Id should be provided on the Unstop registration form.
- Submitted skills must be public on LLM SkillHub, your own work, and must pass review. Skills that are copied, plagiarised, rejected in review, or that break the LLM SkillHub Terms of Service are disqualified.
- Evals are optional. Judges consider eval results only when an eval is included.
- You keep ownership of your skill. It is published under the licence you choose, and LLM SkillHub and the organisers may feature it, with your name, when showcasing the challenge.
- Collect user feedback with the consent of the people giving it, and do not submit anyone’s personal data.
- The judges’ decisions are final. The goodies cannot be exchanged for cash, and the organisers may substitute an item of similar value if one is unavailable.
- The organisers may update the rules, the timeline or the rewards at any time. Changes are announced on Unstop.
- Questions about the challenge go to hello@llmskillhub.com.
- Participation is also subject to the LLM SkillHub Terms of Service and Privacy Policy, and to Unstop’s terms.
See also the LLM SkillHub Terms of Service and Privacy Policy.
One person saving a few tokens may seem small. When thousands of people use better-guided skills, the impact is significant.
Build it. Evaluate it. Publish it. Let the community use it.
A certificate for every valid submission Goodies for the top 20 skill publishers