| title | Agentic Skills |
|---|---|
| emoji | 🎯 |
| colorFrom | red |
| colorTo | purple |
| sdk | static |
| pinned | false |
Plug-and-play skill definitions that turn any AI coding assistant into a domain-aware specialist. Each skill is a
SKILL.mdfile — drop it into your assistant's skills directory, invoke it with$skill-name, and your assistant adopts that skill's operating rules, patterns, and quality checklists.
▶ Live Demo | AIwithSakthivel | HuggingFace Space
| Skill | What it does | Invoke |
|---|---|---|
| frontend-skill | Frontend specialist — React, Vue, Angular, accessibility, responsive UI, performance, component testing | $frontend-skill |
| sci-experiment | Scientific experiment structure — hypothesis-first, reproducible, with configs, notebooks, src modules, baselines, and research documentation | $sci-experiment |
| concept-tutor | Research-grounded science tutor — intuition to formalism, readable formula cards, derivations, paper references, enterprise tradeoffs, and self-assessment for data scientists and ML practitioners | $concept-tutor |
| repo-concept-explainer | Repository-grounded concept explainer — traces implementation, data, tests, and architecture to explain what a concept means inside a specific codebase, with formulas, standalone examples, and design context | $repo-concept-explainer |
| python-comment-helper | Shipping-quality Python documentation pass — Google-style module, class, and function docstrings, useful inline comments, noise removal, and anti-invention discipline for code review | $python-comment-helper |
A skill is a SKILL.md file that defines how your AI assistant should approach a domain. When you prefix a prompt with $skill-name, the assistant loads the skill's context — its operating rules, patterns, and quality checklist — before responding.
$concept-tutor Teach me cross-entropy loss
The assistant responds as a focused research science tutor, applying the lesson structure, formula card pattern, research reference rules, and self-assessment gate from that skill.
Works with any AI coding assistant that supports skill or context files.
git clone https://github.com/AIwithSakthivel/agentic-skills.gitcp -r agentic-skills/frontend-skill ~/.your-assistant/skills/
cp -r agentic-skills/sci-experiment ~/.your-assistant/skills/
cp -r agentic-skills/concept-tutor ~/.your-assistant/skills/
cp -r agentic-skills/repo-concept-explainer ~/.your-assistant/skills/
cp -r agentic-skills/python-comment-helper ~/.your-assistant/skills/$frontend-skill Build a modal dialog with focus trap and keyboard dismiss
$sci-experiment Set up an experiment to test whether feature X improves AUC on this dataset
$concept-tutor Teach me attention mechanisms
$repo-concept-explainer Explain how embedding similarity is computed in this repo
$python-comment-helper Add Google-style docstrings and prepare this module for code reviewGuides modern frontend implementation with maintainable structure, predictable behaviour, accessibility, and performance.
Covers:
- Component-driven architecture (React / Vue / Angular / vanilla JS)
- Responsive layouts and CSS design tokens
- Accessibility — keyboard nav, ARIA, focus management, colour contrast
- Performance — code splitting, render optimisation, image sizing
- Testing — unit, integration, E2E, accessibility assertions
Invoke: $frontend-skill <your request>
Guides data scientists through designing and running reproducible scientific experiments — from hypothesis to findings.
Covers:
- Hypothesis-first structure: question, method, baseline, success criterion, findings
- Experiment layout:
notebooks/,src/,configs/,data/raw,results/,environment.yml - Config management — all parameters (seeds, hyperparameters, paths) in
configs/, never hardcoded - Notebook / src separation — explore in notebooks, extract stable logic to
src/ - Reproducibility checklist — seed pinning, environment lock, deterministic data processing
- Research brief (
README.md) as the primary deliverable - LLM-backed experiment support via
configs/prompts/and a provider interface
Invoke: $sci-experiment <your request>
Turns your AI assistant into a rigorous, research-grounded science tutor for advanced technical learners — especially data scientists and ML practitioners targeting research and production roles.
Covers:
- Intuition-first lessons that move through formalism, derivation, and enterprise application
- Readable formula cards — purpose, plain-English meaning, symbol definitions, read-aloud translation, memory hooks, tiny numeric examples, sensitivity intuition, and common mistakes
- Research trail — seminal papers, top-conference work (NeurIPS, ICML, ICLR, KDD, ACL), surveys, canonical textbooks
- Enterprise lens — accuracy vs latency vs cost, build vs buy, offline vs online metrics, governance, monitoring, and production reliability
- Self-assessment gate — 5–10 questions per lesson covering recall, intuition, formalism, assumptions, derivation, application, comparison, failure modes, research literacy, and enterprise judgment
- Active learning — retrieval questions, derivation exercises, numerical worked examples, paper-reading prompts, implementation tasks
Use when: learning, reviewing, studying, practicing, or deeply evaluating any ML, statistics, data science, NLP, retrieval, causal inference, MLOps, or applied science concept.
Invoke: $concept-tutor <your request>
Explains technical concepts grounded in the actual repository under analysis — source files, tests, configs, datasets, schemas, sample data, prompts, metrics, and architecture.
Covers:
- Repo-first investigation: builds a quick repo map, traces definitions and callers, inspects data samples and test fixtures
- Evidence discipline — separates direct repo evidence from reasoned inference; never invents behavior or formulas the repo doesn't support
- Data-grounded examples — uses real or representative sample data to show how the concept acts in context
- Formula pattern — maps each mathematical term back to the implementation, config, or metric in the codebase
- Standalone exercise — minimal command, snippet, or unit-test shape so the learner can isolate and verify the concept independently
- Design role — explains why the concept exists, what depends on it, and what breaks if it changes
Use when: understanding what a concept means inside a specific repo, tracing an algorithm, explaining a design choice, or verifying how a feature works with real or fixture data.
Invoke: $repo-concept-explainer <your request>
Performs a shipping-quality documentation pass on Python code. Defaults to a tight one-liner docstring; adds Args, Returns, and Raises sections only when the summary line cannot carry the contract. Names the mechanism in the summary line, not just the outcome. Chooses verbs that reflect the nature of the operation. Removes noise and never invents behavior or guarantees not supported by the code.
Covers:
- Module docstrings — always multi-line: summary + elaboration body + closing
"""on its own line; proportional to file complexity - Class docstrings with
Attributesfor public and non-obvious internal classes;__init__on data-structure classes documents fields created, not dependencies - One-liner first — single summary sentence by default; sections only when the one-liner genuinely cannot carry the contract
- Mechanism naming — summary line names the algorithm or scoring method, not just the outcome ("Ranks EDUs by token overlap", not "Ranks EDUs")
- Verb precision — Infers, Rebuilds, Creates, Returns chosen to reflect the nature of the operation, not just its output
- Skip heuristic — omits docstrings for tiny private helpers when name and type hints are self-explanatory
- Inline comments explaining the why: intent, tradeoffs, workarounds, safety constraints, performance choices
- Anti-invention discipline — never documents behavior, guarantees, or business intent not supported by the code
- Documentation pass summary after revising a file
Use when: commenting Python code, preparing a module for code review, converting comments to Google style, cleaning up noisy comments, or documenting source, helper, utility, service, or library files.
Invoke: $python-comment-helper <your request>
The live demo is a static single-page app (index.html) that shows each skill handling a realistic request with an animated response. No build step — just open in a browser.
open index.htmlAlso available as a HuggingFace Space (static HTML). The SKILL.md files are context definitions, not model weights — but the interactive demo shows their effect in a familiar interface.
agentic-skills/
├── index.html ← Live demo UI (GitHub Pages)
├── README.md
├── frontend-skill/
│ ├── SKILL.md
│ └── agents/openai.yaml
├── sci-experiment/
│ ├── SKILL.md
│ ├── agents/openai.yaml
│ └── references/
│ ├── experiment-pattern.md
│ ├── research-docs.md
│ ├── validation-pattern.md
│ └── llm-experiment-pattern.md
├── concept-tutor/
│ ├── SKILL.md
│ └── agents/openai.yaml
├── repo-concept-explainer/
│ ├── SKILL.md
│ └── agents/openai.yaml
└── python-comment-helper/
├── SKILL.md
└── agents/openai.yaml
MIT — use, fork, and extend freely.
Built by AIwithSakthivel