Arm Bandits Expert

Community

damusix

arm-bandits-expert

Implements, evaluates, and deploys multi-armed bandit algorithms — including Thompson Sampling, UCB, epsilon-greedy, LinUCB, EXP3, and contextual bandits. Covers algorithm selection, experiment harnesses, offline evaluation (IPS, Doubly Robust), infrastructure patterns, and correctness verification. Use when the user asks about multi-armed bandits, exploration-exploitation tradeoffs, adaptive experiments, A/B testing alternatives, online optimization, bandit-based recommendation or personalization systems, or contextual bandits.

Overview

Publisher	damusix
Repository	`skills`
Skill name	`arm-bandits-expert`
Stars	63
Forks	3
Bundled files	6
Links	Skill source Repository

Markdown instructions
A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.
Works with any LLM
AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.
6 bundled files
Scripts, templates, and references the model can read while it works. Files are read-only and never executed.
Open source
Published by damusix on GitHub. Read the source before you install it.

Installation

Install the Arm Bandits Expert AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

Open the app and go to Plugins → Skills.
Choose "Install from GitHub".
Paste the skill folder URL below and confirm.
Enable the skill in any chat where you want it available.

Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills

git clone --depth 1 https://github.com/damusix/skills.git /tmp/skills
mkdir -p .claude/skills
cp -r /tmp/skills/arm-bandits-expert .claude/skills/arm-bandits-expert

Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Arm Bandits Expert in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Arm Bandits Expert on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Arm Bandits Expert is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

Loaded only when it is needed
The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.
Switch models mid-chat
Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Multi-Armed Bandits Expert

Guide the user based on their actual need. Do not lecture; respond to what they're doing.

Routing

Assess the user's situation and route to the appropriate reference material. Read the relevant file(s) from skill/references/ before responding.

Entry Paths

"I need to learn about bandits" → Start with tier-1-core-algorithms.md. Progress to tier 2/3 only when the user is ready or asks.

"I need to pick an algorithm" → Use the decision framework below, then read the relevant tier reference for details.

"I need to build/evaluate an experiment" → Read experiment-harness-patterns.md for environment, policy, runner, and offline evaluation abstractions.

"I need to review/debug an implementation" → Read infrastructure-patterns.md for testing patterns and bug checklists. Cross-reference the relevant algorithm tier for formula verification.

"I need to deploy to production" → Read infrastructure-patterns.md for serving, reward pipelines, monitoring, and safety guardrails.

"I need to understand the business case" → Read business-applications.md for domain-specific guidance, real-world examples, and ROI evidence from 60+ named company deployments.

Decision Framework — Picking an Algorithm

Present trade-offs. Never prescribe a single "best" algorithm without context.

Step 0: Does the user specify an algorithm?

If the task names a specific algorithm (e.g. "implement UCB1", "use epsilon-greedy"), implement that algorithm — do not substitute a different one. Only use this decision framework when the user asks for help choosing an algorithm or says something generic like "implement a bandit."

Step 1: What kind of rewards?

Reward type	Candidates
Binary (click/no-click)	UCB1 is the classical default; Thompson Sampling (Beta-Bernoulli) for best empirical performance; epsilon-greedy for simplicity
Continuous (revenue, time)	UCB1, Thompson Sampling (Gaussian/NIG), LinUCB
Adversarial / non-stationary	EXP3, SW-UCB, D-UCB, change-point detectors

Step 2: Do you have context features?

Context	Candidates
No context	Epsilon-greedy, UCB1, Thompson Sampling, Softmax
User/item features available	LinUCB, contextual Thompson Sampling
High-dimensional features	Neural bandits (NeuralUCB/NeuralTS, last-layer Bayesian)

Step 3: What's your constraint?

Constraint	Recommendation
Simplest possible baseline	Epsilon-greedy with decay
Strongest theoretical guarantees	UCB1 (stochastic), EXP3 (adversarial)
Best empirical performance	Thompson Sampling
Delayed feedback	Thompson Sampling (robust to stale posteriors)
Multiple items per round	Combinatorial bandits (CUCB + oracle)
Ranked lists with position bias	Cascading bandits (CascadeUCB1, CascadeLinTS)
Arms change state over time	Restless bandits (Whittle index)
Reward distributions shift	Non-stationary bandits (SW-UCB, GLR-UCB)

Step 4: Maturity reality check

Algorithm	Maturity	Production examples
Epsilon-greedy	Battle-tested	Optimizely, Kameleoon
UCB1	Battle-tested	Widespread
Thompson Sampling	Battle-tested	Yahoo, Stitch Fix, Doordash
LinUCB	Production-proven	Yahoo News, Netflix, Spotify
EXP3	Well-established	Adversarial settings
Bayesian UCB	Production-proven	River, MABWiser
Softmax	Battle-tested	Deep RL action selection
Neural bandits	Early production	Meta ENR (9%+ CTR lift)
Non-stationary (SW/D/GLR-UCB)	Well-established	SMPyBandits, monitoring
Combinatorial bandits	Research	Ad placement (Chen et al.)
Restless bandits	Research → applied	Health interventions (Armman)
Cascading bandits	Early production	Expedia homepage ranking

Build Phases

Core Library — Implement algorithms starting with tier 1 (epsilon-greedy or Thompson Sampling). See tier-1-core-algorithms.md through tier-3-production-algorithms.md.
Experiment Harness — Build environment, runner, and metrics to compare algorithms offline. See experiment-harness-patterns.md.
Production Infrastructure — Add reward pipelines, serving, monitoring, safety guardrails. See infrastructure-patterns.md.

Reference Files

File	Contents
`references/tier-1-core-algorithms.md`	Epsilon-greedy, UCB1, Thompson Sampling — pseudocode, properties, pitfalls
`references/tier-2-practical-algorithms.md`	LinUCB, EXP3, Bayesian UCB, Softmax — context handling, adversarial robustness
`references/tier-3-production-algorithms.md`	Neural, combinatorial, non-stationary, restless, cascading bandits
`references/experiment-harness-patterns.md`	Environment, policy, runner abstractions, metrics, offline evaluation
`references/infrastructure-patterns.md`	Project structure, testing, reward pipelines, serving, monitoring, safety
`references/business-applications.md`	Business decision framework, domain guides, algorithm-to-problem mapping, ROI evidence, failure modes

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Arm Bandits Expert AI skill do?

Why use Arm Bandits Expert on TypingMind?

Because you install it once and use it with any model. Arm Bandits Expert is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Arm Bandits Expert in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/damusix/skills/tree/main/arm-bandits-expert. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Arm Bandits Expert?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Arm Bandits Expert?

As many as you like. As long as a model supports skills, you can use Arm Bandits Expert with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Arm Bandits Expert AI skill free?

It is published on GitHub by damusix. Check the repository for licensing terms. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Caveman Manage

JuliusBrussee

CommunityPopular

Inspect Caveman Cloud's experiment lifecycle and block unsafe execution. Use when asked to start, approve, cancel, promote or roll back a Caveman experiment.

Caveman Setup

Wire a repository through the Caveman Cloud gateway so every LLM request is measured, with no behavior change. Use for "set up caveman" or adding LLM spend observability.

Eng Runbook

An engineering runbook — service overview, alerts table, dashboards links, common procedures with copy-pasteable commands, on-call rotation, and an incident-response checklist. Use when the brief…

Deprecation And Migration

addyosmani

CommunityPopular

Manages deprecation and migration. Use when removing old systems, APIs, or features. Use when migrating users from one implementation to another. Use when migrating a database schema in production,…

Overview

Installation

Install in TypingMind

Install in another agent

Use it in TypingMind

Works with any AI model

Skill instructions

Multi-Armed Bandits Expert

Routing

Entry Paths

Decision Framework — Picking an Algorithm

Step 0: Does the user specify an algorithm?

Step 1: What kind of rewards?

Step 2: Do you have context features?

Step 3: What's your constraint?

Step 4: Maturity reality check

Build Phases

Reference Files

Bundled files

Frequently asked questions

Related AI Skills

Set up your own AI workspace now