About fidget
Context is the bottleneck.
An agent is only as good as what it knows when it sits down to work. fidget is a curated marketplace of contexts — packaged expertise you install in one line, so your agent shows up already fluent in the task at hand.
Why fidget exists
Every capable agent runs into the same wall: it doesn't know what you know. The reference material, the conventions, the hard-won right-way-to-do-it — that lives in someone's head, or scattered across a dozen tabs, and you rebuild it from scratch every time.
fidget exists to end that. We take expertise that's painful to assemble and turn it into context you can install — once, cleanly, in a single command.
What a “context” is
Today, most fidget contexts are distilled reference knowledge — cited, vetted, and shaped so an agent can actually use it. Like agent-harness, our reference for building agentic systems.
But “context” is the unit, not the limit. Anything that makes an agent better at a task — knowledge, conventions, the right way to behave — can become a fidget context.
What makes a fidget context good
Anyone can paste documentation into a prompt. fidget contexts are engineered — built like software, with the same discipline: sourced, verified, and structured to be cheap for an agent to read.
Built from primary sources, then stress-tested
Every claim traces back to a primary source — and has to earn its place. agent-harness distills 25 primary sources into 154 falsifiable claims, each run through a 3-vote adversarial check; only the 150 that survived made the guide, with the votes kept on record. Every claim is cited inline and linked to runnable, type-checked code.
Engineered to spend fewer tokens
A context your agent can't navigate cheaply taxes every turn. fidget contexts route the agent to the one relevant section and read only that — a single manifest lookup plus a ~30-line read, citation already inline — instead of loading a whole document to use a corner of it. Progressive disclosure means it pulls only as deep as the question demands.
Built to stay correct
The navigation an agent relies on can't be allowed to silently rot. Stable anchors, a single source of truth, and CI that fails the build when a link breaks or an index drifts keep every context honest as it changes.
A method, not a one-off
This discipline is codified into a repeatable playbook we apply to every context, whatever the domain. The standard travels; only the subject changes — which is how a catalog stays trustworthy as it grows.
Curated, not crowdsourced
fidget is an index of quality, not a dump. Every context is chosen and held to that bar before it earns a place in the catalog. It's small on purpose — we'd rather ship a handful you can trust than a thousand you have to wade through.
Who it's for
fidget is for builders — whatever you build. A team standardizing how their agents work. A developer grounding an agentic system in real references. An SEO agency teaching an agent to write title tags that rank. A game studio encoding the logic of a skill tree.
The task is yours; the context makes your agent fluent in it.
A refined tinkerer
fidget is named for the good kind of restlessness — the curious, playful urge to pick something up and keep adjusting it until it feels right. That instinct is where good context comes from.
We just refine it: take the tinkering, the trial and error, the “actually — do it this way,” and polish it into something you can depend on. Playful by nature, refined by discipline.
I built fidget because I couldn't stop fidgeting with my own setup — and the best of that work shouldn't have to be rebuilt by everyone, every time. — Aaron, Fidget Softworks
Where it's going
For now, fidget is small and sharp: a few excellent contexts, proving a simple idea — that curated context is worth installing. Where it grows from here depends on what proves genuinely good. The horizon is a trusted home for context across every domain. We'll earn it one context at a time.
Start fidgeting
/plugin marketplace add https://fidget.io/marketplace.json