# Gear
> Gear gives builders and agents a shared inventory of capabilities. Each is a
> focused tool — agent-friendly by design, with CLIs that print grep-friendly
> help, JSON output, and links back to the same documentation humans and agents
> read.
### What's in the catalog
The catalog is organized into five capability categories. Browse the full tables
in
[libraries/README.md](https://github.com/forwardimpact/monorepo/blob/main/libraries/README.md).
### Agent Capability
What the agent surface looks like — entry points, voice, skill data, and the
human-facing output agents produce. CLIs, REPLs, prompt templates, and
documentation site builders.
### Agent Retrieval
How agents fetch and shape context — from raw bytes through typed records to
schema-aware graph and vector inference.
### Agent Self-Improvement
Tooling that closes the Plan-Do-Study-Act loop: trace evaluation, synthetic data
generation, and XmR control charts that distinguish signal from noise.
### Agent Infrastructure
How agent services run — gRPC framework, structured logging, process
supervision, configuration loading, and the bridge that exposes services as MCP
tools.
### Foundations
Cross-cutting primitives — secrets, retries, hashing, project finder, macOS
bundle assembly, and shared test fixtures.
---
### Who it's for
**Builders** composing agentic products. Pull in just the libraries you need;
every CLI works standalone via `npx fit-`.
**Agents** deployed into the field. Each library ships a matching skill in the
`forwardimpact/fit-skills` pack, so agents land on the same docs as humans.
---
## Getting Started
```sh
npm install @forwardimpact/libcli @forwardimpact/libstorage # any subset
npx skills add forwardimpact/fit-skills
```