# Shipyard Documentation

Shipyard is an enterprise-grade agentic AI benchmark and evaluation engine written in Go. It orchestrates isolated sandbox environments, multi-turn tool-call rollouts, ATIF (Agent Trajectory Interchange Format) telemetry, and rubric-based scoring.

## Quick Links
- [Documentation Index](http://localhost:5173/)
- [Full Consolidated Docs for LLMs](http://localhost:5173/llms-full.txt)

## Sections
- [Overview & Architecture](http://localhost:5173/raw/README.md): Core concepts, feature set, workflow lifecycle, and quick start.
- [Quickstart Guide](http://localhost:5173/raw/getting-started.md): Installation, first run, and running an evaluation in under 3 minutes.
- [Configuration (eval.toml)](http://localhost:5173/raw/configuration.md): Evaluation manifest specification, agent overrides, and runtime options.
- [Scaffolding](http://localhost:5173/raw/scaffolding.md): Automated generation of benchmarks, test cases, and rubrics.
- [Agents & Plugins](http://localhost:5173/raw/agents-and-plugins.md): Modular agent framework, mock adapters, and subprocess drivers.
- [Environments & Sandboxing](http://localhost:5173/raw/environments.md): Docker, microVM, and host sandbox lifecycle isolation.
- [Rollouts & ATIF](http://localhost:5173/raw/rollouts-and-atif.md): Agent Trajectory Interchange Format v1.0 specifications and recording.
- [Evaluation & Rubrics](http://localhost:5173/raw/eval-and-rubrics.md): Deterministic, LLM-as-a-judge, and hybrid scoring pipelines.
- [Remote Agent Protocol](http://localhost:5173/raw/remote-agent-protocol.md): HTTP and gRPC protocols for off-cluster agent runners.
- [CLI Reference](http://localhost:5173/raw/cli-reference.md): Full command-line reference for `shipyard`.
- [Error Codes](http://localhost:5173/raw/error-codes.md): Complete list of structured error codes, diagnostic flags, and recovery steps.