Key terms and concepts used throughout the ChooChoo documentation. Each term links to the relevant section for a deeper understanding.

Term	Definition
Context Engineering	The discipline of designing and delivering precise, structured context to AI agents at the right time. ChooChoo is built around this as its core value layer — see Architecture.
Agent Harness	A wrapper around an AI coding tool (Claude Code, Gemini CLI, OpenCode, Codex) that provides structured startup context, enforces behavioral boundaries, and records decisions as traces.
Context Amnesia	The loss of institutional memory that occurs when an AI agent starts every session blank — no knowledge of past decisions, conventions, or constraints. The core problem Context Compilation solves. See Why ChooChoo?.
Context Compilation	The process of scanning a codebase and generating a structured AGENTS.md context document for agents to read at startup. Powered by qmd (semantic indexing) and colgrep.
Agent Benchmark	A task suite used to evaluate agent quality over time. ChooChoo traces feed benchmarks like SWE-Bench and ITBench. See Roadmap for benchmark integration status.
GEPA	Generative Evaluation and Prompt Automation — an approach to automatically improving agent system prompts from trace feedback. Planned in ChooChoo's eval pipeline; see Roadmap.
Agent Trace	A structured record linking a file change to the AI agent that made it, the conversation that motivated it, and the model that ran it. Stored in .choochoo/traces.jsonl.
Agent Rails	Constraints and boundaries declared in AGENTS.md that define what an agent is permitted to do. Enforced by the validation engine.
The Engine	The ChooChoo CLI and Validation Layer. Runs locally and in CI/CD pipelines. See the CLI Reference.
The Map	The accumulated Agent Traces stored in the Context Graph. Powers context retrieval, quality scoring, and benchmark workflows.
The Station	The governance and evaluation web UI. Provides Context Graph visualization, quality scores, risk heatmaps, and audit search. Currently in preview.
ODPS	Open Data Product Standard — defines data product ownership and service level objectives. Official spec. One of the supported schemas ChooChoo validates.
ODCS	Open Data Contract Standard — defines schema and quality rules for data interfaces. Official spec. One of the supported schemas ChooChoo validates.
Arazzo	An OpenAPI-adjacent standard for defining deterministic API workflows. Official spec. One of the supported schemas ChooChoo validates.
OpenAPI	Industry-standard format for describing REST APIs. Official spec. One of the supported schemas ChooChoo validates.
GraphQL SDL	Schema Definition Language for GraphQL APIs. Official spec. One of the supported schemas ChooChoo validates.
AsyncAPI	Standard for describing event-driven and message-based APIs. Official spec. One of the supported schemas ChooChoo validates.
Risk Score	A value from 0.0 to 10.0 calculated for every proposed change, based on agent confidence, scope, and historical quality. Determines whether a change auto-proceeds or requires human approval.
System Card	A machine-readable identity document for an AI agent, stored in system-cards/*.yaml. Declares the agent's capabilities, limitations, and compliance alignment. Referenced during context compilation.

Architecture Layers

For a visual overview of how these concepts relate, see the Architecture page.

• Architecture — The Engine, The Map, and The Station.
• Supported Schemas — ODPS, ODCS, OpenAPI, Arazzo, GraphQL, AsyncAPI, AI System Card.
• Getting Started — Put these concepts into practice.