Glossary

Short definitions for the acronyms and terms used across these docs. Links point to the longer explanation where one exists.

Foundry concepts

Baseline — A reusable building block published by the Foundry team. See the Baselines Overview for the current set.

Shared library — A baseline consumed as a Python package (pip install / uv add). Versioned, back-compat matters. Best for code shared across many agents.

Starter template — A baseline you fork and own. Best for the full app or service shell — agent backends, UIs, workflow workers.

Configuration-first — Agent behavior defined in declarative YAML, overridable per-environment via env vars. One config file, many environments. See Core Concepts.

Protocols

A2A — Agent-to-Agent. A standardized HTTP interface that lets agents discover, call, and stream from one another regardless of who built them. Foundry implements A2A as JSON-RPC 2.0 over HTTP plus a discovery document at /.well-known/agent-card.json. See Core Concepts.

MCP — Model Context Protocol. An industry-standard protocol for connecting agents to tools — file systems, databases, APIs, anything outside the agent's process. See Core Concepts.

JSON-RPC — Lightweight remote-procedure-call protocol over HTTP using JSON envelopes. Foundry agents expose A2A methods (message/send, message/stream, tasks/get, tasks/cancel) over JSON-RPC.

OIDC — OpenID Connect. Identity layer on top of OAuth 2.0. Used by Foundry's React UI starter to authenticate end users against an external identity provider.

Workflows

HIL — Human-in-the-Loop. A workflow step that pauses for a human decision (approval, feedback, edit) before continuing. Foundry pairs this with Temporal's durable execution so the workflow can wait minutes, hours, or days without losing state.

Workflow — In Temporal, the orchestration logic for a long-running process. Must be deterministic. Coordinates Activities.

Activity — In Temporal, a single unit of work that may have side effects (API calls, agent invocations). All A2A agent calls happen inside Activities, never directly in Workflows.

Signal / Query — Temporal primitives. Signals deliver commands to a running workflow (e.g., "approve this plan"). Queries read workflow state without modifying it.

Sessions

Session — Persisted conversation state, addressed by ID. Without sessions, every request to an agent is a cold start. See Core Concepts.

Session backend — Where session state lives. The Strands Base Agent ships file, s3, and dynamodb backends out of the box.

Security & ATO

ATO — Authority to Operate. Formal authorization to deploy a system into a regulated environment (typically U.S. federal). Foundry baselines aim to give teams a head start on the application-layer controls ATO requires.

STIG — Security Technical Implementation Guide. A DISA-published list of controls a system must satisfy. Different STIGs cover different layers (OS, container runtime, application).

ASD STIG — Application Security and Development STIG. The application-layer STIG. Covers session handling, input validation, authentication, cryptography, audit logging, and data handling. Foundry baselines are assessed against this one. See Security & STIG Posture.

DISA — Defense Information Systems Agency. The U.S. DoD agency that publishes STIGs.

Inheritance model — Each STIG finding in a baseline carries a responsibility tag: baseline (work done in the baseline code), delivery (work the adopter does at deploy time), or shared (both sides have work). Adopters only need to address delivery and shared items in their overlay.

Status terms

Used on baseline pages, in the sidebar, and in guides:

Term	Meaning
Available	Baseline is released as OSS, repo public, docs published.
Preview	This content describes a validated internal workflow documented here. The underlying baseline is undergoing final hardening prior to open-source release. Repository links may point to internal sources until publicly available.
Coming soon	Public documentation for this baseline is on the way. It will be published alongside the open-source release.