Appendix: Glossary

The operator-edition vocabulary — what every term in this handbook means, in the order a business leader will encounter it.

The builder edition carries a wider technical vocabulary. This glossary covers only the terms a business leader needs to read, direct, and govern an external operator.

Agents and Roles

Term	Definition
Agent	Software that reads its context, decides what to do, and acts — without being triggered each time by a human. The unit the handbook refers to throughout.
Agent Manager	The human role responsible for setting objectives, monitoring the operator’s behavior, and recalibrating when stagnation or drift appears. Term coined by Harvard Business Review, February 2026.
External Operator	An agent that sits above a SaaS platform and reads it through a standardized protocol. Clawable, in this handbook, is the reference external operator.
Native Agent	An agent built inside a SaaS platform, sharing its database and runtime. FlowPilot, inside FlowWink, is the reference native agent.
Federation	Two or more agents coordinating — an external operator delegating deep work to a native agent, or agents from different businesses transacting with each other. Enabled by the A2A protocol.

Protocols and Surfaces

Term	Definition
MCP (Model Context Protocol)	The standardized layer that makes a SaaS platform legible to an agent. Exposes tools and schemas the agent can call. If a capability is not in MCP, the agent cannot reach it.
A2A (Agent-to-Agent)	The protocol that lets one agent call another and receive a structured response. How an external operator delegates to a native agent.
Skill	A single callable capability the operator can invoke — qualify a lead, create an invoice, publish a post. Each skill has a description, a schema, and a trust level.
MCP Surface	The full set of tools a platform exposes to agents. FlowWink’s MCP surface is 500+ skills at the time of writing.
Harness	The scaffolding around an AI model that turns it into a working agent — the loop, the tools, the permissions, the memory. When the press says “the harness matters more than the model,” it is making this book’s core argument: the value lives in the system around the intelligence, not the intelligence alone.

Identity and Memory

Term	Definition
SOUL.md	The file that defines the operator’s personality — voice, tone, style. Loaded every session.
AGENTS.md	The file that defines how the operator works — operating rules, mandate boundaries, priorities, what to escalate. The governance document.
HEARTBEAT.md	The file that defines what the operator does on its autonomous cycle, when nobody is asking.
Heartbeat	The scheduled cycle the operator runs on its own — reads objectives, checks system state, acts on what it finds, goes back to sleep. Typical cadence: 30 minutes to a few hours.
Daily Memory	The operator’s working journal. One file per day, dated. Raw log of what happened.
Long-term Memory	A curated file that records what mattered across sessions. The operator distills daily memory into it during its own cycles.

Governance

Term	Definition
Approval Gate	A configuration that pauses the operator before a high-risk action and requires human approval. Every skill carries a trust level: `auto`, `notify`, `approve`, or `blocked` (taken out of service entirely).
McKinsey Four-Layer Model	Accountability framework: Design (who built the skill), Deploy (who authorized it), Operate (who monitors it), Review (who audits it). From Trust in the Age of Agents (March 2026).
Stagnation	A long-term failure mode where the operator stops learning and settles into repetitive output. Caused by memory saturation, checklist ossification, no external stimulation.
Drift	A long-term failure mode where the operator’s behavior shifts away from its original design over time. Caused by reflection-loop bias, content exposure, soul mutation.

Maturity Levels

Term	Definition
L1 Reachable	An agent can connect and do the basics — answer questions, execute explicit commands.
L2 Operable	An agent can run real business processes end to end. The threshold for a real deployment.
L3 Resilient	The connection stays reliable, observable, and honest under load. An agent can run autonomously on a schedule.
L4 Federation-Ready	Agents can coordinate with other agents and react to events. The destination.

Platforms Referenced

Term	Definition
OpenClaw	The open-source agent runtime Peter Steinberger released in November 2025; by early 2026 one of the fastest-growing open-source projects ever, with roughly half a million systems running it worldwide. The reference architecture where the agent’s identity, schedule, and memory live in plain text files a human can read and correct.
ClawClass	The multi-tenant deployment layer that runs multiple OpenClaw operators side by side on shared infrastructure. Where Clawable lives in production.
Clawable	The reference external operator in this handbook. An OpenClaw instance deployed on ClawClass that reads FlowWink via MCP.
FlowWink	The self-hosted business operating system used as the reference platform throughout this handbook. Combines CRM, ERP, and CMS in one data model — deal pipeline, quote-to-cash, accounting, contracts, content publishing, HR. Comparable to Odoo but built natively for the agent era.
FlowPilot	FlowWink’s native agent — runs embedded inside the platform, on the same database and runtime. Ships as an opt-in module: switch it off and the platform still works, operated by humans or by an external agent.

If a term appears in the handbook and is not defined here, it is either too specific to the builder edition to carry operator-level meaning, or it is an ecosystem reference with its own documentation. The goal of this glossary is to make the operator-edition text self-contained for a business reader.