Key Concepts

This page defines the core concepts used throughout the CoffeeBreakAI documentation.

Task

A discrete unit of work for the system to perform, such as:

• Implementing or refactoring code
• Generating documentation or tests
• Decomposing an epic into child work items

Tasks are represented internally by a TaskExecution record and move through a well-defined lifecycle.

Agent

An AI-powered worker specialized for a type of task (planning, coding, refactoring, documentation, review simulation, etc.). Agents are registered in an Agent Registry and selected by the Task Router.

Workspace

A logical container for tasks, configurations, integrations, and reviewers. Workspaces let you:

• Separate projects or clients
• Configure thresholds and SLAs
• Control who can review and approve work

Confidence Score

A numeric value (0–1) that expresses how confident the system is in an agent’s output. Workspaces can define minimum confidence thresholds that determine when a human review is required.

Retry

A re-execution attempt when a task fails due to a recoverable error (like a transient network failure). Retries follow a configured Retry Policy.

Trigger Source

Describes why a task is being executed or re-executed (initial creation, retry, reviewer feedback, etc.). Trigger sources are used to drive refinement flows.

Capability Hint

Metadata that describes the skills or domains required for a task. Hints help the router pick the right agent and help the system learn from reviewer feedback over time.

Human-in-the-Loop (HITL) Review

A structured review process where human reviewers approve, reject, or refine agent output before it’s considered complete.

Use these terms as reference while exploring the rest of the documentation.