Framework Glossary

Every term you’ll meet across the seven steps, in plain language, each linked to the page that introduces it.

Acceptance criterion (AC) — A numbered, yes/no statement of what a successful run must do, derived from a real example you were happy with. Captured in Deconstruct as AC1, AC2, … and graded line by line in Test.

Agent (mechanism) — An orchestration mechanism that decides its own path at runtime, using tools on its own judgment and able to run unattended. Chosen in Design when a workflow’s steps depend on what it finds along the way.

Agent (building block) — One of the Orchestration-layer AI building blocks: a configured worker with a mission, tools, and constraints that an orchestrator or a person dispatches. See Agentic Building Blocks. Agent the mechanism means the whole workflow is driven by one agent that decides its own path; agent the building block is any configured agent, including the workers a mechanism-level agent dispatches.

Augmented / Automated (involvement) — The two values of human involvement. Augmented means a human is in the loop during the run — reviewing, steering, or approving at key points. Automated means the AI runs solo and a human only reviews the finished output. Set during Design and classified in the AI Workflow Design Matrix.

Autonomy — How much decision-making the AI has. Deterministic follows fixed rules with no judgment calls. Guided makes bounded decisions within guardrails you set. Autonomous plans, decides, and adapts independently. A fourth value, Human, exists only when classifying an individual step in Design — a step a person performs, with no AI involved — and never appears at the whole-workflow level. See the AI Workflow Design Matrix.

Backlog — The status: backlog value on a Workflow node: a candidate workflow Analyze has registered in your registry but nobody has started building yet.

Baseline — The report card from the round of Test that first reaches Ready. Improve re-runs the same inputs later and compares every line against this baseline to see what changed.

Building blocks (three layers) — The AI building blocks a workflow step can be mapped to, grouped in three layers: Intelligence (Model, Context, Memory, Project), Orchestration (Prompt, Skill, Agent), and Integration (MCP, API, SDK, CLI). Mapped during Design; see Agentic Building Blocks for full definitions.

Capability domain — In a goal-driven workflow, a durable competency the agent draws on at runtime (for example “research” or “synthesis”) — not a step or a pipeline stage. Capability domains replace step-by-step decomposition in Design for goal-driven workflows.

Connector — The mechanism a workflow uses to reach a live system such as a CRM, calendar, or drive. Design works out how the workflow will reach each system — a platform-native connector first, otherwise an MCP server, API, SDK, or CLI; Build has you authorize each one in the account that will actually run the workflow.

Context — One of the Intelligence-layer building blocks: the documents, examples, and reference material a step needs to produce good output. Resolved item by item during Build’s Prepare Context phase.

Context Inventory (C1…) — The table in the Workflow Requirements listing every artifact a workflow needs, each with a stable ID (C1, C2, …), its status, sensitivity, source, and whether the AI can currently access it. Captured in Deconstruct.

Definition type (step-driven / goal-driven) — Which of the two Deconstruct paths a workflow took. Step-driven means you can describe how the work gets done. Goal-driven means you know what “done” looks like and an agent figures out the steps at runtime. Chosen at the start of Deconstruct.

Design Spec — The architectural blueprint Design produces from your Workflow Requirements: platform, mechanism, autonomy, step classifications, skill and agent blueprints. It is a draft — its frontmatter carries approved: false until you say “approve,” and Build refuses to run on an unapproved spec.

Execution mode — A Workflow node field recording how the workflow currently runs: manual (a person still does it, AI hasn’t run it yet), augmented (AI runs it with a human in the loop), or automated (AI runs it unattended). One of the four fixed-value fields on a Workflow node — see the AI Registry Setup guide — kept current by each framework step that touches the node, and checked by Run before a workflow can be put on a schedule.

Fix mode — The path Build takes when you return from a Not Ready verdict in Test: it rebuilds only the building blocks your test results named, leaving everything else installed as-is.

Golden example — A real past output you’d consider exactly right, supplied in Deconstruct. Test compares new output against it to see what’s missing, extra, or different in substance.

Human gate (G1…) — A point in the workflow where a person must review or approve before it continues, captured with a stable ID (G1, G2, …) in Deconstruct and checked off in Test’s report card.

Lens — The scope of an Analyze audit: Individual (your own workflows) or Organizational (a value chain process spanning multiple roles). The lens changes what you’re analyzing, not the matrix you classify it on — see the AI Workflow Design Matrix.

Orchestrator skill — The top-level skill that drives a Skill-mechanism workflow: it invokes the component skills in order and pauses at human gates. Named after the workflow itself and specified in Design’s Orchestrator Prompt Outline.

Packaging — How Design decides the built artifacts will ship: as a Plugin, a Standalone Skill, a Workspace Agent, or Loose Files.

Platform mode — Whether your platform generates source files directly (code mode, e.g. Claude Code, Cursor) or step-by-step GUI instructions (guided mode, e.g. Copilot Studio). Build reads this from the platform registry — it is never something you select.

Process (registry) — A registry node representing an end-to-end business process; every Workflow node belongs to exactly one Process, listed in that Process’s # Workflows section. Set up during your AI Registry interview.

Registry — Your AI Registry: a folder of Markdown “nodes” describing your business, its processes, and its workflows, which every framework step reads from and writes to so you never re-describe your work. See the AI Registry Setup guide.

Report card — The table Test produces per scenario: every acceptance criterion, rule, and human gate listed as Expected | From | Result | Evidence, with Result always exactly Met or Not met.

Requirements (Workflow Requirements) — The Product Requirements Document for a workflow: goal, value and measurement, context inventory, acceptance criteria, example scenarios, rules, and human gates. Produced by Deconstruct and consumed by every later step.

Rule (R1…) — A must-do, must-never-do, or fallback statement about how the work should be done, captured with a stable ID (R1, R2, …) in Deconstruct and checked in Test’s report card.

Run Card — The one-page guide Run writes after your first real run, covering how to start the workflow, what to have ready, and what to check before you act on the output. Saved as run-guide.md.

Run log — The one-line-per-run record (runs.md) that Run starts: date, input, result, and edits needed. Improve reads it as evidence of drift.

Scenario (E1…) — A representative input, with what to look for in the output, captured in Deconstruct’s Example Scenarios and given a stable ID (E1, E2, …). Test runs each one and grades it against the acceptance criteria.

Skill (mechanism) — An orchestration mechanism you start by name; it follows the mapped steps every time, pausing where you said. Chosen in Design when you trigger the work yourself and the same steps repeat each run.

Skill (building block) — Any reusable instruction set the model follows, including the smaller component skills an orchestrator skill invokes. Skill the mechanism means the whole workflow runs as one skill you start by name; skill the building block is any reusable instruction set, including the component skills inside it. See Agentic Building Blocks.

Stable ID — A short label (S1, A1, C1, E1, AC1, R1, G1) that stays the same across every document referencing it, so every framework step can point at exactly the same component. Introduced in Deconstruct and Design.

stale_after — The review-date field on a Workflow node. A workflow is stale once today’s date reaches it; Run sets it on go-live, and Improve resets it at the close of each review.

Workflow node — The registry file (registry/workflows/<slug>.md) that tracks one workflow’s status, mode, and artifacts across every framework step, so any session can pick up where you left off. First written by Analyze; see the AI Registry Setup guide.