Skip to content

Glossary

What you submit to Scope. A request bundles a task prompt, a criteria set, and a profile (or inline runtime configuration). MS Scope creates one or more runs per request. See Submitting requests (Portal).

A single execution attempt of a request by a worker. The first attempt is created when Scope picks the request up; each retry creates a new run on the same request. Logs and reports are produced per run.

The text instructions sent to the agent. Stored in a shared catalog, de-duplicated by text: when you submit a request with task text MS Scope has seen before, the request links to the existing record rather than creating a duplicate. The entity prompt-feature detection runs against. See Managing task prompts.

A criterion is a single, observable statement the judge evaluates the run’s output against. A criteria set is a named, reusable collection of criteria, optionally arranged as a DAG. Referenced from a request by ID. See Defining evaluation criteria.

The component that evaluates a run’s output against the request’s criteria and produces the report. Powered by an LLM.

The structured pass/fail outcome of a run, with rationale per criterion. Produced by the judge, generated on demand.

A reusable, versioned description of an agent runtime configuration (worker, model, agent version, MCP servers, skills, extensions). A profile has a stable identity and a series of immutable versions. See Defining profiles.

An immutable snapshot of a profile’s runtime configuration. Re-running the same profile version always exercises the same agent code, skill commits, and extension versions.

The runtime that drives an AI coding agent during a run. Scope ships three workers — see Coding agents & capabilities.

The protocol the Copilot and Claude Code workers use to talk to their respective agents. Implementation detail for users; mentioned here because it shows up in worker IDs (coder-acp-*).

A Model Context Protocol server. Exposes tools the agent can call during a run. Referenced from a profile by slug.

A packaged Copilot agent skill, identified by a Git path. When referenced from a profile, Scope pins the skill to a specific commit hash so re-runs are reproducible.

A VS Code extension installed for the duration of a run. Only the VS VS Code Copilot coding agent supports extensions.

A boolean characteristic auto-detected on a task prompt (e.g. asks_for_api). Used to slice and group runs across heterogeneous tasks. See Working with prompt features.

The prompt the LLM uses to decide whether a given prompt feature applies to a task prompt. Stored on the feature catalog entry.