Glossary

Ultimate Harness

A runtime-agnostic orchestration layer that standardizes how agentic software work is specified, delegated, sandboxed, verified, reviewed, and promoted.

Runtime

A coding-agent environment capable of performing work: Codex, Claude Code, Pi, oh-my-pi, Hermes, or a future agent runtime.

Runtime adapter

A thin integration layer that translates Ultimate Harness mission packets into runtime-specific prompts, commands, tool invocations, and state collection.

Mission

A bounded unit of agentic work with a goal, inputs, constraints, expected artifacts, verification requirements, and promotion policy.

Mission packet

The portable data structure passed from Ultimate Harness to a runtime adapter. It describes the work without assuming a specific agent implementation.

Workflow profile

A named end-to-end procedure for a class of work, such as research-docs, spec-first-feature, bugfix-contained, adapter-design, or skill-authoring.

Skill

A reusable procedural capability that can be selected for a mission. Skills may be human-readable Markdown first and machine-indexable second.

Sandbox

An isolated execution environment where an agent can read project context and produce changes without directly mutating canonical project state.

Promotion

The controlled process of moving approved sandbox outputs into the canonical working tree, docs, issue tracker, or release state.

Verification result

Structured evidence that a mission satisfied required checks: static analysis, tests, diff review, spec compliance, security review, and human approval.

Audit trail

A durable record linking issue → spec → plan → mission → runtime session → sandbox changes → verification → promotion.

Canonical artifact

A project artifact that has been reviewed/promoted and is accepted as source of truth.

Generated artifact

An artifact produced by an agent or runtime. It is not canonical until verified and promoted.