# Zopu Work OS — Canonical Glossary

> **Purpose:** prevent agents and humans from using overlapping terms inconsistently.

| Term | Definition | Not the same as |
|---|---|---|
| Project | Durable product/repository/context boundary. | Work or chat thread |
| Message | Exact conversational input/output. | Signal |
| Signal | Provenanced evidence that may require attention. | Work |
| Work | Durable desired outcome linking all evidence and lifecycle. | Issue, task, session, PR |
| Work Definition | Versioned product-level contract: problem, outcome, criteria, scope, risk. | Implementation plan |
| Impact Map | Systems/contracts/data/operations potentially affected. | File list only |
| Architecture | Component interaction, ownership, contracts, trust/data flow. | Program design |
| Design Packet | Versioned implementation intent: file/call-flow deltas, types, invariants, slices. | Generated code |
| Vertical Slice | Bounded increment exposing observable behavior across necessary layers. | Horizontal backend/frontend phase |
| Step | Executable graph node; a slice may contain implementation/verify/publish steps. | Work |
| Kit | Versioned executable policy/configuration for resolving Work. | Agent or runtime |
| Kit Builder | Compiler selecting/composing a Kit from approved intent and policy. | Unbounded prompt generator |
| Resolver | Durable engine deciding what must happen next until terminal outcome. | Coding harness |
| Run | Coordination record for resolving a Work/step using one Kit version. | Attempt |
| Attempt | One bounded execution in one runtime/harness context. | Whole Work |
| Actor | Rivet durable identity owning serialized state/lifecycle. | Agent/model |
| Agent | Program/model-directed reasoning role producing proposals or actions. | Actor |
| Harness | Coding/tool-loop implementation such as OMP/OpenCode/Codex/Pi. | Sandbox |
| Runtime | Capability provider executing processes/files/network. | Harness |
| Sandbox | Isolated runtime instance for untrusted/project execution. | Worktree |
| Orb | Product-level execution envelope: runtime, workspace, harness, context, credentials, artifacts. | Agent |
| Worktree | Git checkout isolated for a candidate branch/Work. | Sandbox |
| Candidate | Unverified implementation revision/output. | Verified result |
| Verification Plan | Predeclared checks and evidence required for completion. | Test command only |
| Verification Result | Evidence-bound verdict for an exact revision/environment. | Agent statement |
| Artifact | Durable output/reference: diff, report, screenshot, commit, PR, preview. | Evidence automatically |
| Evidence | Artifact/data supporting a specific claim/check. | Summary prose |
| Question | Structured human input request blocking or affecting Work. | Chat thread |
| Decision | Durable answer/approval bound to exact Work/design/version. | Informal message |
| Review Package | Human-oriented narrative of intent, design, changes, evidence, risks. | Raw Git diff |
| Integration | Composition and verification of multiple slice/candidate changes. | Merge without checks |
| Delivery | Publication/release artifacts and transitions. | Outcome |
| Result | Actual effect compared with desired outcome. | PR merged |
| Learning | Post-work observation proposing future knowledge/process improvement. | Automatic memory mutation |
| Knowledge | Reviewed durable project/product facts, decisions, and conventions. | Harness session memory |
| Policy | Deterministic constraints for risk, tools, permissions, gates, budgets. | Model preference |
| Provenance | Exact origin and chain linking Signal, decision, code, evidence, result. | Generated summary |
| Terminal state | Explicit successful/failed/blocked/cancelled outcome requiring no hidden work. | “Agent stopped” |