Short definitions for terms used by the CLI, its reports, IDE integrations, CI workflows, and AI consumers.
| Term | Plain-language meaning |
|---|---|
| Workspace | A governed collection of one or more registered projects. It is not required to be a Git monorepo. |
| Project | One registered application, service, library, or infrastructure unit inside or linked to a workspace. |
| Workspace manifest | .workspai/workspace.json; workspace identity, profile, engine, and bootstrap metadata. It is not the project registry. |
| Workspace contract | .workspai/workspace.contract.json; the operational registry of projects and their declared ports, APIs, ownership, and relationships. |
| Registry summary | .workspai/workspace-registry.v1.json; the canonical lightweight project count and registry status for UI and CI. |
| Workspace Model | The canonical, deterministic structural description produced from registered projects, manifests, contracts, policies, and detected facts. |
| Knowledge Graph | A proof-backed representation derived from and structurally bound to the canonical Workspace Model. It makes entities and relationships queryable; it cannot replace or mutate the authorizing model. |
| Structural model hash | A stable SHA-256 over model structure. Volatile timestamps, run correlation, live evidence references, and freshness fields are excluded so harmless refreshes do not create false graph drift. |
| Dependency graph | The project-to-project dependency subgraph embedded in the Workspace Model and used by impact and verification. |
| Provider | A deterministic source adapter that emits facts from code, manifests, APIs, infrastructure, documentation, Git, or another supported source. |
| Fact | A normalized observation about the workspace, such as a package, endpoint, import, deployment, or owner. |
| Evidence / proof | The source location and extraction details that justify a fact or relationship. |
| Proof path | A traceable route from a query result or relationship back to its supporting evidence. |
| Artifact | A versioned file written under .workspai/, usually .workspai/reports/, for people or automation to consume. |
| Command projection | The JSON returned on stdout for one command. It can include status and output metadata around a canonical artifact payload. |
| Term | Plain-language meaning |
|---|---|
| Unified runner | workspace intelligence run; the supported way to execute the complete intelligence chain in contract order. |
| Preflight | sync and baseline resolution. These prepare the run but are not stages in the 11-stage intelligence chain. |
| Stage | One contract-defined operation in the intelligence chain. Stage order comes from workspace-intelligence-chain.v1.json. |
| Gate | A decision that can pass, need attention, or block automation according to evidence and policy. |
| Blocked | The command completed, but evidence or policy prevents the requested release/action. The unified runner uses exit code 2. |
| Failed | Execution itself failed. The unified runner uses exit code 1. |
| Needs attention | Evidence is usable but contains warnings or non-blocking gaps. Strict policy can promote it to a blocking result. |
| Fresh / stale | Whether an artifact still matches its governed inputs and dependency closure. A recent timestamp alone does not prove freshness. |
| Snapshot | A stable baseline of the model used for later comparison. |
| Diff | The structural change between the current model and a baseline model or snapshot. |
| Impact | Direct and transitive consequences of a model diff, including verification scope and risk. |
| Verify | The evidence-backed gate over the affected dependency subgraph and workspace policies. |
| Context | A bounded, agent-oriented projection of current workspace evidence. |
| Agent sync | Generation of tool-specific instructions, skills, prompts, indexes, and MCP metadata from canonical evidence. |
| Evaluation | A provenance-aware record of model calls, tool activity, cost, latency, and verified task outcome produced by workspace eval. |
| Verified outcome | A task result supported by the workspace verification path; smaller context alone is not treated as task success. |
| Term | Plain-language meaning |
|---|---|
| Agent grounding | Instructions and evidence references that keep an AI agent inside the correct workspace, contracts, and command loop. |
| Bounded retrieval | Returning only the most relevant entities and proof paths for a question instead of injecting the complete graph or model. |
| Retrieval benchmark | A deterministic comparison of readable proof-source corpus size with one bounded graph response; it is not billing evidence. |
| Token provenance | Whether a token value was provider-reported, counted by a named tokenizer, estimated, or unavailable. |
| MCP | Model Context Protocol; Workspai exposes read-oriented workspace tools through workspace mcp serve. |
| Module recommender | The optional embedding-based FastAPI/NestJS recommendation feature. It is separate from deterministic Workspace Intelligence. |
| Canonical path | The current .workspai path that new writers and consumers should prefer. |
| Legacy path | A .rapidkit compatibility path read for older workspaces; it is not the target for new integrations. |
- Run the full loop: Unified Workspace Intelligence Runner
- Query evidence: Workspace Knowledge Graph
- Find an output: Artifact Catalog
- Look up syntax: Command Reference