workspace-lifecycle skill. It inspects the repository and CI, proposes the lifecycle split, and asks permission before editing. Execution needs separate human approval of the final content. Agents use space.environment for authorized runs and logs; they cannot grant content approval, recover uncertain runs, or destroy cloud resources. See Workspace lifecycle.space namespace in JavaScript code-mode eval, not a CLI. space.list() discovers workspaces in the current project, including closed workspaces and their goals. space.get({ workspaceId }) reads a specific workspace without opening it.space.describe({ method: 'create' }) to read the input schema, then space.create(...) to create a workspace with optional initial Goal, Workflow, and Rubric records. Check the returned ready field. If an instruction write fails after creation, the result includes the created workspace identity and completed writes so the agent can repair that workspace instead of creating a duplicate.setPhase, setRelations, open, close, archive, and restore. These use the workspace lifecycle and checkpoint rules. Keep the revision and placement generation from the latest read; stale changes reject. An agent cannot close or archive its own workspace from a running tool. Use another workspace or the browser for that action. This API does not move or delegate agents.workspaceId to space.goal.get/put, space.workflow.get/put, or space.rubric.get/put to edit another workspace in the same project. Closed workspaces stay closed. Writes require the record's expectedRevision; use zero only when creating an absent record.local://workspace/<path> and local://base/<path>. Workspace agents can read project artifacts but write only their workspace artifacts. Project agents can write project artifacts. Published edits refresh the browser catalog and open artifact views; evidence references still point to their recorded version.plan, code, review, and ship. Archive or release it only after the operational state agrees with the completion claim.